Tested configurations
What has actually been run, where, and how it went. The README’s status table sums this up by vendor; the runs below keep the evidence.
Read the combined catalogue on the website for the frozen table and new run files together. This source page keeps the existing table visible without running a command. The website assembles its view from these repository files; it does not collect results from a service.
Browse the new run files individually, or use the combined catalogue above. For each new measured run,
add a self-contained YYYY-MM-DD-topic-tester-run.md file under docs/tested/, following the
format and checking instructions. The tester’s name and a distinct run identifier are required;
repeat the context instead of saying as above. New results belong in those files, not in the table below.
The website orders new files by date, then filename, descending. Logs and measured commits still belong in the pull request’s Measurements and Testing sections. Do not commit an assembled reading view.
New runs
| Date | Build | OS / desktop | GPU / driver | CPU / RAM | Settings | Where | Result |
|---|---|---|---|---|---|---|---|
| 2026-11-13 | longshot v3.0.0-rc.2-1-g839e85ff (source ef9b723c), its renderer with #162’s commits (9ac9b714), built from source (no DLSS) | Windows 11, 1280x720 window, 2048x1152 output (16:9) | RX 9070 XT, AMD 26.10.2, Vulkan driver 0x8001a0 | Ryzen 7 5700X3D, 64 GB | RX_VK_RECORDER=thread |
the main outdoor area, the standard test: the save loaded, the camera turning at full speed, frame rate unlocked (HOS_FPS_MAX=240), 20 s window; 3 alternating pairs of four arms, iteration 1 rejected (a thread run’s clock 0.908 over 407 and 486 ms frames) |
2 runs (iterations 0 and 2 of the A/B): fps 75.54, 74.54, CPU 67.33, 69.96 ms a frame; the replay waited for the recorder 15.0 ms a second, 7.65 of them at vkResetDescriptorPool |
| 2026-11-13 | longshot v3.0.0-rc.2-1-g839e85ff (source ef9b723c), its renderer with #162’s commits (9ac9b714), built from source (no DLSS) | Windows 11, 1280x720 window, 2048x1152 output (16:9) | RX 9070 XT, AMD 26.10.2, Vulkan driver 0x8001a0 | Ryzen 7 5700X3D, 64 GB | RX_VK_RECORDER=thread, RX_VK_RECORDER_OWN=1 |
the main outdoor area, the standard test: the save loaded, the camera turning at full speed, frame rate unlocked (HOS_FPS_MAX=240), 20 s window; 3 alternating pairs of four arms, iteration 1 rejected (a thread run’s clock 0.908 over 407 and 486 ms frames) |
2 runs (iterations 0 and 2 of the A/B): fps 76.79, 76.34, CPU 64.78, 65.04 ms a frame; the replay waited for the recorder 9.3 ms a second, none at vkResetDescriptorPool |
| 2026-11-13 | longshot v3.0.0-rc.2-1-g839e85ff (source ef9b723c), its renderer with #162’s commits (9ac9b714), built from source (no DLSS) | Windows 11, 1280x720 window, 2048x1152 output (16:9) | RX 9070 XT, AMD 26.10.2, Vulkan driver 0x8001a0 | Ryzen 7 5700X3D, 64 GB | The recorder off | the main outdoor area, the standard test: the save loaded, the camera turning at full speed, frame rate unlocked (HOS_FPS_MAX=240), 20 s window; 3 alternating pairs of four arms, iteration 1 rejected (a thread run’s clock 0.908 over 407 and 486 ms frames) |
2 runs (iterations 0 and 2 of the A/B): fps 66.59, 66.44, CPU 57.95, 60.20 ms a frame |
| 2026-11-13 | longshot v3.0.0-rc.2-1-g839e85ff (source ef9b723c), its renderer with #162’s commits (9ac9b714), built from source (no DLSS) | Windows 11, 1280x720 window, 2048x1152 output (16:9) | RX 9070 XT, AMD 26.10.2, Vulkan driver 0x8001a0 | Ryzen 7 5700X3D, 64 GB | RX_VK_RECORDER=thread, RX_VK_RECORDER_IDLE=wait |
the main outdoor area, the standard test: the save loaded, the camera turning at full speed, frame rate unlocked (HOS_FPS_MAX=240), 20 s window; 3 alternating pairs of four arms, iteration 1 rejected (a thread run’s clock 0.908 over 407 and 486 ms frames) |
2 runs (iterations 0 and 2 of the A/B): fps 74.79, 73.44, CPU 66.12, 67.52 ms a frame; the replay waited for the recorder 22.1 ms a second; the recorder spun 0.32 of a core idle against 0.44 |
| 2026-11-12 | Hookshot: v2.1.0-rc.3-249-g564ea648 and later builds of this change (sparse VRAM mirrors, https://github.com/NotRexed/Hookshot/pull/2794) on Hookshot main c544828f (unmerged then); against the last v2.1.0-rc.3 on 6dcc15c7 | Windows 11, no window (unattended, scripted input) | RTX 4070 SUPER 12 GB, 617.42 | Ryzen 9 9950X3D (SMT off), 32 GB | 2560x1440, DLAA, RX_VK_COMPACT=1, RX_VK_COMPACT_PCT=85, RX_VK_ASTC_BC_ALL=1, RX_VK_IMAGE_BLOCK_MB=32; nothing else running, RX_VK_TEST_ALLOWED_MB=10200; RX_VK_MIRROR_SPARSE=1 against unset (then off) |
the starting forest, standing, the camera turning | run04: on 58.4-58.7 fps, VRAM mean 7,462-7,507 MB; off 58.8 fps, 8,215 MB; screenshots alike |
| 2026-11-12 | Hookshot: v2.1.0-rc.3-249-g564ea648 and later builds of this change (sparse VRAM mirrors, https://github.com/NotRexed/Hookshot/pull/2794) on Hookshot main c544828f (unmerged then); against the last v2.1.0-rc.3 on 6dcc15c7 | Windows 11, no window (unattended, scripted input) | RTX 4070 SUPER 12 GB, 617.42 | Ryzen 9 9950X3D (SMT off), 32 GB | 2560x1440, DLAA, RX_VK_COMPACT=1, RX_VK_COMPACT_PCT=85, RX_VK_ASTC_BC_ALL=1, RX_VK_IMAGE_BLOCK_MB=32; nothing else running, RX_VK_TEST_ALLOWED_MB=10200; RX_VK_MIRROR_SPARSE=1 against unset (then off), and with RX_VK_SPARSE_WAIT=1 |
Castle Town’s south gate to the main outdoor area (an area change) | run03: on 55.4 / 52.5 / 54.8 fps, off 54.0 / 52.7; mirrors with memory 1,267-1,311 of 2,128-2,144 MB; the load’s longest frame 866-961 ms on, 816-902 off, 1,197-1,232 with the binds waited for on the CPU |
| 2026-11-12 | Hookshot: v2.1.0-rc.3-249-g564ea648 and later builds of this change (sparse VRAM mirrors, https://github.com/NotRexed/Hookshot/pull/2794) on Hookshot main c544828f (unmerged then); against the last v2.1.0-rc.3 on 6dcc15c7 | Windows 11, no window (unattended, scripted input) | RTX 4070 SUPER 12 GB, 617.42 | Ryzen 9 9950X3D (SMT off), 32 GB | 2560x1440, DLAA, RX_VK_COMPACT=1, RX_VK_COMPACT_PCT=85, RX_VK_ASTC_BC_ALL=1, RX_VK_IMAGE_BLOCK_MB=32; nothing else running, RX_VK_TEST_ALLOWED_MB=6900; RX_VK_MIRROR_SPARSE=1 against unset (then off) |
the main outdoor area, a scripted five-minute walk | run02 (8 GB allowance): on 59.1-59.3 fps, VRAM peak 6,676-7,018 MB; off 59.0-59.1 fps, 7,028-7,138 MB; mirrors with memory 1,351-1,366 of 1,888 MB |
| 2026-11-12 | Hookshot: v2.1.0-rc.3-249-g564ea648 and later builds of this change (sparse VRAM mirrors, https://github.com/NotRexed/Hookshot/pull/2794) on Hookshot main c544828f (unmerged then); against the last v2.1.0-rc.3 on 6dcc15c7 | Windows 11, no window (unattended, scripted input) | RTX 4070 SUPER 12 GB, 617.42 | Ryzen 9 9950X3D (SMT off), 32 GB | 2560x1440, DLAA, RX_VK_COMPACT=1, RX_VK_COMPACT_PCT=85, RX_VK_ASTC_BC_ALL=1, RX_VK_IMAGE_BLOCK_MB=32; nothing else running, RX_VK_TEST_ALLOWED_MB=6144, RX_VK_MIRROR_MB=1536; RX_VK_MIRROR_SPARSE=1 against unset (then off) |
the main outdoor area, a scripted five-minute walk | run01 (6 GB allowance): three pairs: 57.3 / 55.3 / 58.2 fps against 47.7 / 51.5 / 49.9; frames over 25 ms 625 / 1,160 / 408 against 3,042 / 1,840 / 2,114; VRAM peak 6,394 / 6,200 / 6,387 MB against 7,327 / 6,254 / 6,444. No device lost |
Earlier runs (frozen)
These existing rows keep their original text and order, including their relative references and repeated entries. Do not append new runs here. Frame rates come from the named run’s log at the stated spot, unless the row says otherwise.
| Date | Build | OS / desktop | GPU / driver | CPU / RAM | Settings | Where | Result |
|---|---|---|---|---|---|---|---|
| 2026-11-13 | fix/default-output-2560x1440 b8542e6b (main fbd42024 + this change; longshot v3.0.0-rc.3-16-gb8542e6b): a copy of main’s game folder built again with make_game.py --update, alternated with main |
Windows 11 Pro 26200, a 2560x1440 window in front on a 3840x2160 screen (unattended, input over the control bridge) | RTX 5070 Ti 16 GB, 621.01 | Ryzen 7 7800X3D, 32 GB | No Output resolution chosen (no output_res= line, no HOS_OUTPUT_RES): 2560x1440. Quality preset Custom, Render resolution 75% (1920x1080), game’s own upscaler, Lanczos window filter, frame generation off, no cap (HOS_FPS_MAX=1000), sound off |
Copied saves: the starting forest (child) and the main outdoor area (child), standing, 60 s after 15 s of settling, 3 runs each | oe: output resolution 2560x1440 (none chosen: the game's), nvn: window crop 0,0 2560x1440. Medians of the 3 runs (PresentMon 2.6; GPU time from the log): the starting forest 82.7 fps, 1% low 63.3, frame time 12.04 ms, p99 14.13 ms, GPU 6.9 ms a frame; the main outdoor area 100.8 fps, 1% low 82.3, 9.85 ms, p99 11.54 ms, GPU 6.5 ms. No device lost |
| 2026-11-13 | main fbd42024 (longshot v3.0.0-rc.3-15-gfbd42024), built with make_game, alternated with the row above |
as above | as above | as above | As above; no Output resolution chosen: 2048x1152, so Render resolution 75% is 1536x864 | as above | oe: output resolution 2048x1152 (the game's), nvn: window crop 0,0 2048x1152. the starting forest 81.9 fps, 1% low 64.9, frame time 12.11 ms, p99 14.08 ms, GPU 6.6 ms a frame; the main outdoor area 102.3 fps, 1% low 84.3, 9.76 ms, p99 11.14 ms, GPU 6.1 ms. No device lost |
| 2026-11-14 | fix-split-ub-flush 1c0e220a (https://git.magicbean.dev/MagicBeanDev/Longshot/pulls/189, RX_VK_SPLIT_UBFLUSH default) against its parent on main, fbd42024: both renderer packs (scripts/dev_packs.py) over one base at f647a072, differing only in longshot_vk.dll; the base built without the hot list and profile (the v3.0.0-rc.3 Windows builder has no work folder) |
Windows 11 Pro Insider Preview (26220), 2048x1152 | Intel Arc B390 (Panther Lake iGPU; VRAM mirror of buffer pools off), 32.0.101.9150 |
Core Ultra X7 358H, 32 GB | Default options | the starting forest, the same save, about 2-4 minutes walking between the houses and trees; main, PR, main (test request T-K2T5) | White flashes and flickering textures in both main runs, none with the PR. The over-bright glow on the trees and houses stays with the PR. fps, median of the 5 s windows (first 6 left out): main 21.9 and 21.9, PR 20.5 (one run each way: no speed claim). Main’s runs ended in a GPU hang after 5742 frames (INDIRECT dispatch of 685adb41edc19de7) and a device loss after 2867 (dispatch of 499dbfd0721d2858); the PR’s closed normally after 3991 |
| 2026-11-13 | main 9a186f71 against fix-split-ub-flush 0a5d21e6 (this change on it, https://git.magicbean.dev/MagicBeanDev/Longshot/pulls/189), built from source (llvm-mingw), interleaved with two other Intel PRs’ builds (main, split, pcache, fadd, then again) | Windows 11 Pro (26300), output 1920x1080 | Intel Arc Graphics (Core Ultra 7 155H iGPU, Meteor Lake), 32.0.101.9175 | Core Ultra 7 155H, 32 GB | Render resolution 4, handheld profile, RX_VK_NO_PREDICT=1 on every side, a capture every 60 frames |
A save in the starting forest, then the control bridge’s teleport to the village and Death Mountain Trail, about 130 s each; 3 runs per side | Median over the runs of each run’s median 5 s window (the first 4 after a teleport left out) and GPU time a frame, main against this change: the starting forest 5.5 / 5.4 fps, 160.5 / 168.1 ms (3 runs each); the village 7.4 / 8.1 fps, 120.4 / 107.9 ms (2 / 3); Death Mountain Trail 8.7 / 9.3 fps, 102.0 / 93.6 ms (1 / 1): within the run-to-run spread. Device lost in the village in 2 of 3 runs on both sides. White blocks (bright flat blocks per capture): main 17 captures in 1 run of 3, up to 10.2 % of the picture; this change 0 in 3 runs |
| 2026-11-13 | host and game folder built from main 000f19fb; longshot_vk.so built alone from main fbd42024 and from RX_VK_READBACK_LATE’s change on it, 3a6fbc0c, same flags |
SteamOS 3.9.5 (kernel 7.2.7-valve2), headless | Steam Deck OLED: AMD Custom GPU 0932 (RADV VANGOGH), Mesa 26.2.0-devel; the GPU’s clock fixed at 1600 MHz (Steam’s performance menu) | AMD Custom APU 0932 (4 cores, 8 threads), 16 GB | Minimum preset (RX_VK_LOW_SPEC=1), handheld profile (HOS_HANDHELD=1), the game’s own upscaler, RX_VK_VRAM_MB=2400, RX_VK_TEX_MAX_DIM=1024, RX_VK_DESC_SLOTGEN=1, RX_VK_LOAD=create, HOS_RESERVE_CORE=0; RX_VK_READBACK_LATE as each arm names it (0 names no rule) |
a save loaded by script, standing still: a heavy view (about 2,150 draws a frame), the main outdoor area (about 1,600), the starting forest (about 2,600) | Five arms alternated in each round, each run measured over exactly 720 presents after 3 log windows: main; the change with no rule; newest; 3980; newest,3980. Medians, fps, in that order: heavy view (3 rounds) 24.28 / 24.30 / 30.13 / 24.86 / 30.33; the main outdoor area (3 rounds) 21.88 / 22.00 / 26.40 / 23.08 / 26.48; the starting forest (2 rounds, their mean) 23.44 / 23.32 / 26.30 / 23.86 / 26.78. A readback that waits takes 2.36 ms on main on the heavy view, 0.01 with newest, 1.58 with 3980, 0.03 with both; in the main outdoor area 4.47 ms with no rule, 0.50 with both. The flash detector with the camera turning (RX_VK_FLASH=1, about 1,800 presents a run, the starting forest and the main outdoor area): 0 flashes with no rule and 0 with newest,3980. No errors in 48 runs, no GPU timeout in the kernel log |
| 2026-11-13 | perf/low-vram-mirror-budget 091daa9, a full game build (scripts/dev_packs.py base, llvm-mingw, NVIDIA DLSS SDK); main’s mirrors by RX_VK_MIRROR_MB=512 in it, and 1024 MB the same way; alternated: main, PR, 1024, PR with RX_VK_MIRROR_TIGHT=1, twice |
Windows 11 Home (26200), 2048x1152 | RTX 3060 Laptop 6 GB, 621.14 | Ryzen 7 6800HS, 16 GB | Low (6GB) with Render resolution Native (DLAA) in its Handheld profile: DLSS at 1536x864; frame generation off | the main outdoor area, walking the same route, 75 s a run | fps (two runs each): main 16.9 and 17.7 (950 frames over 50 ms a run), PR 36.2 and 36.1 (about 49), 1024 MB 31.0 and 31.5, PR with the 90% rule 30.5 and 31.4. VRAM used of allowed (median, max): main 80%, 87%; PR 96%, 99%; 1024 MB 89%, 93%; the 90% rule 95%, 99%. No Vulkan errors |
| 2026-11-13 | as above, the renderer built alone from 091daa9 plus an adaptive mirror budget (on backup/adaptive-mirror-budget, not in this PR), run with RX_VK_MIRROR_ADAPT=0 (this PR’s fixed quarter), RX_VK_MIRROR_MB=512 (main’s) and adaptive; alternated, twice |
as above | as above | as above | Docked profile, Render resolution Native (DLAA): DLSS at 2048x1152; and the Low (6GB) preset (1152x648); frame generation off | the main outdoor area, walking the same route, 60 s a run | Docked native: main 12.6 and 12.5 fps (about 780 frames over 50 ms a run, GPU 61 ms), this PR 24.4 and 23.4 (135-153, GPU 31-32 ms), VRAM 97% (max 98%; 5200 of 5063 MB allowed for a moment, nothing failed); the adaptive budget 20.9 and 20.8 (353-406), dropped. Low: this PR 33.2 and 35.2, adaptive 31.9 and 30.8. No Vulkan errors |
| 2026-11-13 | v3.0.0-rc.3 game (source cb01b5a3); its renderer built alone (RX_VK_ONLY, RX_RELEASE, llvm-mingw, NVIDIA DLSS SDK) from main fbd4202 and from perf/low-vram-mirror-budget c17b339, and c17b339 with RX_VK_MIRROR_MB=1024, swapped between runs: main, PR, 1024, main, PR, 1024 |
Windows 11 Home (26200), 2048x1152 | RTX 3060 Laptop 6 GB (Optimus: drives no desktop), 621.14 | Ryzen 7 6800HS, 16 GB (15 GB usable) | Quality preset Low (6GB): its VRAM budget 3.6 GB and its DLSS render size, 1152x648, in every run; frame generation off, no cap | the main outdoor area, walking the same route, about 2 minutes a run | Median of the in-field “presents” lines over both runs: main 17.6 fps (16.7, 18.6), PR 39.2 (39.2, 38.9), 1024 MB 30.7 (22.4, 33.1). From the in-field VRAM: lines (median, max): main 4085 MB used of 4998 allowed (82%, 90%), PR 4818 of 5037 (95%, 98%), 1024 MB 4432 of 5007 (89%, 92%); textures held 1619, 1560 and 1624 MB. Render-target pool “VRAM short: given back”: main 0, PR 5, 1024 MB 0. Mirrors 512, 1498 (a quarter of the 5992 MB the driver reports) and 1024 MB; the largest pool left in system memory 224, 92 and 152 MB. No Vulkan errors |
| 2026-11-13 | fix/evict-writeback-newer-drawing c8e54dc2 (main 514591a3 + this change): the renderer built alone (RX_VK_ONLY, llvm-mingw; v3.0.0-rc.3-26-gc8e54dc2, source 02e9107d) in a game folder of longshot v3.0.0-rc.3-27-gf516654c, alternated with main’s renderer in the same folder |
Windows 11 Pro 26200, a 1707x960 window (unattended, input over the control bridge) | RTX 5070 Ti 16 GB, 621.01 | Ryzen 7 7800X3D, 32 GB | Output resolution 2560x1440 (HOS_OUTPUT_RES), Render resolution 75%, game’s own upscaler, Lanczos window filter, frame generation off, no cap (HOS_FPS_MAX=1000), sound off; RX_VK_TEST_ALLOWED_MB=6144 (a 6 GB card’s allowance) |
A copied save, the debug teleport to the main outdoor area (child), standing 25 s, then the pause menu for 5 s; 3 runs each | The HUD’s labels (d-pad icon, item count, key prompt) drawn right in 3 of 3 runs, no upload over a drawn image line for them; 3, 3 and 2 write-backs skipped, of render targets (1920x1080, 2560x1440, 960x540) unused for 600 frames or more. 104.0, 98.8 and 99.3 fps (medians of the log’s fps lines in the field). One run without the allowance: labels right, 112.1 fps. No device lost |
| 2026-11-13 | main 514591a3: the renderer built alone the same way (v3.0.0-rc.3-25-g514591a3, source 2af40fe2), alternated with the row above |
as above | as above | as above | as above | as above | The labels as noise in 3 of 3 runs, from the area’s load until the pause menu redrew them; 8 upload over a drawn image lines for them in each run, in the frame a 1920x1080 render target over their storage was written back and evicted. 105.5, 95.1 and 100.7 fps. No device lost |
| 2026-11-13 | game folder built from main e51e28e9 (longshot e51e28e9 (source 8dd02aab)); longshot_vk.dll built alone (RX_VK_ONLY, release, llvm-mingw) from fix/virtual-pool-drawn-page-map 4830ea7d (this change on e51e28e9); HUD A/B run 4 of 4 (this PR) |
Windows 11 Pro 26200, fullscreen | RTX 5080 16 GB, 610.88 | Ryzen 9 7900X3D, 64 GB | 2560x1440, DLAA, Reflex off, frame rate cap 90, frame generation off, VRAM budget 11 GB; RX_VK_PARALLEL=4, HOS_PIN=*=FFF; DualSense |
the main outdoor area next to Castle Town by day, walked into Castle Town and its market and back out | The HUD’s labels and D-pad cross drawn at the town gate. No upload over a drawn image line of the bug’s form (dirty, finalized, no “a GPU buffer wrote it”) for the HUD’s textures or render targets; 12 for a 640x360 NVN 0x14 image the texture budget remakes (main has it too). No validation, device-lost or VK_ERROR line; closed from the pause menu |
| 2026-11-13 | game folder built from main e51e28e9 (longshot e51e28e9 (source 8dd02aab)); longshot_vk.dll built alone (RX_VK_ONLY, release, llvm-mingw) from main e51e28e9 (the same file as the full build’s); HUD A/B run 3 of 4 (main) |
Windows 11 Pro 26200, fullscreen | RTX 5080 16 GB, 610.88 | Ryzen 9 7900X3D, 64 GB | 2560x1440, DLAA, Reflex off, frame rate cap 90, frame generation off, VRAM budget 11 GB; RX_VK_PARALLEL=4, HOS_PIN=*=FFF; DualSense |
the main outdoor area next to Castle Town by day, walked into Castle Town and its market and back out | At the town gate the R1 and R2 labels and the D-pad cross are not drawn. upload over a drawn image lines of the bug’s form: the six label and cross textures at one present (5926), 6 other HUD textures, 9 render targets (2560x1440 NVN 0x34 and 0xa among them). No validation, device-lost or VK_ERROR line; closed from the pause menu |
| 2026-11-13 | game folder built from main e51e28e9 (longshot e51e28e9 (source 8dd02aab)); longshot_vk.dll built alone (RX_VK_ONLY, release, llvm-mingw) from fix/virtual-pool-drawn-page-map 4830ea7d (this change on e51e28e9); HUD A/B run 2 of 4 (this PR) |
Windows 11 Pro 26200, fullscreen | RTX 5080 16 GB, 610.88 | Ryzen 9 7900X3D, 64 GB | 2560x1440, DLAA, Reflex off, frame rate cap 90, frame generation off, VRAM budget 11 GB; RX_VK_PARALLEL=4, HOS_PIN=*=FFF; DualSense |
the main outdoor area next to Castle Town by day, walked into Castle Town and its market and back out | The HUD’s labels and D-pad cross drawn at the town gate. No upload over a drawn image line of the bug’s form for the HUD’s textures or render targets; 11 for the 640x360 NVN 0x14 image the texture budget remakes. No validation, device-lost or VK_ERROR line; closed from the pause menu |
| 2026-11-13 | game folder built from main e51e28e9 (longshot e51e28e9 (source 8dd02aab)); longshot_vk.dll built alone (RX_VK_ONLY, release, llvm-mingw) from main e51e28e9 (the same file as the full build’s); HUD A/B run 1 of 4 (main) |
Windows 11 Pro 26200, fullscreen | RTX 5080 16 GB, 610.88 | Ryzen 9 7900X3D, 64 GB | 2560x1440, DLAA, Reflex off, frame rate cap 90, frame generation off, VRAM budget 11 GB; RX_VK_PARALLEL=4, HOS_PIN=*=FFF; DualSense |
the main outdoor area next to Castle Town by day, walked into Castle Town and its market and back out | At the town gate the R1 and R2 labels and the D-pad cross are not drawn, on the first try. upload over a drawn image lines of the bug’s form: the six label and cross textures at one present (7002), 4 other HUD textures, 4 render targets. No validation, device-lost or VK_ERROR line; closed from the pause menu |
| 2026-11-13 | game v3.0.0-rc.3-4-gfdc2840b (main fdc2840b, --fsr); renderer built alone (RX_VK_ONLY, llvm-mingw, LONGSHOT_FSR=1) from fix/psychov35-fsr4 ea988050 |
Windows 11 Pro 26300, fullscreen | RX 9070 XT 16 GB, Adrenalin 26.10.2 | Ryzen 7 9800X3D, 32 GB | 2560x1440, HDR10 (white 200, peak 700 nits), HDR Native, HDR tone mapper Vanilla, AMD FSR 4 Native AA (FSR 4.1.2), Sharpness 50; RX_VK_PSYCHOV35_DLSS=1 |
Forest Temple, indoors; cold start, about a minute standing | 600 of 600 FSR 4 frames in 6 of 6 complete intervals; no PsychoV35 line (default path unchanged) |
| 2026-11-13 | game v3.0.0-rc.3-4-gfdc2840b (main fdc2840b, --fsr); renderer built alone (RX_VK_ONLY, llvm-mingw, LONGSHOT_FSR=1) from fix/psychov35-fsr4 ea988050 |
Windows 11 Pro 26300, fullscreen | RX 9070 XT 16 GB, Adrenalin 26.10.2 | Ryzen 7 9800X3D, 32 GB | 2560x1440, HDR10 (white 200, peak 700 nits), HDR Native, HDR tone mapper PsychoV35, AMD FSR 4 Native AA (FSR 4.1.2), Sharpness 50; RX_VK_PSYCHOV35_DLSS=1 |
Forest Temple, indoors; cold start, about a minute standing | PsychoV35 FSR 4: RGBA16F linear route, FSR 4 made for HDR, linear colour; 600 of 600 FSR 4 frames in 8 of 8 complete intervals. Smooth edges, Sharpness row works, HDR brightness right (#226 fixed) |
| 2026-11-13 | game v3.0.0-rc.3-4-gfdc2840b (main fdc2840b, --fsr); renderer built alone (RX_VK_ONLY, llvm-mingw, LONGSHOT_FSR=1) from fix/psychov35-fsr4 ea988050 |
Windows 11 Pro 26300, fullscreen | RX 9070 XT 16 GB, Adrenalin 26.10.2 | Ryzen 7 9800X3D, 32 GB | 2560x1440, HDR10 (white 200, peak 700 nits), HDR Native, HDR tone mapper PsychoV35, AMD FSR 4 Native AA (FSR 4.1.2), Sharpness 50; RX_VK_PSYCHOV35_DLSS=0 |
Forest Temple, indoors; cold start, about a minute standing | the game's upscale runs (no scene camera this frame); 0 FSR 4 frames in every interval. Jagged, oversharpened, Sharpness row without effect (as on main) |
| 2026-11-13 | game v3.0.0-rc.3-4-gfdc2840b (main fdc2840b, --fsr); renderer built alone (RX_VK_ONLY, llvm-mingw, LONGSHOT_FSR=1) from fix/psychov35-fsr4 1ba927d0 (the same change on #108’s branch) |
Windows 11 Pro 26300, fullscreen | RX 9070 XT 16 GB, Adrenalin 26.10.2 | Ryzen 7 9800X3D, 32 GB | 2560x1440, HDR10 (white 200, peak 700 nits), HDR Native, HDR tone mapper PsychoV35, AMD FSR 4 Native AA (FSR 4.1.2), Sharpness 100; RX_VK_PSYCHOV35_DLSS=0 |
Forest Temple, indoors; cold start, about a minute standing | As the run above: the game’s upscale, 0 FSR 4 frames |
| 2026-11-13 | game v3.0.0-rc.3-4-gfdc2840b (main fdc2840b, --fsr); renderer built alone (RX_VK_ONLY, llvm-mingw, LONGSHOT_FSR=1) from fix/psychov35-fsr4 1ba927d0 (the same change on #108’s branch) |
Windows 11 Pro 26300, fullscreen | RX 9070 XT 16 GB, Adrenalin 26.10.2 | Ryzen 7 9800X3D, 32 GB | 2560x1440, HDR10 (white 200, peak 700 nits), HDR Native, HDR tone mapper PsychoV35, switched to Vanilla and back during play, AMD FSR 4 Native AA (FSR 4.1.2), Sharpness 100; RX_VK_PSYCHOV35_DLSS=1; HDR Native / Extended / Native during play |
Forest Temple, indoors; cold start, about a minute standing, then moving between areas | FSR 4 frames: 19 of 27 complete intervals at 600; 595 to 599 around the tone mapper and HDR switches (1 to 3 frames of unexpected scene colour format each, then the route again), 532 and 522 after an area change, one 351 not explained by the log; one interval of FSR 4 (not Native AA) at 600 |
| 2026-11-13 | host and game folder built from main 000f19fb; longshot_vk.so built alone from main 000f19fb (before) and from this change on it, ffee45f4 (after), same flags |
SteamOS 3.9.5 (kernel 7.2.7-valve2), headless | Steam Deck OLED: AMD Custom GPU 0932 (RADV VANGOGH), Mesa 26.2.0-devel | AMD Custom APU 0932 (4 cores, 8 threads), 16 GB | Minimum preset (RX_VK_LOW_SPEC=1), handheld profile (HOS_HANDHELD=1), the game’s own upscaler, RX_VK_VRAM_MB=2400, RX_VK_TEX_MAX_DIM=1024, RX_VK_DESC_SLOTGEN=1, RX_VK_LOAD=create, HOS_RESERVE_CORE=0 |
a save loaded by script, standing still on a heavy view (about 2,500 draws a frame); the main outdoor area (about 1,600 draws) | Renderers alternated, each run measured over exactly 720 presents after 3 log windows; the render thread’s cycles and instructions by perf stat on that thread (user space). Heavy view, 4 runs each, medians: 77.04 → 73.57 M instructions and 88.30 → 86.08 M cycles a frame, replay 31.67 → 30.45 ms, 23.83 → 23.35 fps (inside the runs’ spread, 23.3 to 24.3 on both sides: the frame is about 42 ms, the rest is waits). the main outdoor area, 2 runs each: 47.01 → 44.70 M instructions, 60.65 → 58.85 M cycles, replay 19.95 → 19.43 ms, 22.58 → 22.61 fps (the GPU’s 24.7 ms a frame is the limit). No errors in 12 runs |
| 2026-11-13 | v3.0.0-rc.3 game (source cb01b5a3); its renderer built alone (RX_VK_ONLY, RX_RELEASE, llvm-mingw, NVIDIA DLSS SDK) from main 0edc137 and from perf/low-vram-mirror-budget d27a462, swapped between runs: main, PR, main, PR |
Windows 11 Home (26200), 2048x1152 | RTX 3060 Laptop 6 GB, 621.14 | Ryzen 7 6800HS, 16 GB (15 GB usable) | Quality preset Auto (this PC), VRAM budget 3.6 GB (Low), NVIDIA DLSS at 33% (683x384) in every run, frame generation off, no cap; no RX_VK_MIRROR_MB |
the main outdoor area, walking, about 2 minutes a run | Median of the in-field “presents” lines: main 19.5 and 21.4 fps (20.1 over both), mirrors 512 MB, a 224 MB pool “stays in system memory”; PR 43.9 and 41.4 fps (42.2 over both), mirrors 1498 MB (a quarter of the 5992 MB the driver reports, which also caps Low’s 1536), the largest pool left out 24 MB. No Vulkan errors in either |
| 2026-11-13 | v3.0.0-rc.3 (source cb01b5a3), the launcher’s build, with RX_VK_MIRROR_MB=512 (main’s mirrors for a 6 GB card on the Low (3.6 GB) VRAM budget) against RX_VK_MIRROR_MB=1536 (perf/low-vram-mirror-budget’s), alternated: main, PR, main, PR |
Windows 11 Home (26200), 2048x1152 | RTX 3060 Laptop 6 GB, 621.14 | Ryzen 7 6800HS, 16 GB (15 GB usable) | Quality preset Auto (this PC), VRAM budget 3.6 GB, NVIDIA DLSS, Render resolution settled at 33% (683x384) in every run (the first main run started at 50% and stepped down), frame generation off, no cap | the main outdoor area, walking, about 2 minutes a run | Median of the in-field “presents” lines: main 17.7 and 20.7 fps (18.8 over both), GPU frame 34-37 ms, a 224 MB pool “stays in system memory” (mirrors 444-488 MB of 512); PR 36.3 and 39.5 fps (38.6 over both), GPU frame 14 ms, 114-120 pools mirrored (1536 MB) |
| 2026-11-13 | Renderer 6b8bc562 / cdee3127; host 7afe6ae1 / 7cfff061; pr-psycho-dlaa | Windows 11 Home 26300, windowed | RTX 4070 12 GB, 621.01 | i7-13700F, 32 GB | 2560x1440 output, 2560x1421 window; DLAA 2560x1440, K, sharpness 0; Native HDR, PsychoV35; white 200 / peak 400 nits for HDR; NVIDIA FG 2x; RX_VK_PSYCHOV35_DLSS=1 | Copied the main outdoor area child save, cold start, 40 s standing after settling | Last two complete 600-frame intervals: 600/600 DLSS/DLAA evaluations and NVIDIA FG sets; zero missing-input fallbacks. Correctness counts, no FPS claim |
| 2026-11-13 | Renderer 6b8bc562 / cdee3127; host 7afe6ae1 / 7cfff061; pr-psycho-on2 | Windows 11 Home 26300, windowed | RTX 4070 12 GB, 621.01 | i7-13700F, 32 GB | 2560x1440 output, 2560x1421 window; DLSS Quality 1707x960, K, sharpness 0; Native HDR, PsychoV35; white 200 / peak 400 nits for HDR; NVIDIA FG 2x; RX_VK_PSYCHOV35_DLSS=1 | Copied the main outdoor area child save, cold start, 40 s standing after settling | Last two complete 600-frame intervals: 600/600 DLSS/DLAA evaluations and NVIDIA FG sets; zero missing-input fallbacks. Correctness counts, no FPS claim |
| 2026-11-13 | Renderer dd02c6cc / cb01b5a3; host 7afe6ae1 / 7cfff061; main-psycho-off2 | Windows 11 Home 26300, windowed | RTX 4070 12 GB, 621.01 | i7-13700F, 32 GB | 2560x1440 output, 2560x1421 window; DLSS Quality 1707x960, K, sharpness 0; Native HDR, PsychoV35; white 200 / peak 400 nits for HDR; NVIDIA FG 2x; switch unset | Copied the main outdoor area child save, cold start, 40 s standing after settling | Last two complete 600-frame intervals: 0/600 NVIDIA FG, 600 missing-input fallbacks; game network used. Correctness counts, no FPS claim |
| 2026-11-13 | Renderer 6b8bc562 / cdee3127; host 7afe6ae1 / 7cfff061; pr-vanilla-sdr | Windows 11 Home 26300, windowed | RTX 4070 12 GB, 621.01 | i7-13700F, 32 GB | 2560x1440 output, 2560x1421 window; DLSS Quality 1707x960, K, sharpness 0; SDR, Vanilla; white 200 / peak 400 nits for HDR; NVIDIA FG 2x; switch unset | Copied the main outdoor area child save, cold start, 40 s standing after settling | Last two complete 600-frame intervals: 600/600 DLSS/DLAA evaluations and NVIDIA FG sets; zero missing-input fallbacks. Correctness counts, no FPS claim |
| 2026-11-13 | Renderer dd02c6cc / cb01b5a3; host 7afe6ae1 / 7cfff061; main-vanilla-sdr | Windows 11 Home 26300, windowed | RTX 4070 12 GB, 621.01 | i7-13700F, 32 GB | 2560x1440 output, 2560x1421 window; DLSS Quality 1707x960, K, sharpness 0; SDR, Vanilla; white 200 / peak 400 nits for HDR; NVIDIA FG 2x; switch unset | Copied the main outdoor area child save, cold start, 40 s standing after settling | Last two complete 600-frame intervals: 600/600 DLSS/DLAA evaluations and NVIDIA FG sets; zero missing-input fallbacks. Correctness counts, no FPS claim |
| 2026-11-13 | Renderer dd02c6cc / cb01b5a3; host 7afe6ae1 / 7cfff061; main-psycho-off | Windows 11 Home 26300, windowed | RTX 4070 12 GB, 621.01 | i7-13700F, 32 GB | 2560x1440 output, 2560x1421 window; DLSS Quality 1707x960, K, sharpness 0; Native HDR, PsychoV35; white 200 / peak 400 nits for HDR; NVIDIA FG 2x; switch unset | Copied the main outdoor area child save, cold start, 40 s standing after settling | Last two complete 600-frame intervals: 0/600 NVIDIA FG, 600 missing-input fallbacks; game network used. Correctness counts, no FPS claim |
| 2026-11-13 | Renderer 6b8bc562 / cdee3127; host 7afe6ae1 / 7cfff061; pr-psycho-on | Windows 11 Home 26300, windowed | RTX 4070 12 GB, 621.01 | i7-13700F, 32 GB | 2560x1440 output, 2560x1421 window; DLSS Quality 1707x960, K, sharpness 0; Native HDR, PsychoV35; white 200 / peak 400 nits for HDR; NVIDIA FG 2x; RX_VK_PSYCHOV35_DLSS=1 | Copied the main outdoor area child save, cold start, 40 s standing after settling | Last two complete 600-frame intervals: 600/600 DLSS/DLAA evaluations and NVIDIA FG sets; zero missing-input fallbacks. Correctness counts, no FPS claim |
| 2026-11-13 | Renderer 6b8bc562 / cdee3127; host 7afe6ae1 / 7cfff061; pr-vanilla-hdr | Windows 11 Home 26300, windowed | RTX 4070 12 GB, 621.01 | i7-13700F, 32 GB | 2560x1440 output, 2560x1421 window; DLSS Quality 1707x960, K, sharpness 0; Native HDR, Vanilla; white 200 / peak 400 nits for HDR; NVIDIA FG 2x; switch unset | Copied the main outdoor area child save, cold start, 40 s standing after settling | Last two complete 600-frame intervals: 600/600 DLSS/DLAA evaluations and NVIDIA FG sets; zero missing-input fallbacks. Correctness counts, no FPS claim |
| 2026-11-13 | Renderer dd02c6cc / cb01b5a3; host 7afe6ae1 / 7cfff061; main-vanilla-hdr | Windows 11 Home 26300, windowed | RTX 4070 12 GB, 621.01 | i7-13700F, 32 GB | 2560x1440 output, 2560x1421 window; DLSS Quality 1707x960, K, sharpness 0; Native HDR, Vanilla; white 200 / peak 400 nits for HDR; NVIDIA FG 2x; switch unset | Copied the main outdoor area child save, cold start, 40 s standing after settling | Last two complete 600-frame intervals: 600/600 DLSS/DLAA evaluations and NVIDIA FG sets; zero missing-input fallbacks. Correctness counts, no FPS claim |
| 2026-11-13 | fix/occlusion-queries 3f67aad0 (main 839e85ff + this change: the renderer built alone and put in a copy of main’s folder) against main 839e85ff, alternated | Linux (CachyOS, kernel 7.3.3), a private headless sway 2048x1152, scripted input, no sound device | RTX 3090 24 GB, 620.70.06 | Ryzen 9 7900X, 30 GB (swapping in every run) | Default options, shaders prepared, cap 240; RX_VK_OCC_LOG=1 both sides |
The pristine starting-area save: the starting forest standing 40 s, then the camera turning at full speed 30 s; Castle Town Market (debug teleport) the same; 4 runs each | Medians of the runs’ medians, main / this change. starting-area standing: 2,862 / 2,632 draws, GPU 10.1 / 10.1 ms, frame 16.54 / 15.95 ms. starting-area turning: 1,928 / 1,767 draws, GPU 9.00 / 9.25 ms, frame 14.43 / 14.35 ms. Market standing: 2,926 / 2,464 draws, GPU 8.60 / 8.07 ms, frame 16.55 / 15.34 ms. Market turning: GPU 5.38 / 5.42 ms, frame 12.84 / 12.82 ms, longest frame 71 / 46 ms. 22-60% of reports read 0, written 1-4 recordings after the report, 0 left visible by a draw that did not run. Nothing missing in screenshots at the same spot |
| 2026-11-13 | fix/occlusion-visible-unless-known (main fdc2840b + this change: the renderer built alone, RX_VK_ONLY, as main’s for the comparison, in copies of one main folder) against main fdc2840b, alternated |
Linux (CachyOS, kernel 7.3.3), a private headless sway 2048x1152, scripted input, no sound device | RTX 3090 24 GB, 620.70.06 | Ryzen 9 7900X, 30 GB (swapping in every run) | Default options, shaders prepared, cap 240; RX_VK_OCC_LOG=1 both sides |
The pristine starting-area save: the starting forest standing 40 s, then turning 30 s; Castle Town Market (debug teleport) the same; 4 runs each (P M M P, then M X M X with X the same code), 2 with RX_VK_OCC_QUERIES=0 |
Medians of the runs’ values, main / this change / queries off. starting-area standing: 2,600 / 2,614 / 2,864 draws, frame 15.47 / 15.20 / 15.32 ms. Market standing: 2,433 / 2,440 / 2,928 draws, GPU 7.53 / 7.40 / 7.95 ms, frame 15.67 / 16.12 / 17.34 ms. Market turning: frame 13.36 / 13.60 / 13.05 ms. Validation layer, one short session each: main 02741 and 00807 to the cap, this change none. 0 flashes either side. 0 queries left unbegun in play |
| 2026-11-12 | fix/load-wait-rule 3a297868 (main 4dd972a8 + this change; the commit before the merge of main that brought the rename: same code, old hookshot_ file names), with a test-only switch added that is not part of the change. The merged branch has not run in a game |
Linux, a private headless sway, scripted input | RTX 3090, - | - | A folder of its own prepared by --prepare-shaders, RX_VK_LOAD=create; nothing injected |
A short session: the first area, a teleport, a second area; three sessions (the first start after a preparation, one between the injected ones below, the first start after the second preparation) | Waits the rule counts (over 20 ms): 27 ms (8 first uses met a load, 31 ms in all), 0 ms (23 first uses, 32 ms in all, the longest 5.9 ms), 0 ms (11 first uses, 6.6 ms in all). No line from the safety net; no marker (longshot_load_start.txt) and no .waited file after any of them |
| 2026-11-12 | as above | as above | as above | as above | as above, with long waits injected: the test-only switch, which is not part of the change, makes the first N first uses that meet an unfinished load take a set time longer, inside the time the safety net measures. The rule’s own code ran unchanged; no long wait could be produced on this PC without it | as above, six sessions in the order given; “next start” is a start without RX_VK_LOAD, to see what the default picks |
One wait of 300 ms: no line, no file, next start create. Two of 300 ms (601 ms counted): “over 500 ms. Once is let pass”, .waited left. One of 300 ms right after: no line, no file (the strike was taken and not renewed), next start create. Two of 300 ms (600 ms): “Once is let pass”, .waited. Waits of 600 ms right after: “over 500 ms, as the session before this one was”, the marker, next start start. Two of 300 ms with the marker there (600 ms): “Once is let pass”, the marker and .waited |
| 2026-11-12 | as above | as above | as above | as above | The marker and .waited both in the folder |
A preparation (--prepare-shaders), no game |
Both files removed; the next start picks create |
| 2026-11-12 | as above | as above | as above | as above | RX_VK_LOAD=create, one wait of 2,100 ms injected by the same test-only switch (not part of the change) |
A short session, as above | 2,136 ms counted: “over 2000 ms, so the next start loads every listed program before the title again”, the marker; the next start picks start |
| 2026-11-12 | as above | as above | as above | as above | RX_VK_LOAD=create, no load workers (RX_VK_LOAD_WORKERS=0), nothing injected |
A short session, as above | 0 ms counted: 1,157 first uses ran their load themselves, 1,743 ms in all, the longest 16.2 ms. No line, no file (the rule before this change, over 500 ms in all, would have left the marker) |
| 2026-11-12 | main 4dd972a8 + this change, HOS_VIDEO_INPUT_LIMIT=0 (as main) against the default (16) |
Linux (CachyOS, kernel 7.3.3), a private headless sway, scripted input, no sound device (HOS_AUDIO=0) |
RTX 3090 24 GB, 620.70.06 | Ryzen 9 7900X, 30 GB (the PC was swapping in every run) | 2048x1152, default options, cap 60 | A new game from an empty data folder: the first movie (A after 11 s, about 820 pictures of it), then the second (9,254 pictures) to its end | Three runs each, interleaved. Pictures skipped, first movie: 244, 247, 251 without the limit; 8, 198, 1 with it. Second movie: 123, 700, 98 without; 0, 502, 0 with. In the second run of each side the PC swapped hard and frames were late (177 and 146 frames over 33 ms; 0 to 2 in the others): those skips are the late frames’. Waits of over 90 ms between two pictures in a movie’s first 5 s: 44, 43, 43 and 20, 21, 18 without; 1, 0, 0 and 0, 9, 0 with. The second movie ends within 12 ms of its clock on both sides |
| 2026-11-13 | Hookshot fix/commit-limit 905cb5e1 (NotRexed/Hookshot#2972, ported here as fix/commit-limit), against Hookshot main e0e60920, alternated | Windows | RTX 4060 Ti, - | -, 31 GB (115 GB commit limit) | 3440x1440, muted, the commit rule idle (default RX_VK_COMMIT_RESERVE_MB) |
Route run: starting-area, the main outdoor area, the village, 2 runs each | This change: 53 and 57 fps median, p99 40.5 and 36.1 ms, private commit at end 28.9 and 28.6 GB, no commit limit near line (the rule never fired). Main: 51 and 50 fps, p99 43.2 and 42.1 ms, 33.4 and 34.8 GB. Run-to-run variation, not the change: it costs nothing when idle |
| 2026-11-13 | as above, this change only | as above | as above | as above | as above, RX_VK_COMMIT_RESERVE_MB=38397 (the rule forced, about 25 GB of the game’s commit before it kicks in) |
as above, 1 run | 54 fps median, p99 38.7 ms, max 881 ms (the run before, rule not firing: 58, 35.6, 881). 52 commit limit near lines, texture budget down by the shortfall each time (e.g. 3661 to 2812 MB), headroom 37.4 to 38.4 GB; private commit peak 26.7 GB (30.3 without the rule), VRAM blocks peak 9.0 GB (12.9). The 5 s wait at the limit itself not reached |
| 2026-11-13 | v2.1.0-rc.3-313-g9b6072c + this PR, the split-submission fix (#189) and compute stages let go of the 32-lane pin as https://github.com/NotRexed/Hookshot/pull/2564 does, all three in each of the 12 runs (one build), for https://github.com/NotRexed/Hookshot/issues/999; the same patch rebased on main e0e6092 for the PR, built from source (llvm-mingw) | Windows 11 Pro (26300), 1920x1080, scripted input | Intel Arc Graphics (Core Ultra 7 155H iGPU, Meteor Lake), 32.0.101.9175 | Core Ultra 7 155H, 32 GB | Render resolution 4, handheld profile; RX_VK_PASS_TIMING=2, a screenshot every 60 frames; old/new alternated, the game closed normally |
the starting forest outside the player character’s house, standing, 300 s | No igvk64.dll+0x79ddf3 crash in 12 runs (main 9b6072c without them: 4 of 6): the three changes together, not this one alone. The background save ran 3 times and was put off once. Device lost in compute programs in 4 of 12, as on plain main |
| 2026-11-13 | v2.1.0-rc.3-313-g9b6072c + this PR, the pipeline cache fetch lock (https://git.magicbean.dev/MagicBeanDev/Longshot/pulls/190) and compute stages let go of the 32-lane pin as https://github.com/NotRexed/Hookshot/pull/2564 does, one build for every run, for https://github.com/NotRexed/Hookshot/issues/999; the same patch rebased on main e0e6092 for the PR, built from source (llvm-mingw) | Windows 11 Pro (26300), 1920x1080, scripted input | Intel Arc Graphics (Core Ultra 7 155H iGPU, Meteor Lake), 32.0.101.9175 | Core Ultra 7 155H, 32 GB | Render resolution 4, handheld profile; RX_VK_PASS_TIMING=2, a screenshot every 60 frames; old/new alternated, the game closed normally; old: the same build with RX_VK_SPLIT_UBFLUSH=1 |
the starting forest outside the player character’s house, standing, 300 s | White blocks over the forest in 9, 7 and 11 frames of the old runs that ran on, 0 in every new run (an earlier series of this change alone: 7, 1, 1 against 0, 0, 0); fps 4.7-5.7 old, 5.7-6.0 new. Device lost in compute programs (ab7699ee, aba5e6af, 499dbfd0) in 2 of 8 old and 5 of 8 new runs over both series. The same programs lose it with the float-add loop on main and the other Intel builds (12 of 18 runs in https://git.magicbean.dev/MagicBeanDev/Longshot/pulls/191’s series; 499dbfd0 is a float add to a storage buffer, ab7699ee likely one), none with https://git.magicbean.dev/MagicBeanDev/Longshot/pulls/191’s native add (0 of 6), and none in 6 route runs of a build with this change and https://git.magicbean.dev/MagicBeanDev/Longshot/pulls/191 together (13/11): most likely the loop’s, not this change’s, though 5 of 8 against 2 of 8 is not explained |
| 2026-11-13 | v2.1.0-rc.3-330-g82de567a game (main); its renderer built alone (RX_VK_ONLY, llvm-mingw, NVIDIA DLSS SDK) from main d4a69456 and from fix/dlss-preset-settle e8c02084, swapped between runs |
Windows 11 Pro (26300), headless (offscreen 3440x1440) | RTX 5080 16 GB, 621.24 | - | Upscaler NVIDIA DLSS, Render resolution Native (DLAA), output 3440x1440, frame rate cap 240, frame generation off; the DLSS model row set through the control bridge: K, then M and L 11 frames apart, then M and K, then M alone | the starting forest (child, the debug teleport’s arrival), standing; 4 runs, before/after/after/before | Steady, both builds: K 7.4-7.6 ms of GPU a frame (80-84 fps, 140-145 W), L 10.6-10.7 ms (66-67 fps, 198-202 W), M 9.7-9.9 ms (70-72 fps, 186-188 W); the last before run’s scene lighter (K 6.8, L 9.6 ms). K to M to L: before, features for M and L, DLSS’s own 1419 MB, VRAM up 1254 and 1143 MB in 15 s (nvidia-smi); after, L’s alone, 813 MB, up 637 and 637 MB. L to M to K: before up 806 and 827 MB, after 126 and 189 MB. M alone: 657-722 MB both. Issue #17 |
| 2026-11-13 | main @ 316f9319; historical hold-to-toggle run 1 |
NixOS 26.11 / Linux 7.2.7; private headless Sway 1.12 + Xwayland | RX 7900 XT, RADV Mesa 26.2.3 | Ryzen 5 7600X, 32 GB | 1920×1080, 100% render resolution, FG/HDR Off, Uncapped, VSync Auto; stock; mesh fallback, udmabuf, audio off | Castle Town wooden door, copied save | Last five presents readings: 54.1, 55.4, 54.1, 53.6, 53.7; median 54.10 FPS. Same copied save, separate warm-up. Recorded before the review fixes; not a measurement of the current head. |
| 2026-11-13 | this PR @ 0f4b3374; historical hold-to-toggle run 2 |
NixOS 26.11 / Linux 7.2.7; private headless Sway 1.12 + Xwayland | RX 7900 XT, RADV Mesa 26.2.3 | Ryzen 5 7600X, 32 GB | 1920×1080, 100% render resolution, FG/HDR Off, Uncapped, VSync Auto; ZL Toggle On; no held button during FPS samples; mesh fallback, udmabuf, audio off | Castle Town wooden door, copied save | Last five presents readings: 54.4, 53.7, 53.1, 53.0, 52.8; median 53.10 FPS. Same copied save, separate warm-up. Recorded before the review fixes; not a measurement of the current head. |
| 2026-11-13 | main @ 316f9319; historical hold-to-toggle run 3 |
NixOS 26.11 / Linux 7.2.7; private headless Sway 1.12 + Xwayland | RX 7900 XT, RADV Mesa 26.2.3 | Ryzen 5 7600X, 32 GB | 1920×1080, 100% render resolution, FG/HDR Off, Uncapped, VSync Auto; stock; mesh fallback, udmabuf, audio off | Castle Town wooden door, copied save | Last five presents readings: 54.7, 53.9, 53.7, 54.5, 53.9; median 53.90 FPS. Same copied save, separate warm-up. Recorded before the review fixes; not a measurement of the current head. |
| 2026-11-13 | this PR @ 0f4b3374; historical hold-to-toggle run 4 |
NixOS 26.11 / Linux 7.2.7; private headless Sway 1.12 + Xwayland | RX 7900 XT, RADV Mesa 26.2.3 | Ryzen 5 7600X, 32 GB | 1920×1080, 100% render resolution, FG/HDR Off, Uncapped, VSync Auto; ZL Toggle On; no held button during FPS samples; mesh fallback, udmabuf, audio off | Castle Town wooden door, copied save | Last five presents readings: 54.4, 54.2, 54.1, 53.8, 54.7; median 54.20 FPS. Same copied save, separate warm-up. Recorded before the review fixes; not a measurement of the current head. |
| 2026-11-13 | rev137-bc0c1292 (source f94c58fd); Git bc0c1292, hold-to-toggle with main f3fe1763 merged |
NixOS 26.11, Linux 7.2.7; private headless Sway 1.12 + Xwayland | RX 7900 XT, RADV Mesa 26.2.3 | Ryzen 5 7600X, 32 GB | 1920×1080, render 100%, game upscaler, FG/HDR Off, Native mouse, Toggle ZR/ZL On, mesh fallback, udmabuf, audio off; copied save | the castle Town (Market_Child), indoors by the wooden door |
10m06s functional session, clean control-socket exit. ZR press at frames 21466–21481 stayed latched after release; 20 relative XTest mouse events of +20 horizontal counts moved the slingshot aim from the wall to the door without firing (14 seeds before/after; queue empty at frame 23052). ZL press at 18803–18818 held the shield with queue empty at 18946; press at 18954–18969 released it. Opening Insert cleared a later shield latch. Inspected Controls: ZL/ZR On, Plus/Minus disabled, revised help. No crash/device-lost/hang marker. No FPS comparison: another isolated CPU build ran concurrently. |
| 2026-11-13 | feat/mono-audio review follow-up on main d9648863; mono-output-review (source 528a52f6) |
NixOS 26.11 / Linux 7.2.7, private headless Sway/Xwayland | RX 7900 XT 20 GB, RADV Mesa 26.2.3 | Ryzen 5 7600X, 32 GB | 1920×1080, 100% render resolution, game upscaler, FG/HDR Off, Uncapped, VSync Auto; compute mesh fallback; HOS_AUDIO=0; shader prediction budget 60 s |
Castle Town wooden door, copied save, stationary; one live Stereo → Mono → Stereo → Mono run, 20 s per sample after settling | Mono: 0 unequal L/R frames out of 962,048 / 963,072, with 959,216 / 960,560 non-silent frames. Stereo: 960,175 / 960,120 unequal out of 962,048 each. Last-five native FPS medians: Stereo 56.0 / 56.6, Mono 56.6 / 56.4; medians of the two 56.3 / 56.5 (one scene, not a general cost claim). Audio output row on Mono and its live/restart help inspected in a screenshot; saved audio_output=3; no separate Mono row. |
| 2026-11-13 | rev138-2f42b40e01e3 (source a5035217), fix/damage-taken-review 2f42b40e |
NixOS 26.11, Linux 7.2.7; private headless Sway 1.12/Xwayland | RX 7900 XT 20 GB, RADV Mesa 26.2.3 | Ryzen 5 7600X, 32 GB | 1920x1080, game’s upscaler, FG/HDR Off; mesh fallback, audio disabled; copied save | Castle Town roof; first gate-harness attempt | Excluded from gate proof: a three-quarter hit removed 12 (16 to 4) with Quadruple saved, but the harness did not record Debug-tools state at impact. The later run below records it at the damage hook. No product failure claimed from this timing attempt |
| 2026-11-13 | rev138-2f42b40e01e3 (source a5035217), fix/damage-taken-review 2f42b40e |
NixOS 26.11, Linux 7.2.7; private headless Sway 1.12/Xwayland | RX 7900 XT 20 GB, RADV Mesa 26.2.3 | Ryzen 5 7600X, 32 GB | 1920x1080, game’s upscaler, FG/HDR Off; mesh fallback, audio disabled; copied save | Castle Town roof; paired controlled falls | With Quadruple saved, tools Off: raw hit 3, life 16 to 13; tools On again: raw hit 3, life 13 to 1. Quarter-heart units. Read-only hardware breakpoint confirmed tools state and saved choice at each hit; bridge read life after impact. Two setup attempts that did not rise were excluded. Screenshots inspected; no FPS claim |
| 2026-11-13 | rev138-2f42b40e01e3 (source a5035217), fix/damage-taken-review 2f42b40e |
NixOS 26.11, Linux 7.2.7; private headless Sway 1.12/Xwayland | RX 7900 XT 20 GB, RADV Mesa 26.2.3 | Ryzen 5 7600X, 32 GB | 1920x1080, game’s upscaler, FG/HDR Off; Debug tools On, Heart protection Off for measured damage; mesh fallback, audio disabled; copied save | Castle Town roof; four controlled four-quarter hits | Normal: 16 to 12; Double: 12 to 4; Half: 4 to 2, then 2 to 0 and the death animation. Raw hit 4 confirmed at every impact by read-only hardware breakpoint; no guest-memory writes. Two protected height-calibration falls (raw 5 and 4) left life at 16. Screenshots inspected; no FPS claim |
| 2026-11-12 | d4a69456 + PsychoV35 DLSS port, renderer source 878c957c; host 7afe6ae1; DLAA cold start | Windows 11 Home, build 26300 | RTX 4070 12 GB, NVIDIA 621.01 | i7-13700F, 32 GB | 2560x1440 DLAA (RX_VK_DLSS=2), K; Native HDR PsychoV35, white 200 / peak 400 nits; sharpness 0; NVIDIA FG 2x; RX_VK_PSYCHOV35_DLSS=1 | Copied the main outdoor area child save, 30 s standing after settling | RGBA16F input/output, HDR feature; repeated 600/600 DLAA evaluations and 600/600 NVIDIA FG sets with zero missing-input fallbacks in steady intervals. Method and limits |
| 2026-11-12 | Same renderer and host; extra tone-mapper transition run | Windows 11 Home, build 26300 | RTX 4070 12 GB, NVIDIA 621.01 | i7-13700F, 32 GB | 2560x1440 output, DLSS Quality K; Native HDR PsychoV35, white 200 / peak 400 nits; sharpness 0; NVIDIA FG 2x; switch on | Same copied save, 30 s standing; Vanilla then PsychoV35 | DLSS and NVIDIA FG run; tone-mapper changes recover. Bridge refused Native/DLAA setting requests without changing the setting; DLAA was subsequently tested by a separate cold start |
| 2026-11-12 | d4a69456 + PsychoV35 DLSS port, renderer source 878c957c; host 7afe6ae1; on run 1 | Windows 11 Home, build 26300 | RTX 4070 12 GB, NVIDIA 621.01 | i7-13700F, 32 GB | 2560x1440 output, 2560x1421 window; DLSS Quality 1707x960, K; Native HDR PsychoV35, white 200 / peak 400 nits; sharpness 0; NVIDIA FG 2x; RX_VK_PSYCHOV35_DLSS=1 | the main outdoor area child save, cold start, 30 s standing after settling | RGBA16F input/output accepted, HDR DLSS feature; repeated 600/600 DLSS evaluations and 600/600 NVIDIA FG sets, zero missing-input fallbacks in steady intervals. Method and limits |
| 2026-11-12 | Same renderer and host; off run 1 | Windows 11 Home, build 26300 | RTX 4070 12 GB, NVIDIA 621.01 | i7-13700F, 32 GB | Same scene/settings as on run 1, RX_VK_PSYCHOV35_DLSS=0 | Same copied save, cold start, 30 s standing | No DLSS feature; repeated 0/600 NVIDIA FG sets and 600/600 missing-input fallbacks in steady intervals |
| 2026-11-12 | Same renderer and host; off run 2 | Windows 11 Home, build 26300 | RTX 4070 12 GB, NVIDIA 621.01 | i7-13700F, 32 GB | Same scene/settings as off run 1 | Same copied save, cold start, 30 s standing | Reproduced 0/600 NVIDIA FG sets and 600/600 missing-input fallbacks |
| 2026-11-12 | Same renderer and host; on run 2 | Windows 11 Home, build 26300 | RTX 4070 12 GB, NVIDIA 621.01 | i7-13700F, 32 GB | Same scene/settings as on run 1 | Same copied save, cold start, 30 s standing; then Vanilla/PsychoV35 switches and a camera turn | Reproduced 600/600 DLSS evaluations and NVIDIA FG sets in steady intervals; tone-mapper switching returned to RGBA16F, history reset; actor motion collected |
| 2026-11-12 | feat/auto-preset (the Auto (this PC) branch, on main), game and renderer built from it | Ubuntu 26.04.1, kernel 7.0, a private KWin virtual output 2560x1440 at 60 Hz (headless runs, scripted input) | RTX 4070 Ti 12 GB, 595.99.02 | Ryzen 7 9800X3D, 32 GB | Quality preset Auto (this PC); frame generation off; caps 60 and 90 (HOS_FPS_MAX), and 4K output set by hand for the step-down run |
The test save’s spawn point, then the starting forest and the main outdoor area | First start: tier 4 (High), DLSS Quality, output 2560x1440. Same PC again: kept. RX_VK_AUTO_PRESET_VRAM_MB=6144: Low, DLSS Performance, output 2048x1152; back to 12 GB: High again. A row changed: Custom, and Auto left it. In play: 1440p at 60, the GPU at 8.4 of 16.7 ms: kept; 4K at a 90 cap, frames missed (about 80 fps), the GPU at 10.6 of 11.1 ms: 67% -> 58% -> 50%; 8 GB faked at 60, the GPU at 8.3 ms: Balanced (58%) -> Quality (67%). No validation, device-lost or “not made” line |
| 2026-11-12 | v2.1.0-rc.3-256-g51da046-dirty (source 0212cb04): main with the frame generation maker shown, built from source (llvm-mingw, NVIDIA DLSS SDK with nvngx_dlssg.dll 310.9.16) |
Windows 11 Pro 25H2 (26200), 3440x1440 window | RTX 5060 Ti 16 GB, 617.42 | Core Ultra 7 270K Plus, 32 GB | Upscaler NVIDIA DLSS Quality, Frame generation by Auto, frame rate cap 240, V-Sync Auto | Fire Temple, a save, standing | 2x: the row Auto: DLSS FG, the FPS counter Base 63, DLSS FG at 126-129 FPS (DLSS-G made 600 of 600 presents). Frame generation Auto: Auto: Built-in FG, its line why, Base 71, Built-in FG at 240 FPS. The pause menu and the title screen: Auto: Built-in FG (no DLSS frame). Built-in chosen: Built-in FG. 6x, 4x, 2x, Auto in turn: no fault |
| 2026-11-12 | v2.1.0-rc.3-250-gdea9b646 (source 5821b893): main c544828f + this change (before #2711); against v2.1.0-rc.3-248-gc544828f (source 08d5a9cc), main | Windows 11, no window (unattended, scripted input) | RTX 4070 SUPER 12 GB, 617.42 | Ryzen 9 9950X3D (SMT off), 32 GB | 2560x1440, DLAA, default options; nothing else running, RX_VK_TEST_ALLOWED_MB=6900 (an 8 GB card’s allowance; this card cannot page) |
the main outdoor area, a scripted five-minute walk | This change: 58.9 fps, 27 of 29 VRAM lines past the allowance (worst +1,393 MB), no “room in the blocks: not squeezed” line. Main: 58.7 fps, 26 of 28 past (worst +1,641 MB), 26 such lines. The texture budget was already 0 MB on main there; what held the VRAM was the image pool’s blocks made at the start (#2438, fixed since by #2711): main with RX_VK_IMAGE_PREALLOC=0 was past the allowance in 0 of 29 lines, 59.3 fps. No device lost |
| 2026-11-12 | as above | as above | as above | as above | 2560x1440, DLAA, default options; nothing else running, RX_VK_TEST_ALLOWED_MB=6144 (a 6 GB card’s) |
Castle Town to the main outdoor area (an area change) | This change: 51.2 fps, 4 of 6 VRAM lines past the allowance (worst +2,465 MB), no “not squeezed” line. Main: 51.2 fps, 4 of 6 (worst +2,333 MB), 4 such lines; with RX_VK_IMAGE_PREALLOC=0 worst +799 MB. No device lost |
| 2026-11-12 | v2.1.0-rc.3-253-g2fb430bd (source e6840837): main c544828f + #2585 + this change’s first version, whose 8 ms bound was then the default (now RX_VK_TRANSLATE_WAIT_MS=8) |
Windows 11, no window (unattended, scripted input) | RTX 4070 SUPER 12 GB, 617.42 | Ryzen 9 9950X3D (SMT off), 32 GB | 2560x1440, DLAA, default options; nothing else running, RX_VK_TEST_ALLOWED_MB=10200; RX_VK_NO_PROGRAM_CACHE=1 (a cold program bank); the 8 ms bound against RX_VK_TRANSLATE_WAIT_MS=-1 (waited out, as main) |
From the title into Castle Town, then out to the main outdoor area | Two runs each. Bound: 50.0 / 47.4 fps, frames over 100 ms 22 / 23, over 250 ms 8 / 6; 1.80 / 1.70 s waiting for shaders; 20,592 / 23,208 draws left out. Waited out: 44.9 / 45.3 fps, 39 / 34, 11 / 11; 4.94 / 4.85 s; 4,404 / 4,368 left out (pipelines still building). Screenshots every 60 frames: nothing missing; the HUD at the first frame after the load was missing waited out, present with the bound. No device lost |
| 2026-11-12 | as above | as above | as above | as above | as above, with the program bank warm | as above | 47.2 fps both ways, no waits for shaders and no draws left out either way |
| 2026-11-12 | v2.1.0-rc.3-259-g31ddd7a5 (source 0a02ef62): main edaef2a9 + this change and other open PRs | Windows 11, no window (unattended, scripted input) | RTX 4070 SUPER 12 GB, 617.42 | Ryzen 9 9950X3D (SMT off), 32 GB | 2560x1440, DLAA, RX_VK_COMPACT=1, RX_VK_COMPACT_PCT=85, RX_VK_ASTC_BC_ALL=1, RX_VK_IMAGE_BLOCK_MB=32; nothing else running, RX_VK_TEST_ALLOWED_MB=10200 |
Castle Town to the main outdoor area, in a folder of the same translator with the stamp file removed | The start dropped the 1,233 MB cache (“1 files made for another translator dropped (1233 MB)”) and the stage libraries were made again; the next start dropped nothing. No device lost |
| 2026-11-12 | v2.1.0-rc.3-154-g2e790fa9 (source 8c169744): main 15646d18 + this change; against v2.1.0-rc.3-153-g15646d18-dirty: main with a temporary measuring patch | Windows 11, no window (unattended, scripted input) | RTX 4070 SUPER 12 GB, 617.42 | Ryzen 9 9950X3D (SMT off), 32 GB | 2560x1440, DLAA, RX_VK_COMPACT=1, RX_VK_COMPACT_PCT=85, RX_VK_ASTC_BC_ALL=1, RX_VK_IMAGE_BLOCK_MB=32; nothing else running, RX_VK_TEST_ALLOWED_MB=10200 |
the main outdoor area, a scripted five-minute walk, in a folder whose pipeline cache had grown to 2,364 MB | The first start dropped the 2,364 MB file; the prepare made a new one of 681 MB. Working set mean 17,708 / 17,550 MB (peak 18,535 / 18,318) against 18,693 MB (peak 19,349) without the change. No pipeline waits, no draws left out, no device lost |
| 2026-11-12 | v2.1.0-rc.3-252-g221410ba (source 39fd6b45): main ee29dd55’s exe, and the same exe with this change’s manifest line patched in (the segment heap) | Windows 11 (26200), no window (unattended, scripted input) | RX 6700 XT 12 GB, AMD 26.10.1 | Ryzen 5 5600, 16 GB | 2560x1440 output, AMD FSR 4 Performance, the game’s caches (warm), RX_VK_NO_PREDICT=1 RX_VK_PRECOMPILE=0 |
the main outdoor area by day, standing, 150 s; four pairs alternated | fps median NT heap / segment heap: 50.8 / 50.1, 51.2 / 50.7, 49.5 / 51.3, 47.3 / 49.2; GPU 13.0-13.6 ms in both; peak private 21.1-21.4 GB in both. No device lost |
| 2026-11-12 | v2.1.0-rc.3-299-gdae19300 (source c7435484): main dae19300 + this change, its own hookshot.exe | Windows 11 (26200), no window (--prepare-shaders) |
RX 6700 XT 12 GB, AMD 26.10.1 (Vulkan driver 0x800161) | Ryzen 5 5600 (6 cores, 12 threads), 16 GB | Cold: empty Hookshot caches, a new folder path (AMD’s driver cache empty for it) | The shader preparation, no game | The segment heap: vk: preparation: 12 workers for the shader programs; first part (vk: prepared ... in) 982 s; 2,153 predicted pipelines in the first 300 s; peak private memory 12.1 GB |
| 2026-11-12 | as above, the same hookshot.exe with the manifest’s heapType line taken out (the NT heap) |
as above | as above | as above | as above | as above | 6 workers ... (half the CPUs); first part 1,478 s; 1,115 predicted pipelines in the first 300 s; peak private memory 9.0 GB |
| 2026-11-12 | v2.1.0 (source e561602e) | Arch Linux, KDE Plasma Wayland, 2560x1440 window (mailbox, udmabuf on) | RX 9070 XT 16 GB, Mesa RADV 26.2.6 | Core i5-13600KF, 32 GB | 2560x1440 output, V-Sync off, parallel recorder on (RX_VK_RECORDER=thread, RX_VK_BATCH=0), RX_VK_MIRROR_MB=4096, idle live-import cap on (RX_VK_IMPORT_HIGH=25, RX_VK_IMPORT_LOW=20: ~3,858 MB high / ~3,086 MB low on the 15,434 MB system heap budget) |
Long session across the game world (176,640 presents, ~1.5 h) | 63.5-65.6 fps at 641-708 draws, 67.2-68.3 fps at 186 draws throughout the session. 257 idle live imports (7,872 MB in all) given back; live imports stayed at 50-64 pools (2,256-2,720 MB), capture submit 0.18-0.43 ms (scan 0.08 ms over 3,923-4,251 MB), queue submit 0.32-0.73 ms, readbacks of GPU results waited 0.64-1.14 ms each. No device lost |
| 2026-11-12 | v2.1.0 (source e561602e) | as above | as above | as above | 2560x1440 output, V-Sync off, default longshot_env.txt (default 2,048 MB mirror budget, default recorder, imports_give_back off on the 15,434 MB system heap budget since budget >= 12 GB) |
Long session across the game world (684,360 presents, ~2 h) | At 10 min (40,920 presents): 60+ fps, 55 live imports (2,000 MB), capture submit 0.27 ms (scan 0.05 ms over 734 ranges, 2,417 MB), queue submit 0.55 ms, readback wait 2.19 ms each (270 of 3,000). After ~2 h (684,120 presents): faded to 44-48.4 fps at 303-408 draws (vi: real-time step on: ratio 1.352, game's frame rate 44); 221 live imports (7,392 MB; 67 retired given back, 2,992 MB), working set 15,351 MB (peak 16,107 MB, 1,511 MB swap), capture submit 0.50 ms (scan 0.09 ms over 334 ranges, 7,218 MB), queue submit 1.11-1.14 ms, readback wait 6.29-7.03 ms each (312-340 of 3,000). No device lost |
| 2026-11-11 | v2.1.0-rc.3-154-g3c89158e (source 1eba3e78): main 15646d18 + this change | Windows 11, no window (unattended, scripted input) | RTX 4070 SUPER 12 GB, 617.42 | Ryzen 9 9950X3D (SMT off), 32 GB | 2560x1440, DLAA, RX_VK_COMPACT=1, RX_VK_COMPACT_PCT=85, RX_VK_ASTC_BC_ALL=1, RX_VK_IMAGE_BLOCK_MB=32; nothing else running, RX_VK_TEST_ALLOWED_MB=10200, RX_VK_FRAME_TRACE=1 |
From the title into Castle Town, then out to the main outdoor area (an area change) | Two runs (3,480 and 3,600 frames traced). The new columns filled in: summed late_wait_ms 260 / 319 ms, readback_wait_ms 13 / 13 ms, tex_frame_ms 323 / 555 ms; late_wait_ms was never over its frame’s memsync_ms (summed 1,671 / 1,808 ms). No device lost |
| 2026-11-12 | v2.1.0-rc.3-120-g1bde1868 (source 62267c81), the native-zstd branch built from source (llvm-mingw, no DLSS) | Windows 11, 5120x1440 (32:9) | RX 9070 XT, Vulkan driver 0x8001a0 | Ryzen 7 5700X3D, 64 GB | HOS_NATIVE_ZSTD=2 |
the main outdoor area, the standard test: the save loaded, the camera turning at full speed, frame rate unlocked (HOS_FPS_MAX=240), 20 s window; 2 warm-ups, then 4 runs an arm, alternating |
fps 43.7-45.0, CPU 111.5-116.3 ms a frame, p99 28.5-30.0 ms, worst 45.0-56.7 ms, no frame over 100 ms; save load 28.0-28.7 s; 3 core escapes |
| 2026-11-12 | v2.1.0-rc.3-120-g1bde1868 (source 62267c81), the native-zstd branch built from source (llvm-mingw, no DLSS) | Windows 11, 5120x1440 (32:9) | RX 9070 XT, Vulkan driver 0x8001a0 | Ryzen 7 5700X3D, 64 GB | HOS_NATIVE_ZSTD=1 |
the main outdoor area, the standard test: the save loaded, the camera turning at full speed, frame rate unlocked (HOS_FPS_MAX=240), 20 s window; 2 warm-ups, then 4 runs an arm, alternating |
fps 42.3-44.8, CPU 112.8-114.9 ms a frame, p99 28.3-36.4 ms, worst 54.4-333.1 ms, 4 frames over 100 ms and 2 over 250, all in the first run; save load 25.9-28.9 s; 4 core escapes |
| 2026-11-12 | v2.1.0-rc.3-120-g1bde1868 (source 62267c81), the native-zstd branch built from source (llvm-mingw, no DLSS) | Windows 11, 5120x1440 (32:9) | RX 9070 XT, Vulkan driver 0x8001a0 | Ryzen 7 5700X3D, 64 GB | Defaults | the main outdoor area, the standard test: the save loaded, the camera turning at full speed, frame rate unlocked (HOS_FPS_MAX=240), 20 s window; 2 warm-ups, then 4 runs an arm, alternating |
fps 44.0-44.9, CPU 112.3-117.9 ms a frame (process cores / fps), p99 28.7-37.6 ms, worst 55.8-220.6 ms, 2 frames over 100 ms in 4 runs, none over 250; save load 26.2-29.2 s; 4 core escapes. Control for HOS_NATIVE_ZSTD below |
| 2026-11-12 | v2.1.0-rc.3-250-ge375837c (main c544828f + this change + a test patch that makes every mirror refused for the budget MIRROR_REFUSED, so that this card has refused pools at all) against v2.1.0-rc.3-249-gcd9e28dd (main + the same test patch) |
Windows 11, no window (unattended, scripted input) | RTX 4070 SUPER 12 GB, 617.42 | Ryzen 9 9950X3D (SMT off), 32 GB | 2560x1440, DLAA, RX_VK_MIRROR_MB=256 (most pools refused a mirror); nothing else running |
Castle Town to the main outdoor area (an area change), a five-minute the main outdoor area walk, the starting forest | The “vk: imports:” line at the end. Main: 20 retired pools still imported (1,488 MB) after the area change, 13 (992 MB) in the walk, only 2 ever given back. This change: 0 still imported in all three, 21 (1,472 MB), 15 (1,040 MB) and 11 (512 MB) given back. No device lost, no error |
| 2026-11-12 | v2.1.0-rc.3-154-g2e790fa9 (source 8c169744): main 15646d18 + #2684; RX_VK_IMPORT_CHUNK_MB=4 (this change’s default) against unset (16) |
Windows 11, no window (unattended, scripted input) | RTX 4070 SUPER 12 GB, 617.42 | Ryzen 9 9950X3D (SMT off), 32 GB | 2560x1440, DLAA, RX_VK_COMPACT=1, RX_VK_COMPACT_PCT=85, RX_VK_ASTC_BC_ALL=1, RX_VK_IMAGE_BLOCK_MB=32; nothing else running, RX_VK_TEST_ALLOWED_MB=6144, RX_VK_MIRROR_MB=1536 |
the main outdoor area, a scripted five-minute walk | Two pairs: VRAM peak 5,865 / 5,887 MB with 4 MB chunks against 6,135 / 6,129 MB with 16 (allowance 6,144); the GPU’s longest frame 13.5 / 12.5 ms against 24.1 / 21.7 ms; 53.0 / 52.4 fps against 50.3 / 54.9 (no frame-rate claim). No device lost |
| 2026-11-12 | as above | Windows 11, no window (unattended, scripted input) | RTX 4070 SUPER 12 GB, 617.42 | Ryzen 9 9950X3D (SMT off), 32 GB | 2560x1440, DLAA, RX_VK_COMPACT=1, RX_VK_COMPACT_PCT=85, RX_VK_ASTC_BC_ALL=1, RX_VK_IMAGE_BLOCK_MB=32; nothing else running, RX_VK_TEST_ALLOWED_MB=10200 |
the starting forest, standing, the camera turning | 4 MB chunks: 58.2 fps, mirrors 843 MB, VRAM mean 6,672 MB, longest import 44 ms; 16 MB: 57.0 fps, 1,154 MB, 7,281 MB, 66 ms. No hitch from a pool across a chunk’s end. No device lost |
| 2026-11-12 | v2.1.0-rc.3-153-g15646d18-dirty (source f0a7c2bd): main 15646d18 with a temporary measuring patch (the mirrors’ padding counted); RX_VK_IMPORT_CHUNK_MB=4 against unset (16) |
Windows 11, no window (unattended, scripted input) | RTX 4070 SUPER 12 GB, 617.42 | Ryzen 9 9950X3D (SMT off), 32 GB | 2560x1440, DLAA, RX_VK_COMPACT=1, RX_VK_COMPACT_PCT=85, RX_VK_ASTC_BC_ALL=1, RX_VK_IMAGE_BLOCK_MB=32; nothing else running, RX_VK_TEST_ALLOWED_MB=10200 |
the main outdoor area, a scripted five-minute walk; Castle Town to the main outdoor area (an area change) | Walk: mirrors 1,236-1,268 MB with 4 MB chunks against 1,645-1,657 MB with 16, 58.1 / 58.6 fps against 59.0 / 58.2, imports 91-94 (0.8 s in the driver) against 59-65 (0.7 s). Area change: VRAM mean 7,859 against 8,357 MB, 50.5 against 50.4 fps. No device lost |
| 2026-11-11 | v2.1.0-rc.3-156-g50a6887e (source 44083c56): main 15646d18 + #2683 + this change; against v2.1.0-rc.3-154-g3c89158e (source 1eba3e78): main 15646d18 + #2683 | Windows 11, no window (unattended, scripted input) | RTX 4070 SUPER 12 GB, 617.42 | Ryzen 9 9950X3D (SMT off), 32 GB | 2560x1440, DLAA, RX_VK_COMPACT=1, RX_VK_COMPACT_PCT=85, RX_VK_ASTC_BC_ALL=1, RX_VK_IMAGE_BLOCK_MB=32; nothing else running, RX_VK_TEST_ALLOWED_MB=10200 |
From the title into Castle Town (the title-to-game load), then out to the main outdoor area | Two runs each. The frame at the title-to-game load (the movie’s end) took 258 and 259 ms without the change; with it no frame in that stretch was over 60 ms. No device lost |
| 2026-11-12 | v2.1.0-rc.3-260-gb7d175e4f (source 43edcbeb), stock fixed build before rebase, no local counters | Arch Linux, kernel 7.2.15-1-omarchy, headless | Intel UHD 770 (RPL-S), Mesa ANV 26.2.5-arch1.1 | Core i9-13900K, 93 GB reported by the game | 1920x1080, Minimum, FG off, precompile/warm/prediction off, 32 GiB process limit; warmed save/cache; Khronos validation 1.4.363 (RX_VK_VALIDATION=1), duplicate-message cap off |
the starting forest, child, standing; 60 s of confirmed play | VUID-vkCmdDrawIndirect-pStrides-04913: 0 (stock main 51da0464 control: 968 overall, 957 in its 60 s play window). No other errors or new message IDs; existing unused-fragment-output warnings remain. Exit 0, no device loss or GPU fault. Forest/HUD render with this iGPU’s known false colours. No performance claim; full per-ID counts in PR #2796 |
| 2026-11-11 | v2.1.0-rc.3-246-g87991bc8 (source 6d41d308), main built from source (llvm-mingw, no DLSS) | Windows 11, 1280x720 window (mailbox, V-Sync Auto) | RTX 3080 10 GB, 621.14 | Core i9-12900K, 64 GB | Default options, uncapped (HOS_FPS_MAX=1000); a warm-up run first, no pipeline compiled while measuring |
the main outdoor area (adult), by the Lost Woods bridge where the Cheats teleport puts Link, standing; 120 s from 45 s after the teleport | Two runs: 78.4 and 79.3 fps, median frame 12.18 and 12.23 ms, 30 and 37 frames over 20 ms, 1,335-1,453 draws. Readbacks of GPU results: about 1,550 of 3,000 waited, 0.04-0.07 ms each (medians). Against the nightly below for #2602 |
| 2026-11-11 | v2.1.0-rc.3-193-g55b976c0 (source ae9e7c1b), the last nightly before #2602, built from source as above | as above | as above | as above | as above | as above | Two runs, interleaved with the two above: 59.7 and 60.8 fps, median frame 16.39 and 16.32 ms, 264 and 22 frames over 20 ms, 1,355-1,445 draws. Readbacks of GPU results: 172 of 3,000 waited, 5.80 and 5.56 ms each (medians) |
| 2026-11-12 UTC | main 51da0464 / PR2649 84a006d5, matching host and renderer | Windows 11, windowed | RTX 5070 Ti 16 GB, NVIDIA 621.01 | Ryzen 7 9800X3D, 32 GB | 1280x720, 60 fps cap, FG off, precompile/prediction off, RX_VK_PLAY_CACHE=0; identical 1,124,480,456-byte base plus one play file and warmed translator inputs restored before each process | starting-area balcony checkpoint, 30 s after loaded state; eight fresh processes: A/A then AB, BA, AB | All exit 0; checkpoint and final cache hashes match. Paired peak private reductions 924.9 / 1,064.3 / 900.3 MiB, but A/A varied 979.9 MiB. Working-set peak deltas -8.7 / -59.9 / +74.3 MiB; no consistent steady-memory, physical-memory, FPS or startup benefit established |
| 2026-11-12 UTC | PR2649 23c50087 against main 51da0464; reported by dankman42 | Windows 11, windowed | RTX 3080 10 GB, NVIDIA 621.14 | i9-12900K, 64 GB | 1280x720, default options; fixed 1,615,665,731-byte base plus three play files restored each run; one warm-up per build, three interleaved pairs | Adult the village save, teleport to the main outdoor area, 30 s, clean quit; three base saves per measured run | Reported median peak private 26,418 vs 27,831 MB, working set level; all saves published and a further start read the saved cache. Earlier-head external evidence, not a local or current-head rerun; no FPS/Linux/AMD/Intel test |
| 2026-11-12 UTC | v2.1.0-rc.3-234-g7820853d (source 3e8f38f4), matching file-backed-cache host/renderer/image | Windows 11, windowed | RTX 5070 Ti 16 GB, NVIDIA 621.01 | Ryzen 7 9800X3D, 32 GB | 1720x720 requested window, 3440x1440 configured output, game upscaler 67%, 60 fps cap, FG off | starting-area balcony checkpoint, three fresh processes | Loaded checkpoint and rendered scene/HUD; native-menu Quit saved the base cache, 1,124,480,456 to 1,124,543,762 bytes; next process read the saved cache. All exit 0. Functional load/save/read, not a speed or campaign result |
| 2026-11-11 | v2.1.0-rc.3-46-gddafc1f (source c872b22b): the nightly with #2314’s gpu/vk/vk_memory.cpp (97c1a202) swapped in |
Ubuntu 26.04.1, kernel 7.0.0, GNOME 50 Wayland | RTX 4070 12 GB, 615.71.09 | Ryzen 7 5800X3D, 64 GB | 2560x1440 output, DLSS DLAA (model K), frame generation off, Balanced preset (texture budget 6144 MB), Reflex on, RX_VK_PARALLEL=4 |
the starting forest, the stream by the village: fresh launch, then again after about 10 minutes of exploring | Fresh 59.0 fps at about 2,505 draws (memory sync 3.9 ms); after exploring 49.5 fps at about 2,760 draws (memory sync 6.3 ms). 99.96% of the texture-shadow copies came from staging. No device lost |
| 2026-11-11 | as above | as above | as above | as above | as above, RX_VK_SHADOW_STAGE=0 |
as above | Fresh 63.2 fps at about 2,431 draws (memory sync 3.0 ms); after exploring 33.1 fps at 2,431-2,669 draws (memory sync 15.9 ms). No device lost |
| 2026-11-11 | v2.1.0-rc.3-46-gddafc1f (source d13bde75), nightly | as above | as above | as above | as above, without #2314 | the starting forest, the stream by the village, about 2 minutes after launch | 59.2 fps (56.6-60.0 over 29 windows of 120 presents) at about 2,262 draws; replay submit 12.1 ms, memory sync 4.6 ms. No device lost |
| 2026-11-11 | as above | as above | as above | as above | 2560x1440 output, DLSS DLAA (model K), frame generation 3x, texture budget 6144 MB, Reflex on, RX_VK_PARALLEL=4 |
the starting forest, the stream by the village, shortly after launch; then free play | Standing 60.9 fps (59.1-61.4 over 24 windows of 120 presents) at about 2,455 draws. About 8 minutes after launch, in the main outdoor area just north of Lake Hylia by the scarecrows, the device was lost after a 5.4 s GPU memory import: VK_ERROR_DEVICE_LOST, kernel Xid 109 then Xid 31, VRAM 10.4 of 10.8 GB allowed. Log on #1000 |
| 2026-11-11 | host v2.1.0-main-local (source fc2ed0d0); renderer main 49f75dc8 + this branch (FSR 3.1 on the temporal chain) | Gentoo Linux, kernel 7.3.3, KDE Plasma Wayland, windowed | RX 9070 XT 16 GB, Mesa RADV 26.2.6 | Ryzen 9 9950X3D, 32 GB | 2560x1440 output, AMD FSR 3.1 at 100% and Quality (67%), 60 fps cap; RX_VK_FSR3_CHAIN=0 for the comparison runs |
the starting forest (the lost-woods path, a 28 s camera turn), Castle Town market at noon | FSR 3.1 drew every frame on the chain (600 of 600 per status line), no evaluation failures. starting-area, 4 runs each: 59.3 fps on the chain, 59.2 with its own inputs (1% low 48.1 and 46.4). Market, 3 runs each: 52.5 and 53.1 fps (runs vary by about 0.9). Standing still at 67%: edge flicker 1.67 against 3.01, 1-pixel outlines 2 against 12 per megapixel; at 100% the same within noise |
| 2026-11-11 | v2.0.14 (source f2796384) + the value_rec last-part fix (renderer only) |
Anatase 20261005.5 (Fedora-based), Steam Gaming Mode (gamescope, Flatpak Steam), 1280x720 window | ROG Ally X: Ryzen Z1 Extreme iGPU, RADV PHOENIX, Mesa 26.2.3 | Ryzen Z1 Extreme, 24 GB (15.6 GB visible) | 1920x1080 output, game’s upscaler 67%, FG 2x, pacing on; HOS_GUEST_MEMFD=1, RX_VK_VRAM_MB=3072 |
Death Mountain Trail, walking and a fight | 57 s recording: 0 one-frame luminance spikes; 31.3 fps median (the game’s own frames, over 300 draws) |
| 2026-11-11 | host v2.1.0-main-local (source fc2ed0d0); renderer main 49f75dc8 + #1941 + #1942 + #2243 + this change, built with FSR 4 | Gentoo Linux, kernel 7.3.3, KDE Plasma Wayland, windowed | RX 9070 XT 16 GB, Mesa RADV 26.2.6 | Ryzen 9 9950X3D, 32 GB | 2560x1440 HDR10 (HDR Native, paper white 110 nits, peak 400), AMD FSR 4 Native AA through fsr4-vk (FSR 4.1.1, FP8) | the starting forest (the lost-woods path) | The game presented PQ and FSR 4 was made HDR (log: HDR colour, linear); 600 of 600 frames drawn, no device lost. Brightest pixel PQ code 132 (the paper white) with RX_VK_FSR4_HDR=0, 135-145 with the change (two runs); the rest of the scene within one code value, brightness steady frame to frame. SDR unchanged |
| 2026-11-11 | v2.1.0-rc.3-41-g59b9bb75 (main fab8cb3 + the native float add, unmerged then) | Arch Linux, no window (unattended, scripted input) | RTX 4090, 615.78.08 | Core i9-13900K, 96 GB | 2048x1152 output, Resolution 100%, Game speed per frame, RX_VK_PASS_TIMING=2; float adds native (OpAtomicFAddEXT) |
the village (child, day), standing still | GPU time 7.2 ms a frame (median of 56 120-present windows); 59.3 fps mean (60 cap). No device lost |
| 2026-11-11 | as above | as above | as above | as above | as above, RX_VK_NATIVE_FADD=0 (the compare-and-swap loop, as before) |
as above | GPU time 7.8 ms a frame (median of 57); 59.3 fps mean. No device lost |
| 2026-11-11 | as above | as above | as above | as above | as above, float adds native | Death Mountain Trail (adult, night), standing still | GPU time 5.4 ms a frame (median of 57); 59.4 fps mean. No device lost |
| 2026-11-11 | as above | as above | as above | as above | as above, RX_VK_NATIVE_FADD=0 |
as above | GPU time 5.8 ms a frame (median of 57); 59.6 fps mean. No device lost |
| 2026-11-11 | v2.1.0 (source 633eb435) with the 3D BC attachment fix | Windows 10 Enterprise 19045, offscreen | GTX 1080, 581.80 | Core i9-7900X, 64 GB | 1920x1080 offscreen, compute mesh fallback, RX_VK_SKIP_PROG unset |
New game, opening cutscene into the first dungeon and the companion dialogue | Recorded all 128 noise draws. The dialogue rendered and the run reached 15,840 presents without device loss. A standalone Vulkan probe lost the device on a direct BC4 volume attachment clear, then read back all 16,384 blocks correctly through the separate attachment and copy path |
| 2026-11-11 | PR #2357, author report; exact build not recorded | Not recorded | Not recorded | Not recorded | FOV 50% to 150%; also with ultrawide enabled | Gameplay aiming; area not recorded | The PR author reports aiming works across the tested FOV range and with ultrawide mode. Manual gameplay confirmation; no performance measurements reported |
| 2026-11-11 | v2.1.0-rc.3-53-g49f75dc-dirty (source 52a1c006), local feat/adjustable-fov |
Fedora 44, headless | RX 7800 XT 16 GB, RADV Mesa 26.1.8 | Ryzen 7 5700X, 32 GB | 1920x1080, game’s upscaler, FG off, RX_VK_PRECOMPILE=0; copied saves and separate settings |
Dodongo’s Cavern 2F, stationary by the chest | Scripted Gameplay menu changes 100% to 125% and back to 100%. Inspected captures: wider scene at 125%, original framing restored at 100%, HUD placement unchanged; new slider and help fit at 1280x720. Options file saved 125 then 100. No performance comparison; Windows, ultrawide, aiming and cutscene transitions not exercised in that run |
| 2026-11-10 | v2.0.13 (source cfc565a2) + the value_rec last-part fix (renderer only) |
Anatase 20261005.5 (Fedora-based), KDE Plasma desktop, windowed | ROG Ally X: Ryzen Z1 Extreme iGPU, RADV PHOENIX, Mesa 26.2.3 | Ryzen Z1 Extreme, 24 GB (15.6 GB visible) | 1920x1080 output, FSR 3.1 Quality, FG 2x, pacing on, cap 120 | the main outdoor area, walking and fights | 42 s recording: 3 one-frame luminance spikes, 0 of 15/255 or more; 103-137 sync waits in 5 s (1.4-1.8 s); FPS overlay 38-49 |
| 2026-11-10 | v2.0.13 (source cfc565a2) | as above | as above | as above | as above | the main outdoor area, Castle Town | 69 s recording: 104 one-frame luminance spikes, 15 of 15/255 or more (HUD stretched across the screen, grass flat white, whole-screen flashes); 4-109 sync waits in 5 s (31-232 ms); FPS overlay 45-70 |
| 2026-11-11 | v2.1.0-rc.3-58-g06ce5785 (main 49f75dc + the compaction hold’s fix and reserve, and the compaction on by default; unmerged then) | Windows 11, no window (unattended, scripted input) | RTX 4070 SUPER 12 GB, 617.42 | Ryzen 9 9950X3D (SMT off), 32 GB | 2560x1440, DLAA, no switch set (the compaction on by default) against RX_VK_COMPACT=0; nothing else running |
the main outdoor area, a scripted five-minute walk (not standing still) | Two runs each, counted from the end of the load. On: 59.8, 59.9 fps; 66/12/2 and 36/2/1 frames over 25/33/50 ms; 1% low 37.7, 40.9 fps; the driver’s count 9.05-9.07 GB in the mean (85% of what it allows), 88-91% at most; the write-back shadows never given back. Off: 56.0, 56.0 fps; 633/60/15 and 408/72/28; 1% low 25.8, 24.1 fps; the count 10.21-10.29 GB in the mean (95-96%), 97-98% at most; the shadows given back 8 times in each. No validation, device-lost or “not made” line |
| 2026-11-11 | as above | as above | as above | as above | 2560x1440, DLAA, RX_VK_TEST_ALLOWED_MB=7000, no other switch against RX_VK_COMPACT=0 |
the main outdoor area, standing, the camera turning for 50 s (not still) | One run each, counted from the end of the load. On: 60.0 fps, 1,203 MB of textures uploaded, 4,498 retired, 1 frame over 25 ms, 1% low 45.4 fps. Off: 59.9 fps, 2,660 MB, 8,332 retired, 8 frames over 25 ms, 1% low 40.7 |
| 2026-11-11 | v2.1.0-rc.3-50-g3acf477a (main 4d68bc8 + the same two commits, rebased; unmerged then) against v2.1.0-rc.3-48-g4d68bc83 (main) | Windows 11, no window (unattended, scripted input) | RTX 4070 SUPER 12 GB, 617.42 | Ryzen 9 9950X3D (SMT off), 32 GB | 2560x1440, DLAA, RX_VK_COMPACT=1, RX_VK_TEST_ALLOWED_MB=7000; nothing else running |
the main outdoor area, standing, the camera turning for 50 s (not still) | One run each, counted from the end of the load. The branch: 60.0 fps, 1,187 MB of textures uploaded, 3,445 retired, no frame over 25 ms, 1% low 47.2 fps. Main: 59.6 fps, 6,842 MB, 13,290 retired, 16 frames over 25 ms, 1% low 39.1. While the save loaded, the branch’s count passed the allowance once (7,016 of 7,000 MB) and the write-back shadows were given back once; main’s did not. No validation, device-lost or “not made” line |
| 2026-11-11 | as above | as above | as above | as above | 2560x1440, DLAA, RX_VK_COMPACT=1, RX_VK_ASTC_BC_ALL=1, RX_VK_TEST_ALLOWED_MB=9500 |
the main outdoor area, a scripted five-minute walk (not standing still) | One run each, counted from the end of the load. The branch: 29/8/1 frames over 25/33/50 ms, the write-back shadows never given back, 4,753 MB uploaded, 1% low 42.3 fps. Main: 104/26/6, the shadows given back 5 times, 7,383 MB, 1% low 33.6 |
| 2026-11-11 | v2.0.11-782-gf0c6e73 (main 3932421 + the compaction hold’s fix and reserve, unmerged then) | Windows 11, no window (unattended, scripted input) | RTX 4070 SUPER 12 GB, 617.42 | Ryzen 9 9950X3D (SMT off), 32 GB | 2560x1440, DLAA, RX_VK_COMPACT=1, RX_VK_TEST_ALLOWED_MB=7000; nothing else running |
the main outdoor area, standing, the camera turning for 50 s (not still) | Two runs: 60.0, 60.0 fps; 1,284 and 1,284 MB of textures uploaded; 3,814-4,044 textures retired; 7 and 5 frames over 25 ms; 1% low 42.4, 44.3 fps. No validation, device-lost or “not made” line |
| 2026-11-11 | v2.0.11-780-g3932421 (main) | as above | as above | as above | as above | as above | Two runs: 59.0, 59.2 fps; 6,573 and 5,954 MB uploaded; 12,622-13,701 textures retired; 42 and 27 frames over 25 ms; 1% low 34.8, 35.3 fps. Without RX_VK_COMPACT: 60.0 fps, 1,313 MB, 3,481 retired, 6 frames over 25 ms, 1% low 42.5 |
| 2026-11-11 | v2.0.11-782-gf0c6e73 (as above) | as above | as above | as above | 2560x1440, DLAA, RX_VK_COMPACT=1, RX_VK_ASTC_BC_ALL=1, RX_VK_TEST_ALLOWED_MB=9500 |
the main outdoor area, a scripted five-minute walk (not standing still) | Two runs: 86/17/1 and 92/13/1 frames over 25/33/50 ms; the write-back shadows never given back; GPU time over 12.5 ms in 0 of 126 and 2 of 125 windows of 120 presents; 1% low 35.9, 36.9 fps. Without the cap (11.2 GB allowed): 96/19/0 and 65/8/1, shadows never given back, 1% low 35.9, 38.2 |
| 2026-11-11 | v2.0.11-780-g3932421 (main) | as above | as above | as above | as above | as above | Two runs: 120/29/10 and 93/16/3 frames over 25/33/50 ms; the shadows given back 4 and 5 times; GPU time over 12.5 ms in 8 of 124 and 11 of 124 windows; 1% low 30.6, 35.5 fps. Without the cap: 93/24/7 and 97/28/11, shadows given back twice in each, 1% low 33.2, 31.2 |
| 2026-11-11 | host v2.1.0-main-local (source fc2ed0d0); renderer main 49f75dc8 + #1941 + #1942 + #2243 + this change, built with FSR 4 | Gentoo Linux, kernel 7.3.3, KDE Plasma Wayland, windowed | RX 9070 XT 16 GB, Mesa RADV 26.2.6 | Ryzen 9 9950X3D, 32 GB | 2560x1440 HDR10 (HDR Native, paper white 110 nits, peak 400), AMD FSR 4 Native AA through fsr4-vk (FSR 4.1.1, FP8) | the starting forest (the lost-woods path) | The game presented PQ and FSR 4 was made HDR (log: HDR colour, linear); 600 of 600 frames drawn, no device lost. Brightest pixel PQ code 132 (the paper white) with RX_VK_FSR4_HDR=0, 135-145 with the change (two runs); the rest of the scene within one code value, brightness steady frame to frame. SDR unchanged |
| 2026-11-11 | v2.1.0 (source 633eb435) with the 3D BC attachment fix | Windows 10 Enterprise 19045, offscreen | GTX 1080, 581.80 | Core i9-7900X, 64 GB | 1920x1080 offscreen, compute mesh fallback, RX_VK_SKIP_PROG unset |
New game, opening cutscene into the first dungeon and the companion dialogue | Recorded all 128 noise draws. The dialogue rendered and the run reached 15,840 presents without device loss. A standalone Vulkan probe lost the device on a direct BC4 volume attachment clear, then read back all 16,384 blocks correctly through the separate attachment and copy path |
| 2026-11-11 | main d7a73062 + input-origin/game-clock log witnesses; numeric-input prediction candidate renderer (matched API14) |
Windows 11 Home N, 26300 | RX 9070 XT 16 GB, 32.0.32015.12012 | Ryzen 7 5700X3D, 64 GB | 2560x720, 32:9 camera, serial, FG off, wall-clock scripted save load, periodic captures | the main outdoor area, load only, 95 seconds | 4,219 consecutive native trace rows with all seven wall-clock phase fields ordered. All 14 logged game-clock timestamps exactly match native trace endpoints; the input origin is logged. Original frame 2400 shows terrain and HUD. No CPU-time, movement-speedup or clean-validation claim; validation off. First game-clock baseline endpoint is not logged |
| 2026-11-11 | Host and clean renderer fab8cb37; candidate GPU write-back observations on that base |
Windows 11, windowed | RTX 5070 Ti, driver 0x9b404000 |
Ryzen 7 9800X3D, 32 GB | 3840x2160 output, main scene 2880x1620, 1280x720 window, frame generation off, cap 144, exclusive scripted input | starting-area balcony, fresh copy of preserved checkpoint per process | Six 30 s windows in clean/off/on/on/off/clean order: mean FPS 75.249 / 75.523 / 74.242 by mode. Observations on: 95/97 reports, 0 unavailable chunks; changed chunks 15.11/15.03%, covering 13.06/13.00% of compared volume. All six exit normally, original checkpoint unchanged. Two windows per mode; no temperature/clock telemetry, equivalence or culling claim. |
| 2026-11-10 | main 117eae7 (git describe: v2.0.11-730-g117eae7), built with make_game; log: longshot unknown (source 4ca6c9ae) |
NixOS 26.11, Linux 7.2.7; headless Sway 1.12 + Xwayland (unattended, scripted input) | RX 7900 XT (RADV NAVI31), Mesa 26.2.3 | Ryzen 5 7600X, 32 GB | 2048x1152 game output, 1920x1080 window, Auto render resolution, game’s own upscaler, Lanczos window filter, FG 2x, udmabuf on, audio off; mesh mode: RX_VK_FORCE_MESH_FALLBACK=1 only |
the starting forest, the player character’s house balcony (from a save); standing still for the sample, then scripted movement | 139 s total; 46.6 fps median of the last five stationary samples. No GPU hang; stopped by the test |
| 2026-11-10 | as above | as above | as above | as above | as above; mesh mode: RX_VK_TASK_MESH_FALLBACK=1 only |
as above | 139 s total; 43.8 fps median of the last five stationary samples. No GPU hang; stopped by the test |
| 2026-11-10 | as above | as above | as above | as above | as above; mesh mode: native mesh (both fallback switches unset) | as above | 137 s total; 45.4 fps median of the last five stationary samples. No GPU hang; stopped by the test |
| 2026-11-10 | as above | as above | as above | as above | as above; mesh mode: RX_VK_FORCE_MESH_FALLBACK=1 only |
as above | 139 s total; 46.5 fps median of the last five stationary samples. No GPU hang; stopped by the test |
| 2026-11-10 | as above | as above | as above | as above | as above; mesh mode: RX_VK_TASK_MESH_FALLBACK=1 only |
as above | 139 s total; 44.0 fps median of the last five stationary samples. No GPU hang; stopped by the test |
| 2026-11-10 | as above | as above | as above | as above | as above; mesh mode: RX_VK_FORCE_MESH_FALLBACK=1 only |
as above | 139 s total; 46.5 fps median of the last five stationary samples. No GPU hang; stopped by the test |
| 2026-11-10 | as above | as above | as above | as above | as above; mesh mode: RX_VK_TASK_MESH_FALLBACK=1 only |
as above | 138 s total; 43.9 fps median of the last five stationary samples. No GPU hang; stopped by the test |
| 2026-11-10 | as above | as above | as above | as above | as above; mesh mode: native mesh (both fallback switches unset) | as above | 198 s total; 44.9 fps median of the last five stationary samples. No GPU hang; stopped by the test |
| 2026-11-10 | as above | as above | as above | as above | as above; mesh mode: RX_VK_FORCE_MESH_FALLBACK=1 only |
as above | 231 s total; 46.1 fps median of the last five stationary samples. No GPU hang; stopped by the test |
| 2026-11-10 | as above | as above | as above | as above | as above; mesh mode: RX_VK_TASK_MESH_FALLBACK=1 only |
as above | 179 s total; 43.9 fps median of the last five stationary samples. No GPU hang; stopped by the test |
| 2026-11-10 | v2.0.14 (source 3a9998c3), Longshot Builder; later v2.0.14 + the scale-image SAMPLED fix (as in #1940) | Windows 11 Pro 26200, fullscreen | RTX 5090, 621.14 | Ryzen 7 5800X3D, 64 GB | 3840x2160 output, DLSS Quality (67%), Steam Controller (2026) | Castle Town market, the starting forest | Plays well. On the release: HDR Expanded a black picture, Native flat (#1613); with the fix Expanded is right, frame generation 3x too. One crash in nvoglv64.dll (0xc0000005) after about 17,000 frames with frame generation 5x. Steam Controller started from Steam: no gyro until Steam’s SDL_GAMECONTROLLER_IGNORE_DEVICES and SDL_GAMECONTROLLER_ALLOW_STEAM_VIRTUAL_GAMEPAD were cleared (#1897) |
| 2026-11-10 | cc80fde on 9447e1d, Vulkan API 14 |
Windows 11 Pro 26300, windowed | RTX 5070 Ti, 621.01 | Ryzen 7 9800X3D, 32 GB | 3440x1440 game output, fixed FG 2x, parallel replay (2 workers, batch 1), pacing off | the player character’s house in starting-area, copied checkpoint | 6,688 presents logged. Swapchain 1444x720 → 1440x829 → 1444x720, then 1444x720 after minimize/restore. The scene returned and the window closed normally. Surface loss and deferred admission were exercised with injected results in the source fixture. |
| 2026-11-10 | v2.0.11-449-g1987d8c (main caaee0b + the VRAM compaction branch, unmerged then) | Windows 11, no window (unattended, scripted input) | RTX 4070 SUPER 12 GB, 617.42 | Ryzen 9 9950X3D (SMT off), 32 GB | 2560x1440, DLAA, RX_VK_COMPACT=1; other programs open (10.2-10.5 GB of VRAM allowed) |
the main outdoor area, a scripted five-minute walk (not standing still) | Two runs: 58.2, 58.6 fps mean; VRAM 8.3, 7.9 GB; GPU time over 12.5 ms a frame in 6 of 122 and 10 of 123 120-present windows; 88, 62 frames over 33 ms; 1% low 21.1, 26.5 fps. No device lost |
| 2026-11-10 | as above | as above | as above | as above | as above, without the switch | as above | Two runs: 53.2, 54.4 fps mean; VRAM 9.5, 9.6 GB; the write-back shadows given back in 73 of 111 and 80 of 114 windows; GPU time over 12.5 ms in 69 of 111 and 55 of 114; 107, 112 frames over 33 ms; 1% low 19.4, 20.6 fps. No device lost |
| 2026-11-10 | host v2.0.11-438-gcd544a7 (source 8bdd7a54); renderer main 8bac8aa + #1502 + the host-clock capture change |
Windows 11 Home N, 26300 | RX 9070 XT 16 GB, 32.0.32015.12012 | Ryzen 7 5700X3D, 64 GB | 2560x720 window, 32:9 camera, serial, FG off, dense capture, core validation; fresh prepared shader bank | Boot, title, save load; gameplay not reached | 96 s: 1,476 consecutive original capture records; all 1,461 records covered by the flushed native trace align one-to-one with native returns. Movement window rejected: no filled gameplay interval. No new validation IDs; existing 02251, and two 08737 errors during full preparation. No gameplay speed or performance claim |
| 2026-11-10 | v2.0.11-447-gcbe49eb, sampler admission branch (main 8bac8aa + sampler checks + staging fixture repair) | Windows 11, windowed | RTX 5070 Ti, 621.01 | Ryzen 7 9800X3D, 32 GB | 3440x1440 picture, native upscaling, fixed FG 2x, 1 MB recording streams; serial and parallel replay (2 workers, batch 1) | the starting forest, inside the player character’s house; copied checkpoint | Both runs rendered the scene, maximized/restored (client 1444x720 to 1440x829 and back), and exited normally. Six representative PNGs decoded at 3440x1440; original checkpoint hashes unchanged. Healthy-path validation only; no FPS comparison or native OOM recovery claim |
| 2026-11-10 | host 0f5d423a (API13 diagnostic backport), renderer main dadecd53 |
Windows 11 | RX 9070 XT 16 GB, 32.0.32015.12012 | Ryzen 7 5700X3D, 64 GB | 32:9, serial, FG off, GPU profiling/validation off; menu-only CPU program bank plus fixed warmed application pipeline caches; sparse screenshots | the main outdoor area, first entry, control first | 96 s; reached gameplay, terrain checked, no fatal error, original saves unchanged. 557 readiness waits totaling 30.725 s; longest native-present interval 32.568 s. Empty 40-60 s window rejected. Driver/OS caches uncontrolled; no FPS claim |
| 2026-11-10 | same host, renderer d39e988c (#1573) |
as above | as above | as above | identical nine-file starting cache hashes and settings | the main outdoor area, first entry, candidate second | 96 s; terrain/grass/castle checked, no fatal error, original saves unchanged. 815 readiness waits totaling 3.120 s; longest native-present interval 2.114 s. CPU decision tests do not prove GPU writes; no speedup claim |
| 2026-11-10 | same host, renderer main dadecd53 |
as above | as above | as above | identical nine-file starting cache hashes and settings | the main outdoor area, first entry, control repeated third | 96 s; terrain checked, no fatal error, original saves unchanged. 557 readiness waits totaling 0.923 s; longest native-present interval 2.128 s. Unchanged control also lost its long freeze: do not attribute the first/second difference to #1573 |
| 2026-11-10 | main b2bb214 (+ #1447 #1451 #1520 #1660 #1619 #1448 #1654 #1612 #1600, unmerged then) | Ubuntu 26.04.1, KDE Plasma 6.6.6 Wayland | RTX 4070 Ti 12 GB, 595.99.02 | Ryzen 7 9800X3D, 32 GB | 3840x2160 output, DLSS Quality, no mods | the castle road (child) | 48-50 fps. Same spot at 2560x1440 output: 56.3 fps. No device lost, saves written |
| 2026-11-10 | main b2bb214 + #1459 (Dear ImGui menu) | as above | as above | as above | as above, windowed | Title, the castle road | F1 menu, help box, prompts, the game’s font and the FPS badge all correct; 60 fps |
| 2026-11-10 | main b2bb214 | as above | as above | as above | HDR on | start-up | KDE offers an HDR10 surface (vk: HDR: an HDR10 surface (format 58)) |
| 2026-11-09 | main 7eac5bf | as above | as above | as above | 3840x2160, DLSS Quality, no mods | the castle road | 52.8 fps (replay submit 17.5 ms, memory sync 4.8 ms, GPU 12.8 ms) |
| 2026-11-09 | main 7eac5bf | as above | as above | as above | fullscreen (HOS_FULLSCREEN=1) |
the castle road | fullscreen works, no device lost |
| 2026-11-09 | v2.0.11-142 (7dbdcdc) | as above | as above | as above | 3840x2160, DLSS Quality | Water Temple 1F, the castle road, the first dungeon 2F | Water Temple walls missing with the 23.3 GB dump, drawn with stand-in textures (#1308) or a complete dump. Slingshot and bow aiming: no fps change (51-52 fps, #1305). Pause / unpause: no freeze |
Limits of the original RTX 4070 Ti runs
- Real mouse aiming with Mouse camera = Native after https://github.com/NotRexed/Hookshot/pull/1462 (the camera states were traced, but not tested with a real mouse).
- Frame generation at 120 Hz or above (the displays here top out at 75 Hz).
- HDR output on an HDR monitor (the surface is offered, the picture was not judged on one).