[Crash] VK_ERROR_DEVICE_LOST on RX 6900 XT, always right after "pool 0x7fde000000..0x7fe0000000 took in 2 pools"

rbx-br commented

2026-11-07 13:28:55 +00:00

(Migrated from github.com)

Setup: hookshot.exe, AMD Radeon RX 6900 XT, driver 32.0.21045.20006 (Adrenalin 26.10.2), Intel Xeon E5-2697 v3, 64 GB RAM, Windows 11. The game runs at a steady 60 fps, then crashes.

What happens:
The AMD driver crashes (TDR) and the game freezes. The log shows submit N failed (-4) followed by device lost. After that, every pipeline fails with -4 and I close the window manually.

Pattern I found across 3 runs (logs attached):
In all three logs, the lines right before the failure are a series of pool ... remade larger / took in 2 pools messages, and the last memory line is always the same one:
vk: pool 0x7fde000000..0x7fe0000000 took in 2 pools
The very next thing in the log is submit failed (-4).

Run 1 (Stock): crashed at submit
Run 2 (With RX_VK_GPU_UNSWIZZLE=0): crashed at submit
Run 3 (With RX_VK_GPU_UNSWIZZLE=0 and RX_VK_CHECKPOINTS=1): crashed at submit

Things I tried (no change):

RX_VK_GPU_UNSWIZZLE=0 (confirmed applied: log shows "unswizzled on the CPU")
RX_VK_CHECKPOINTS=1 (no marker output appeared in the log, only the same "device lost" hint)

Happy to test any build or flag you suggest.

built-in log
hookshot_log.txt

my log
log-partida.txt

**Setup:** hookshot.exe, AMD Radeon RX 6900 XT, driver 32.0.21045.20006 (Adrenalin 26.10.2), Intel Xeon E5-2697 v3, 64 GB RAM, Windows 11. The game runs at a steady 60 fps, then crashes. **What happens:** The AMD driver crashes (TDR) and the game freezes. The log shows `submit N failed (-4)` followed by `device lost`. After that, every pipeline fails with -4 and I close the window manually. **Pattern I found across 3 runs (logs attached):** In all three logs, the lines right before the failure are a series of `pool ... remade larger` / `took in 2 pools` messages, and the last memory line is always the same one: `vk: pool 0x7fde000000..0x7fe0000000 took in 2 pools` The very next thing in the log is `submit failed (-4)`. - Run 1 (Stock): crashed at submit - Run 2 (With RX_VK_GPU_UNSWIZZLE=0): crashed at submit - Run 3 (With RX_VK_GPU_UNSWIZZLE=0 and RX_VK_CHECKPOINTS=1): crashed at submit **Things I tried (no change):** - RX_VK_GPU_UNSWIZZLE=0 (confirmed applied: log shows "unswizzled on the CPU") - RX_VK_CHECKPOINTS=1 (no marker output appeared in the log, only the same "device lost" hint) Happy to test any build or flag you suggest. built-in log [hookshot_log.txt](https://github.com/user-attachments/files/34519678/hookshot_log.txt) my log [log-partida.txt](https://github.com/user-attachments/files/34519948/log-partida.txt)

NotRexed commented

2026-11-07 19:41:27 +00:00

(Migrated from github.com)

Linked here from closed duplicates, with log data: #233 (RTX 4060 Ti, checkpoint dispatch 4b03b3040d7cf076, frame 4053), #266 (GTX 1660 SUPER, unbound storage buffers in 53caac5f38279735 / 7984161db3e053b5, RX_VK_TAAU_KEYS=- reported to avoid it), #108 and #273 (RX 6600: device lost as new computes come in, "pool ... not imported" retries, mirrors given back), #243 (RX 6600, Kakariko cemetery), plus reports without logs: #29, #170, #178, #181, #202 (Legion Go S / ROG Ally Z1E), #223, #253, #258, #264.

👍 1

Tempus3214 commented

2026-11-07 20:27:30 +00:00

(Migrated from github.com)

hookshot_log.txt

still getting hang right after the first cutscene, heres my log, rx 6600.

[hookshot_log.txt](https://github.com/user-attachments/files/34528175/hookshot_log.txt) still getting hang right after the first cutscene, heres my log, rx 6600.

sforsbergg commented

2026-11-07 20:46:27 +00:00

(Migrated from github.com)

Managed to show a couple frames at varu bad framerate then the screen turned black again

hookshot_log.txt

Managed to show a couple frames at varu bad framerate then the screen turned black again [hookshot_log.txt](https://github.com/user-attachments/files/34528460/hookshot_log.txt)

Sebgoataura commented

2026-11-07 21:49:49 +00:00

(Migrated from github.com)

hi, now I got through the big deku tree but then it just crashed

hookshot_log.txt

hi, now I got through the big deku tree but then it just crashed [hookshot_log.txt](https://github.com/user-attachments/files/34529388/hookshot_log.txt)

ghostday05 commented

2026-11-08 00:49:13 +00:00

(Migrated from github.com)

hi crashed after entering the mojo tree my logs :

hookshot_log.txt

hi crashed after entering the mojo tree my logs : [hookshot_log.txt](https://github.com/user-attachments/files/34532712/hookshot_log.txt)

NotRexed commented

2026-11-08 02:34:50 +00:00

(Migrated from github.com)

Found and fixed (9bb52a9), in the release https://github.com/NotRexed/Hookshot/releases/tag/device-lost-fix. The renderer keeps a table of where the game's memory lives on the GPU, and the CPU rewrote it while earlier GPU work was still running: right after the game's memory pools merged, the foliage culling shader (53caac5f) could follow an entry into memory that wasn't ready, and the GPU faulted (it stopped for ~2 s, then 'device lost'). Table changes now reach the GPU in submission order. Here that took starts from ~1 crash in 7 to 0 in 14, with the same fps. Please rebuild with this release and tell me if you still get a device lost; if so, attach hookshot_log.txt (the lines after 'device lost: reading' say where).

Found and fixed (9bb52a9), in the release https://github.com/NotRexed/Hookshot/releases/tag/device-lost-fix. The renderer keeps a table of where the game's memory lives on the GPU, and the CPU rewrote it while earlier GPU work was still running: right after the game's memory pools merged, the foliage culling shader (53caac5f) could follow an entry into memory that wasn't ready, and the GPU faulted (it stopped for ~2 s, then 'device lost'). Table changes now reach the GPU in submission order. Here that took starts from ~1 crash in 7 to 0 in 14, with the same fps. Please rebuild with this release and tell me if you still get a device lost; if so, attach hookshot_log.txt (the lines after 'device lost: reading' say where).

pr0vId3nce commented

2026-11-08 05:41:51 +00:00

(Migrated from github.com)

Tested main at 4cfa08f (includes 9bb52a9) on an RTX 4080 (Windows 11, Ryzen 9 7900X): 6 New Game runs, 1 with no caches and 5 with warm caches, each through the opening, the Deku Tree cutscene and into Kokiri Forest. No device lost, no failed submits, and no 53caac5f lines in any log.

For comparison, 3 earlier runs of e40306e (before 9bb52a9) on this PC also had no device lost, so this card may not hit the bug often. Treat this as "no regression and no crash in 6 runs", not proof of the fix. I can run more, or the old ordering (RX_VK_TABLE_NOW=1) as a control, if that helps.

Tested `main` at 4cfa08f (includes 9bb52a9) on an RTX 4080 (Windows 11, Ryzen 9 7900X): 6 New Game runs, 1 with no caches and 5 with warm caches, each through the opening, the Deku Tree cutscene and into Kokiri Forest. No device lost, no failed submits, and no `53caac5f` lines in any log. For comparison, 3 earlier runs of e40306e (before 9bb52a9) on this PC also had no device lost, so this card may not hit the bug often. Treat this as "no regression and no crash in 6 runs", not proof of the fix. I can run more, or the old ordering (`RX_VK_TABLE_NOW=1`) as a control, if that helps.

nicholasgower commented

2026-11-08 09:27:42 +00:00

(Migrated from github.com)

Current main has 9bb52a9, the guest page-table update ordering in the foliage culling. The last retest here was six clean runs on an RTX 4080, so this remains the canonical device-lost tracker rather than a new bug.

If you still see device lost on a build with 9bb52a9, attach hookshot_log.txt; the lines after device lost: reading say where it happened. Otherwise no action is needed on this issue.

Current `main` has `9bb52a9`, the guest page-table update ordering in the foliage culling. The last retest here was six clean runs on an RTX 4080, so this remains the canonical device-lost tracker rather than a new bug. If you still see `device lost` on a build with `9bb52a9`, attach `hookshot_log.txt`; the lines after `device lost: reading` say where it happened. Otherwise no action is needed on this issue.

wimw85 commented

2026-11-08 10:15:45 +00:00

(Migrated from github.com)

Still device lost on a build with 9bb52a9: Linux, Steam Deck (RADV VANGOGH), reproducible

You asked for the lines after device lost: reading from builds that already include 9bb52a9. Two runs on a Steam Deck, both ending in a GPU hang at the same point.

Setup

Steam Deck, AMD Custom GPU 0405 (RADV VANGOGH), Vulkan 1.4.350, SteamOS build 20261023.100, kernel 7.2.7-valve2.
15.2 GB RAM visible (BIOS UMA frame buffer set to 1 GB; with 4 GB the machine ran out of memory instead).
Hookshot Linux AppImage of main at 3601726 (source 50b3813b), built on the Deck itself, game v131072. Program cache warm (39,918 keys).
hookshot_options.ini: quality preset Custom (no VRAM budget), resolution Auto, frame generation off, game speed Real time, handheld profile on.

What happens
Intro and title run at 53 to 60 fps. When the save loads (Kokiri Forest), textures go from about 130 MB to about 3.5 GB within seconds (run 1: 1500 textures, 3496 MB), then the GPU hangs. Kernel:

amdgpu: ring gfx_0.0.0 timeout, signaled seq=66295, emitted seq=66296
amdgpu: Ring gfx_0.0.0 reset failed
amdgpu: GPU reset begin!   (MODE2 reset, "GPU reset succeeded")

Hookshot (run 2):

vk: table entry UNBOUND at frame 1435: prog ae384cff0ad0f467 ssbo 0 of 3 (c0+0x880) addr 0 size 0 (before 0x107f7e8000)
vk: table entry UNBOUND at frame 1435: prog 918adef1afaa39d4 ssbo 0 of 3 (c0+0x880) addr 0 size 0 (before 0x107f7e8000)
vk: table entry UNBOUND at frame 1435: prog d13aac3cec1a4bb5 ssbo 0 of 3 (c0+0x880) addr 0 size 0 (before 0x107f7e8000)
vk: texture 0x10a7bb6c00 6144x6144x1 x1 (NVN 0x7b, 191 MB): its 1 top level(s) left out (47 MB)
radv/amdgpu: The CS has been cancelled because the context is lost. This context is guilty of a hard recovery.
vk: submit 10046 failed (-4)
vk: device lost: reading the fault info (RX_VK_CHECKPOINTS=1 for markers)

Run 1 ended the same way at frame 3324, after the same table entry (c0+0x880, addr 0x107f7e8000, size 0x9596000, programs 10dc355939390bdc and 918adef1afaa39d4) was unbound and bound again.

What I tried

RX_VK_BATCH=0: same hang (a plain submit ... failed (-4), no longer a batched one), so batching is not the cause.
RX_VK_CHECKPOINTS=1: no marker lines after device lost: reading the fault info. I do not know if RADV supports them.
There are no 53caac5f lines in either log, so it may be another shader than the one in your fix.
Free memory fell from about 11 GB to 0.57 GB in about a minute before the hang in run 2.

I can run RX_VK_TABLE_NOW=1 as a control, or any other switch or build you want tested, and can attach the full logs. The measurements were taken on the device; I used an AI assistant (Claude Code) to read the logs and write this up.

**Still `device lost` on a build with 9bb52a9: Linux, Steam Deck (RADV VANGOGH), reproducible** You asked for the lines after `device lost: reading` from builds that already include 9bb52a9. Two runs on a Steam Deck, both ending in a GPU hang at the same point. **Setup** - Steam Deck, `AMD Custom GPU 0405 (RADV VANGOGH)`, Vulkan 1.4.350, SteamOS build 20261023.100, kernel 7.2.7-valve2. - 15.2 GB RAM visible (BIOS UMA frame buffer set to 1 GB; with 4 GB the machine ran out of memory instead). - Hookshot Linux AppImage of `main` at `3601726` (source 50b3813b), built on the Deck itself, game v131072. Program cache warm (39,918 keys). - `hookshot_options.ini`: quality preset Custom (no VRAM budget), resolution Auto, frame generation off, game speed Real time, handheld profile on. **What happens** Intro and title run at 53 to 60 fps. When the save loads (Kokiri Forest), textures go from about 130 MB to about 3.5 GB within seconds (run 1: 1500 textures, 3496 MB), then the GPU hangs. Kernel: ``` amdgpu: ring gfx_0.0.0 timeout, signaled seq=66295, emitted seq=66296 amdgpu: Ring gfx_0.0.0 reset failed amdgpu: GPU reset begin! (MODE2 reset, "GPU reset succeeded") ``` Hookshot (run 2): ``` vk: table entry UNBOUND at frame 1435: prog ae384cff0ad0f467 ssbo 0 of 3 (c0+0x880) addr 0 size 0 (before 0x107f7e8000) vk: table entry UNBOUND at frame 1435: prog 918adef1afaa39d4 ssbo 0 of 3 (c0+0x880) addr 0 size 0 (before 0x107f7e8000) vk: table entry UNBOUND at frame 1435: prog d13aac3cec1a4bb5 ssbo 0 of 3 (c0+0x880) addr 0 size 0 (before 0x107f7e8000) vk: texture 0x10a7bb6c00 6144x6144x1 x1 (NVN 0x7b, 191 MB): its 1 top level(s) left out (47 MB) radv/amdgpu: The CS has been cancelled because the context is lost. This context is guilty of a hard recovery. vk: submit 10046 failed (-4) vk: device lost: reading the fault info (RX_VK_CHECKPOINTS=1 for markers) ``` Run 1 ended the same way at frame 3324, after the same table entry (`c0+0x880`, addr `0x107f7e8000`, size `0x9596000`, programs `10dc355939390bdc` and `918adef1afaa39d4`) was unbound and bound again. **What I tried** - `RX_VK_BATCH=0`: same hang (a plain `submit ... failed (-4)`, no longer a batched one), so batching is not the cause. - `RX_VK_CHECKPOINTS=1`: no marker lines after `device lost: reading the fault info`. I do not know if RADV supports them. - There are no `53caac5f` lines in either log, so it may be another shader than the one in your fix. - Free memory fell from about 11 GB to 0.57 GB in about a minute before the hang in run 2. I can run `RX_VK_TABLE_NOW=1` as a control, or any other switch or build you want tested, and can attach the full logs. The measurements were taken on the device; I used an AI assistant (Claude Code) to read the logs and write this up.

NotRexed commented

2026-11-08 12:43:55 +00:00

(Migrated from github.com)

Release https://github.com/NotRexed/Hookshot/releases/tag/fast-updates no longer gives back memory the GPU used, on Windows (it was meant for a Linux AMD driver; fuxau traced the Arc device lost to it in #17, and the AMD logs show device lost right after it). If you had device lost, could you try this release and tell us? On Linux/Steam Deck the old behaviour stays; there, please try RX_VK_KEEP_RETIRED=1 and say whether the device lost goes away.

Release https://github.com/NotRexed/Hookshot/releases/tag/fast-updates no longer gives back memory the GPU used, on Windows (it was meant for a Linux AMD driver; fuxau traced the Arc device lost to it in #17, and the AMD logs show device lost right after it). If you had device lost, could you try this release and tell us? On Linux/Steam Deck the old behaviour stays; there, please try `RX_VK_KEEP_RETIRED=1` and say whether the device lost goes away.

wimw85 commented

2026-11-08 13:03:55 +00:00

(Migrated from github.com)

Steam Deck, RX_VK_KEEP_RETIRED=1: the GPU hang is gone, but it runs out of memory at the same point

Tried as you asked, on the build from my earlier comment (main at 3601726, which is 60 commits after f2203e4, so it has the switch).

Result

With RX_VK_KEEP_RETIRED=1 the log shows imports: 50 live (1296 MB), 37 retired still imported (880 MB), 0 retired given back (0 MB). Before it was 38 retired given back (1088 MB) right before the hang.
There is no ring gfx_0.0.0 timeout and no GPU reset any more (both earlier runs had them).
Instead the load of the save ends in radv/amdgpu: Not enough memory for command submission (kernel: amdgpu [drm] *ERROR* Not enough memory for command submission!), followed by batched submit ... failed (-4) and device lost: reading the fault info.
Same place every time: about 1500 textures (3486 MB in one run, 3444 MB in the other) while free memory falls from about 5.5 GB to 0.45 GB in 14 seconds.

Two runs

Quality preset Custom: ends as above at 1500 textures, 3486 MB.
Quality preset Minimum (the log shows textures 132 MB (budget 2200 MB)): ends the same way at 1500 textures, 3444 MB, so the 2200 MB texture budget does not hold while the save loads.

Machine: Steam Deck, RADV VANGOGH, 15.2 GB RAM visible (BIOS UMA frame buffer 1 GB, GTT 8 GB). The log says boot: image 96 MB ..., 9809 MB of memory for the game, so the guest plus 3.4 GB of textures plus about 2.2 GB of imports does not seem to fit.

So with the switch the failure changes from a GPU hang to an out-of-memory. I can try anything that caps the textures during the load (a different budget switch, handheld profile settings, or a build you want tested) and can attach the full logs.

(Measured on the device; the write-up was prepared with an AI assistant, Claude Code.)

**Steam Deck, `RX_VK_KEEP_RETIRED=1`: the GPU hang is gone, but it runs out of memory at the same point** Tried as you asked, on the build from my earlier comment (main at `3601726`, which is 60 commits after `f2203e4`, so it has the switch). **Result** - With `RX_VK_KEEP_RETIRED=1` the log shows `imports: 50 live (1296 MB), 37 retired still imported (880 MB), 0 retired given back (0 MB)`. Before it was `38 retired given back (1088 MB)` right before the hang. - There is **no `ring gfx_0.0.0 timeout` and no GPU reset** any more (both earlier runs had them). - Instead the load of the save ends in `radv/amdgpu: Not enough memory for command submission` (kernel: `amdgpu [drm] *ERROR* Not enough memory for command submission!`), followed by `batched submit ... failed (-4)` and `device lost: reading the fault info`. - Same place every time: about 1500 textures (3486 MB in one run, 3444 MB in the other) while free memory falls from about 5.5 GB to 0.45 GB in 14 seconds. **Two runs** - Quality preset Custom: ends as above at 1500 textures, 3486 MB. - Quality preset Minimum (the log shows `textures 132 MB (budget 2200 MB)`): ends the same way at 1500 textures, **3444 MB**, so the 2200 MB texture budget does not hold while the save loads. **Machine:** Steam Deck, RADV VANGOGH, 15.2 GB RAM visible (BIOS UMA frame buffer 1 GB, GTT 8 GB). The log says `boot: image 96 MB ..., 9809 MB of memory for the game`, so the guest plus 3.4 GB of textures plus about 2.2 GB of imports does not seem to fit. So with the switch the failure changes from a GPU hang to an out-of-memory. I can try anything that caps the textures during the load (a different budget switch, handheld profile settings, or a build you want tested) and can attach the full logs. (Measured on the device; the write-up was prepared with an AI assistant, Claude Code.)

lenorocraft3-png commented

2026-11-08 13:16:54 +00:00

(Migrated from github.com)

Same as wimw85, but on Windows + NVIDIA: the Minimum (4GB) texture budget
is not held after loading a save.

GPU: GTX 1650 Super 4 GB (driver 621.14), Ryzen 5 5500, 16 GB RAM, Windows 11
Build: fast-updates (4187feb5), preset Minimum (4GB), Handheld profile On

Title screen: 60 fps, textures ~115 MB.
After loading a save in Kokiri Forest:
"VRAM: 3567 MB used of 3116 MB allowed ... textures 1589 MB (budget 1044 MB)"
-> 4 fps. Same on low-vram (textures up to 2.7 GB).
No device lost. On the older foliage build I had ~30 fps here.
Log attached.

hookshot_log.txt

Same as wimw85, but on Windows + NVIDIA: the Minimum (4GB) texture budget is not held after loading a save. GPU: GTX 1650 Super 4 GB (driver 621.14), Ryzen 5 5500, 16 GB RAM, Windows 11 Build: fast-updates (4187feb5), preset Minimum (4GB), Handheld profile On Title screen: 60 fps, textures ~115 MB. After loading a save in Kokiri Forest: "VRAM: 3567 MB used of 3116 MB allowed ... textures 1589 MB (budget 1044 MB)" -> 4 fps. Same on low-vram (textures up to 2.7 GB). No device lost. On the older foliage build I had ~30 fps here. Log attached. [hookshot_log.txt](https://github.com/user-attachments/files/34543870/hookshot_log.txt)

rbx-br commented

2026-11-08 13:32:11 +00:00

(Migrated from github.com)

Update on RX 6900 XT with the latest release (fast-updates, source 4187feb5). This time the crash happened while loading a save, and the game closed by itself about 30 seconds after the driver failure.

Setup: RX 6900 XT, driver 32.0.21045.20006 (Adrenalin 26.10.2), Xeon E5-2697 v3, 64 GB RAM, Windows 11 25h2.

What the log shows (attached, line numbers refer to it):

Line 747: view of texture 0x1133ccb000 (vk 139) as vk 95 ... not made, 16 times (layers 0 to 15).
Lines 765-779: six mesh draw: program 1a70. Line 42 says task and mesh programs built without the driver's optimizer (AMD; ...).
Lines 836 and 855: slow submit: 362.7 ms and a 171.7 ms dispatch.
Lines 878-886: three storage buffer NOT bound (reads zeros) with values that look like garbage (size 0x3f800000, which is the bit pattern of 1.0f, and addr 0x3e0000003ba478e8 size 0x44ca2618).
Lines 888-904: ten graphics pipeline ... not made (-4) (VK_ERROR_DEVICE_LOST), followed by five compute pipelines failing with -4 (0ba2f076e4a94acf, 53fcaee4dc45c6ee, 26fe0c70ab9fc547, 685adb41edc19de7, 6637ac1d052d22d4).
Line 935: pool 0x7fff4bda00+0x8 not imported (VkResult -1000072003).
Line 938: the TAAU network fd029a317a9a4b94 is stood in for ... its pipeline was not made.
Line 940: slow submit: 30423.3 ms (draw 30327.7 ms), a ~30 s stall.
Lines 945-946, right after the stall, the process itself dies:
CRASH: exception 0xc0000005 (write 0x28d004b1000, guest 0xfffff28d004b1000) at amdvlk64.dll+0x2f1c0fb
host backtrace: amdvlk64.dll+0x2f1c0fb amdvlk64.dll+0x4953eb amdvlk64.dll+0x459901 amdvlk64.dll+0x1abe07 hookshot_vk.dll+0x14efe2 hookshot_vk.dll+0x1c001b ..., on the Presentation Thread. Guest backtrace: lr 0x80590358 0x80df26c0 0x834a2ec8.

So after the device is lost, the GPU stalls for about 30 s and then the emulator crashes with an access violation inside the AMD Vulkan driver (amdvlk64.dll). The game closing on its own matches this.

It does not look like VRAM pressure: the last VRAM: line says 2648 MB used of 15569 MB allowed.

hookshot_log.txt

Update on RX 6900 XT with the latest release (fast-updates, source 4187feb5). This time the crash happened while loading a save, and the game closed by itself about 30 seconds after the driver failure. **Setup:** RX 6900 XT, driver 32.0.21045.20006 (Adrenalin 26.10.2), Xeon E5-2697 v3, 64 GB RAM, Windows 11 25h2. **What the log shows (attached, line numbers refer to it):** - Line 747: `view of texture 0x1133ccb000 (vk 139) as vk 95 ... not made`, 16 times (layers 0 to 15). - Lines 765-779: six `mesh draw: program 1a70`. Line 42 says `task and mesh programs built without the driver's optimizer (AMD; ...)`. - Lines 836 and 855: `slow submit: 362.7 ms` and a 171.7 ms dispatch. - Lines 878-886: three `storage buffer NOT bound (reads zeros)` with values that look like garbage (`size 0x3f800000`, which is the bit pattern of 1.0f, and `addr 0x3e0000003ba478e8 size 0x44ca2618`). - Lines 888-904: ten `graphics pipeline ... not made (-4)` (VK_ERROR_DEVICE_LOST), followed by five compute pipelines failing with -4 (0ba2f076e4a94acf, 53fcaee4dc45c6ee, 26fe0c70ab9fc547, 685adb41edc19de7, 6637ac1d052d22d4). - Line 935: `pool 0x7fff4bda00+0x8 not imported (VkResult -1000072003)`. - Line 938: `the TAAU network fd029a317a9a4b94 is stood in for ... its pipeline was not made`. - Line 940: `slow submit: 30423.3 ms` (draw 30327.7 ms), a ~30 s stall. - Lines 945-946, right after the stall, the process itself dies: `CRASH: exception 0xc0000005 (write 0x28d004b1000, guest 0xfffff28d004b1000) at amdvlk64.dll+0x2f1c0fb` host backtrace: `amdvlk64.dll+0x2f1c0fb amdvlk64.dll+0x4953eb amdvlk64.dll+0x459901 amdvlk64.dll+0x1abe07 hookshot_vk.dll+0x14efe2 hookshot_vk.dll+0x1c001b ...`, on the Presentation Thread. Guest backtrace: `lr 0x80590358 0x80df26c0 0x834a2ec8`. So after the device is lost, the GPU stalls for about 30 s and then the emulator crashes with an access violation inside the AMD Vulkan driver (amdvlk64.dll). The game closing on its own matches this. It does not look like VRAM pressure: the last `VRAM:` line says 2648 MB used of 15569 MB allowed. [hookshot_log.txt](https://github.com/user-attachments/files/34544245/hookshot_log.txt)

Scuchimi commented

2026-11-08 13:34:45 +00:00

(Migrated from github.com)

Same flickering here on an RTX 3050 (release fast-updates, source 4187feb5, Windows).

Frame generation Off and Frame pacing Off: the picture still flickers in Kokiri Forest.
Quality preset was "Minimum (4GB)". The log shows "5446 MB of VRAM allowed: all large ASTC textures kept as BC3".
VRAM near the limit: 4855 MB used of 5143 MB allowed.
About 45-50 fps in menus, about 23 fps in gameplay.
No crash, the game closed normally.

Log attached.

hookshot_log.txt

Same flickering here on an RTX 3050 (release fast-updates, source 4187feb5, Windows). - Frame generation Off and Frame pacing Off: the picture still flickers in Kokiri Forest. - Quality preset was "Minimum (4GB)". The log shows "5446 MB of VRAM allowed: all large ASTC textures kept as BC3". - VRAM near the limit: 4855 MB used of 5143 MB allowed. - About 45-50 fps in menus, about 23 fps in gameplay. - No crash, the game closed normally. Log attached. [hookshot_log.txt](https://github.com/user-attachments/files/34544269/hookshot_log.txt)

sforsbergg commented

2026-11-08 21:18:56 +00:00

(Migrated from github.com)

still black screen for me .

im on windows 11 on ROG Ally Z1 Extreme. tried vram set to auto and at 4GB.

hookshot_log.1.txt
hookshot_log.txt

still black screen for me . im on windows 11 on ROG Ally Z1 Extreme. tried vram set to auto and at 4GB. [hookshot_log.1.txt](https://github.com/user-attachments/files/34555383/hookshot_log.1.txt) [hookshot_log.txt](https://github.com/user-attachments/files/34555384/hookshot_log.txt)

sforsbergg commented

2026-11-08 21:37:22 +00:00

(Migrated from github.com)

with vram set to 6gb the game makes the whole system reboot.

hookshot_log.txt

with vram set to 6gb the game makes the whole system reboot. [hookshot_log.txt](https://github.com/user-attachments/files/34555729/hookshot_log.txt)

NotRexed commented

2026-11-09 05:44:43 +00:00

(Migrated from github.com)

Tracked in #996, follow there. That issue collects every report of AMD driver crashes, device lost and freezes on Windows, and this one is linked from it with its GPU, version and log details. The tracker summarises the device-lost causes found here and what was fixed since. If it still happens on v2.0.6 (https://github.com/NotRexed/Hookshot/releases/tag/v2.0.6; Update next to Play in the launcher), please add a comment there with your hookshot_log.txt.

Tracked in #996, follow there. That issue collects every report of AMD driver crashes, device lost and freezes on Windows, and this one is linked from it with its GPU, version and log details. The tracker summarises the device-lost causes found here and what was fixed since. If it still happens on v2.0.6 (https://github.com/NotRexed/Hookshot/releases/tag/v2.0.6; Update next to Play in the launcher), please add a comment there with your `hookshot_log.txt`.

zackfairccff-bot commented

2026-11-09 15:28:15 +00:00

(Migrated from github.com)

Steam Deck LCD / SteamOS — Vulkan crashes and rendering issues

Hi! I've been testing Hookshot on my Steam Deck LCD running SteamOS, using the AMD RADV VANGOGH Vulkan driver.

I successfully built and launched the game natively on Linux.

Results:

The intro and initial cutscenes work.
I can reach Kokiri Forest and control Link.
Performance is sometimes around 25 FPS, with severe stuttering.
Vegetation flickers or disappears.
Some objects, including doors, become invisible.
Lighting is severely overexposed in certain scenes.
The game sometimes crashes with "VK_ERROR_DEVICE_LOST (-4)".

Configurations tested:

"RX_VK_NO_FOLLOW=1" — allowed the game to launch and progress.
"RX_VK_NO_MESH=1" — allowed gameplay but worsened rendering, with missing vegetation and geometry.
"RX_VK_KEEP_RETIRED=1" — tested without achieving stable gameplay.
Other Vulkan memory and batching options have also been investigated.

The logs include "storage buffer NOT bound", "batched submit failed (-4)", and "device lost".

I'm happy to provide screenshots, full logs, and test experimental Linux builds or Vulkan configurations.

Is there a recommended fix or branch specifically for Steam Deck / RADV VANGOGH?

Thanks for your work on Linux support!

Steam Deck LCD / SteamOS — Vulkan crashes and rendering issues Hi! I've been testing Hookshot on my Steam Deck LCD running SteamOS, using the AMD RADV VANGOGH Vulkan driver. I successfully built and launched the game natively on Linux. Results: - The intro and initial cutscenes work. - I can reach Kokiri Forest and control Link. - Performance is sometimes around 25 FPS, with severe stuttering. - Vegetation flickers or disappears. - Some objects, including doors, become invisible. - Lighting is severely overexposed in certain scenes. - The game sometimes crashes with "VK_ERROR_DEVICE_LOST (-4)". Configurations tested: - "RX_VK_NO_FOLLOW=1" — allowed the game to launch and progress. - "RX_VK_NO_MESH=1" — allowed gameplay but worsened rendering, with missing vegetation and geometry. - "RX_VK_KEEP_RETIRED=1" — tested without achieving stable gameplay. - Other Vulkan memory and batching options have also been investigated. The logs include "storage buffer NOT bound", "batched submit failed (-4)", and "device lost". I'm happy to provide screenshots, full logs, and test experimental Linux builds or Vulkan configurations. Is there a recommended fix or branch specifically for Steam Deck / RADV VANGOGH? Thanks for your work on Linux support!