Game: AMD driver timeout when loading any save on RX 7900 XTX #986

Open
opened 2026-11-16 00:50:22 +00:00 by BeanIssues · 3 comments
Collaborator

Reported by Curtis on Discord

Longshot version / channel

Initial report: v3.0.0-rc.3 (source cb01b5a3), release candidate. Follow-up reproduction: nightly-287-0407ae24 (source 932a85a8), Nightly.

What happened / steps

Initial report: launching from the Longshot Launcher, selecting Continue at the title screen, then Load Save causes an immediate AMD driver timeout. The reporter says this happens with all saves and autosaves, including saves from different points in the game.

Ian tested nightly-287-0407ae24: initially the game started, but then crashed; Task Manager showed 28/32 GB RAM in use and all displays flickered and went black. The included crash dialog says Longshot ended without a crash report, possibly because it was stopped by the system, often for running out of memory. The AMD Bug Report Tool reports a driver timeout. Nightly log recorded repeated VRAM full fallback messages for images at 2560x1440, 1280x720 and 1040x1040, putting them in system memory. Earlier screenshots show AMD metrics at 100% GPU utilization.

Where in the game / how often

Initial reproduction occurs immediately while loading any tested save/autosave. Nightly test establishes the game can start, but then fails during the run; exact time/location and repeat frequency for this nightly run are unknown.

Graphics card and driver version

AMD Radeon RX 7900 XTX (24 GB), AMD proprietary driver 25.9.1 (LLPC) in the nightly report. The log also lists AMD Radeon(TM) Graphics. In the initial RC report, settings selected device 1: AMD Radeon(TM) Graphics despite the reporter identifying the RX 7900 XTX; nightly options record gpu_device=0, and the nightly report lists both devices. Exact device actually used in each run remains to be established.

CPU, RAM and operating system

AMD Ryzen 7 7700X (16 threads), 32 GB DDR5-6000 (31 GB usable), Windows 11 Home 26H2, build 26300.9974 (the report records build 26300). During the nightly failure, Task Manager showed 28/32 GB RAM in use.

Settings / environment

Initial RC bug report records High (12GB) quality preset, 2048x1152 render resolution, frame generation Auto (Built-in FG), frame pacing On, AMD Anti-Lag 2 On, GPU device 1: AMD Radeon(TM) Graphics, and VRAM budget 7.5 GB. Nightly report records quality preset 6, ultrawide enabled, output resolution 2560x1440, frame generation off, frame pacing on, upscaling value 2, GPU device 0, memory budget 5. AMD FSR 4 loader was present. Other settings are in the attached reports.

Logs

Initial run recorded: [hos Presentation Thread] vk: device lost: reading the fault info (RX_VK_CHECKPOINTS=1 for markers).

Nightly report's log tail has no fatal/crash/device-lost marker in the unclean record; it does contain the VRAM-full fallback lines described above. The process ended without a clean exit, and the report does not prove whether the system stopped it for memory pressure or the driver timeout caused termination.

Still unknown

  • Exact point/map and time in the nightly run when the failure occurred
  • Whether the nightly failure is reproducible and whether it happens on all saves
  • Which of the two listed GPUs Longshot used in each run; why the initial report selected integrated AMD graphics while the reporter identified the RX 7900 XTX
  • Whether disabling FSR 4 / changing quality or output resolution / selecting the RX 7900 XTX explicitly changes the behavior

Related but not established as the same issue: #717 concerns artifacts and driver timeouts on an RX 580, and #973 concerns a black-screen freeze on an RX 580 after loading a save.

Attachments

The original report's report.txt and logs, plus screenshots showing the timeout and metrics, are attached to the issue. The follow-up nightly archive contained a crash report and screenshot; its report contents are summarized above. The screenshots show the Longshot unexpected-close dialog, AMD driver timeout dialog, and RX 7900 XTX metrics.

The device-lost log line itself is clarified by PR #670, but that change only improves the diagnostic message and does not fix the crash. It is not in a release yet; it comes with the next nightly.

Reported by [Curtis on Discord](https://discord.com/channels/1569849200047427645/1570776416906252308/1570776416906252308) ### Longshot version / channel Initial report: v3.0.0-rc.3 (source cb01b5a3), release candidate. Follow-up reproduction: nightly-287-0407ae24 (source 932a85a8), Nightly. ### What happened / steps Initial report: launching from the Longshot Launcher, selecting Continue at the title screen, then Load Save causes an immediate AMD driver timeout. The reporter says this happens with all saves and autosaves, including saves from different points in the game. Ian tested nightly-287-0407ae24: initially the game started, but then crashed; Task Manager showed 28/32 GB RAM in use and all displays flickered and went black. The included crash dialog says Longshot ended without a crash report, possibly because it was stopped by the system, often for running out of memory. The AMD Bug Report Tool reports a driver timeout. Nightly log recorded repeated `VRAM full` fallback messages for images at 2560x1440, 1280x720 and 1040x1040, putting them in system memory. Earlier screenshots show AMD metrics at 100% GPU utilization. ### Where in the game / how often Initial reproduction occurs immediately while loading any tested save/autosave. Nightly test establishes the game can start, but then fails during the run; exact time/location and repeat frequency for this nightly run are unknown. ### Graphics card and driver version AMD Radeon RX 7900 XTX (24 GB), AMD proprietary driver 25.9.1 (LLPC) in the nightly report. The log also lists AMD Radeon(TM) Graphics. In the initial RC report, settings selected device 1: AMD Radeon(TM) Graphics despite the reporter identifying the RX 7900 XTX; nightly options record `gpu_device=0`, and the nightly report lists both devices. Exact device actually used in each run remains to be established. ### CPU, RAM and operating system AMD Ryzen 7 7700X (16 threads), 32 GB DDR5-6000 (31 GB usable), Windows 11 Home 26H2, build 26300.9974 (the report records build 26300). During the nightly failure, Task Manager showed 28/32 GB RAM in use. ### Settings / environment Initial RC bug report records High (12GB) quality preset, 2048x1152 render resolution, frame generation Auto (Built-in FG), frame pacing On, AMD Anti-Lag 2 On, GPU device 1: AMD Radeon(TM) Graphics, and VRAM budget 7.5 GB. Nightly report records quality preset 6, ultrawide enabled, output resolution 2560x1440, frame generation off, frame pacing on, upscaling value 2, GPU device 0, memory budget 5. AMD FSR 4 loader was present. Other settings are in the attached reports. ### Logs Initial run recorded: `[hos Presentation Thread] vk: device lost: reading the fault info (RX_VK_CHECKPOINTS=1 for markers)`. Nightly report's log tail has no fatal/crash/device-lost marker in the unclean record; it does contain the VRAM-full fallback lines described above. The process ended without a clean exit, and the report does not prove whether the system stopped it for memory pressure or the driver timeout caused termination. ### Still unknown - Exact point/map and time in the nightly run when the failure occurred - Whether the nightly failure is reproducible and whether it happens on all saves - Which of the two listed GPUs Longshot used in each run; why the initial report selected integrated AMD graphics while the reporter identified the RX 7900 XTX - Whether disabling FSR 4 / changing quality or output resolution / selecting the RX 7900 XTX explicitly changes the behavior Related but not established as the same issue: #717 concerns artifacts and driver timeouts on an RX 580, and #973 concerns a black-screen freeze on an RX 580 after loading a save. ### Attachments The original report's report.txt and logs, plus screenshots showing the timeout and metrics, are attached to the issue. The follow-up nightly archive contained a crash report and screenshot; its report contents are summarized above. The screenshots show the Longshot unexpected-close dialog, AMD driver timeout dialog, and RX 7900 XTX metrics. ### Related change The device-lost log line itself is clarified by PR #670, but that change only improves the diagnostic message and does not fix the crash. It is not in a release yet; it comes with the next nightly.
Author
Collaborator
Files from the Discord report: - [report.txt](https://git.magicbean.dev/attachments/92148aa3-e884-4b5d-b85f-2d110ee7a436) - [longshot_log.txt](https://git.magicbean.dev/attachments/468b138e-864b-4290-a785-46b5d981cf13) - [longshot_log.1.txt](https://git.magicbean.dev/attachments/80b818c0-037e-45e8-958d-531b05d24011) - [While_its_crashing.png](https://git.magicbean.dev/attachments/37c44eba-7ec7-4ab5-90c4-ae7a23fed68f) - [post-crash.png](https://git.magicbean.dev/attachments/9195d01a-90c0-4eca-83c6-6650aba4a7a3) - [Game_up_before_crash.png](https://git.magicbean.dev/attachments/05a95bf5-ae4b-41b4-b00b-daa95fcf428d)
Author
Collaborator

Updated with more from ian_ShmianTTV on Discord: Added Ian's nightly-287 reproduction and crash report: device list, 25.9.1 driver, 2560x1440 settings, 28/32 GB RAM observation, and VRAM-full fallback log lines; noted unresolved reproduction and device-selection details.

Files:

Updated with more from [ian_ShmianTTV on Discord](https://discord.com/channels/1569849200047427645/1570776416906252308/1571749600539443271): Added Ian's nightly-287 reproduction and crash report: device list, 25.9.1 driver, 2560x1440 settings, 28/32 GB RAM observation, and VRAM-full fallback log lines; noted unresolved reproduction and device-selection details. Files: - [report-20261116T044838Z.txt](https://git.magicbean.dev/attachments/3aba637d-b2aa-4664-9224-6f6cc3bc8ced) - [Screenshot_2026-11-15_224832.png](https://git.magicbean.dev/attachments/f21ead91-6ba8-44fa-b1dd-0937cfb39d11)
Author
Collaborator

From Curtis on Discord:

Ian clarified that the RX 7900 XTX was the active GPU during the nightly-287-0407ae24 test. This resolves the uncertainty about which listed GPU was active for that run; initial v3.0.0-rc.3 selection remains as recorded.

From [Curtis on Discord](https://discord.com/channels/1569849200047427645/1570776416906252308/1570776416906252308): Ian clarified that the RX 7900 XTX was the active GPU during the nightly-287-0407ae24 test. This resolves the uncertainty about which listed GPU was active for that run; initial v3.0.0-rc.3 selection remains as recorded.
Sign in to join this conversation.
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
MagicBeanDev/Longshot#986
No description provided.