Game: Longshot crashes on every startup with RX 7600 XT #1391

Open
opened 2026-11-17 10:19:16 +00:00 by BeanIssues · 5 comments
Collaborator

Reported on Discord

Longshot version and channel

Longshot v3.0.0-rc.7 (release candidate, source e5cea4aa), according to the manual crash report.

What happened

Longshot crashes during startup every time, initially showing a black screen. The dialog reads:

Longshot closed unexpectedly during the start.

What went wrong: The graphics driver failed. Update your graphics driver, and turn off overlays and tools hooked into the game.

The reporter said automatic report submission was unavailable. Disabling overlays allowed the intro animation to appear, but the game then froze and stopped working. The reporter confirms their graphics drivers are up to date.

Where in the game

At startup/title intro.

How often

Every startup.

Graphics card and driver version

AMD Radeon RX 7600 XT (16 GB); AMD proprietary driver 26.10.1 (LLPC), from the manual crash report.

CPU, RAM and operating system

AMD Ryzen 7 5800XT 8-Core Processor (3.80 GHz), 16 GB RAM, Windows 11 build 26200.

Log evidence

The attached longshot_log.1.txt is from Longshot v3.0.0-rc.7 and records Vulkan GPU hang/device loss at the intro. It reports no GPU submission finishing for 30 seconds, followed by a device fault and fatal GPU-hung message. The log shows 120 presents at 8.7 fps before the hang.

Still unknown

It is unclear whether longshot_log.1.txt is the complete original logs/longshot_log.txt or a renamed/copy of it. The earlier value 20261117T101259Z was explicitly retracted by the reporter.

Longshot #1225 is a graphics-driver-failed startup crash on Intel Arc; this report is on AMD and is not known to be the same issue.

Reported on [Discord](https://discord.com/channels/1569849200047427645/1572191838125690972/1572191838125690972) ### Longshot version and channel Longshot v3.0.0-rc.7 (release candidate, source e5cea4aa), according to the manual crash report. ### What happened Longshot crashes during startup every time, initially showing a black screen. The dialog reads: ```text Longshot closed unexpectedly during the start. What went wrong: The graphics driver failed. Update your graphics driver, and turn off overlays and tools hooked into the game. ``` The reporter said automatic report submission was unavailable. Disabling overlays allowed the intro animation to appear, but the game then froze and stopped working. The reporter confirms their graphics drivers are up to date. ### Where in the game At startup/title intro. ### How often Every startup. ### Graphics card and driver version AMD Radeon RX 7600 XT (16 GB); AMD proprietary driver 26.10.1 (LLPC), from the manual crash report. ### CPU, RAM and operating system AMD Ryzen 7 5800XT 8-Core Processor (3.80 GHz), 16 GB RAM, Windows 11 build 26200. ### Log evidence The attached `longshot_log.1.txt` is from Longshot v3.0.0-rc.7 and records Vulkan GPU hang/device loss at the intro. It reports no GPU submission finishing for 30 seconds, followed by a device fault and fatal GPU-hung message. The log shows 120 presents at 8.7 fps before the hang. ### Still unknown It is unclear whether `longshot_log.1.txt` is the complete original `logs/longshot_log.txt` or a renamed/copy of it. The earlier value `20261117T101259Z` was explicitly retracted by the reporter. ### Related Longshot #1225 is a graphics-driver-failed startup crash on Intel Arc; this report is on AMD and is not known to be the same issue.
Author
Collaborator

Files from the Discord report:

Files from the Discord report: - [image.png](https://git.magicbean.dev/attachments/f7895387-6437-4c30-afe1-9a34c4714796)
Author
Collaborator

From Discord:

The reporter attached a manually generated crash report (report-20261117T102448Z.txt). It identifies Longshot v3.0.0-rc.7 (release candidate, source e5cea4aa), Windows 11 build 26200, Vulkan, AMD Radeon RX 7600 XT with AMD proprietary driver 26.10.1 (LLPC), Ryzen 7 5800XT, 16 GB RAM. It reports a 40-second run at the title/intro, 10 crash/unclean-end records in the last day, and reason: device lost; the log end records a GPU hang with no submission finishing for 30 seconds. The attachment includes only the end of the log, not the full logs/longshot_log.txt. Still unknown: whether driver update or disabling overlays/hooks changes the crash. The earlier 20261117T101259Z remains unexplained.

From [Discord](https://discord.com/channels/1569849200047427645/1572191838125690972/1572191838125690972): The reporter attached a manually generated crash report (`report-20261117T102448Z.txt`). It identifies Longshot v3.0.0-rc.7 (release candidate, source e5cea4aa), Windows 11 build 26200, Vulkan, AMD Radeon RX 7600 XT with AMD proprietary driver 26.10.1 (LLPC), Ryzen 7 5800XT, 16 GB RAM. It reports a 40-second run at the title/intro, 10 crash/unclean-end records in the last day, and `reason: device lost`; the log end records a GPU hang with no submission finishing for 30 seconds. The attachment includes only the end of the log, not the full `logs/longshot_log.txt`. Still unknown: whether driver update or disabling overlays/hooks changes the crash. The earlier `20261117T101259Z` remains unexplained.
Author
Collaborator

Updated with more from Discord: Added confirmation that the AMD graphics drivers are up to date; kept prior log/overlay findings.

Files:

Updated with more from [Discord](https://discord.com/channels/1569849200047427645/1572191838125690972/1572197167555420191): Added confirmation that the AMD graphics drivers are up to date; kept prior log/overlay findings. Files: - [longshot_log.1.txt](https://git.magicbean.dev/attachments/8efaa204-7e36-4d32-b4c7-1aa9131989c7)
Author
Collaborator

From Discord:

The reporter clarified that longshot_log.1.txt is logs/longshot_log.txt, just renamed. Please treat the attached file as the game's log; disregard the retracted timestamp noted in the issue.

From [Discord](https://discord.com/channels/1569849200047427645/1572191838125690972/1572191838125690972): The reporter clarified that `longshot_log.1.txt` is `logs/longshot_log.txt`, just renamed. Please treat the attached file as the game's log; disregard the retracted timestamp noted in the issue.
Owner

Finding from the attached log: this looks like the same hang as #1376, and very likely #1429. I don't think it's new in rc.6 or rc.7.

What the log shows

  • Both logs here (#1391: RX 7600 XT; #1429: Radeon 8060S) end the same way: AMD's Windows driver 26.10.1 (LLPC), fullscreen, Frame generation 4x with present ahead on, and exactly 7 submissions sent but not finished (the timeline at 897, 904 sent / 78, 85 sent). The device fault is empty and the driver never reports the device lost, so there was no TDR. The queue is waiting on something, not running a shader that hung.
  • Neither log has a vk: fsr3 inputs: or vk: fsr3fi: ... makes line, so FSR 3.1 frame generation (#1103, new in rc.6) never ran before the hang. It is not the cause.
  • In #1391 the game started behind another window: hid: game window not in front (foreground window: ... "Chrome_WidgetWin_1"). In #1429, starting from the launcher or its shortcut hangs and starting longshot.exe directly works. #1376 is the same hang about 1 to 20 s after Alt+Tab on an RX 9070 XT, same driver, at 2x.
  • Before the hang the 7600 XT ran at 120 presents, 8.7 fps, with present 29.93 ms a frame in the replay. That is about one 100 ms wait per frame-generation set. The display thread waits up to 100 ms for the set's first copy (vk_framegen.cpp display_main), and a copy only finishes once its swapchain image is given back.

The cause, as far as the code shows
present_image (gpu/vk/vk_replay.cpp) submits the copy into the swapchain image on the replay's queue, waiting on the acquire's semaphore (waits[0].semaphore = f.acquired). AMD's Windows driver returns from vkAcquireNextImageKHR at once and leaves the wait to that semaphore. While the fullscreen window is behind another one, the image comes back late or not at all. When it doesn't come back, the GPU-side wait blocks every later submission of the game's work, and the GPU hang watch ends the game 30 s later.

Frame generation 4x with present ahead makes this more likely: a set holds 4 acquired images at once. Also, present_image has only 3 frame slots, so the set's 4th acquire first waits, with no time limit, for the set's own first copy to finish, before any of the set has been presented.

Status: the agent on #1376 has a local fix for exactly this wait (acquire with a fence, waited on the CPU, and the image held when it doesn't come back in 100 ms), not opened as a PR yet. That fix should cover this issue and #1429 too. Its Test Farm run on Windows AMD should include starting from the launcher with Frame generation 4x through the intro. Not opening a second PR for the same wait.

For the reporter, to confirm on rc.7: put RX_VK_PRESENT_AHEAD=0 in longshot_env.txt (next to longshot_options.ini), or set Frame generation Off, and start the game with nothing else in front of it.

🤖 Generated with Claude Code

Finding from the attached log: this looks like the same hang as [#1376](https://git.magicbean.dev/MagicBeanDev/Longshot/issues/1376), and very likely [#1429](https://git.magicbean.dev/MagicBeanDev/Longshot/issues/1429). I don't think it's new in rc.6 or rc.7. **What the log shows** - Both logs here (#1391: RX 7600 XT; #1429: Radeon 8060S) end the same way: AMD's Windows driver 26.10.1 (LLPC), fullscreen, Frame generation 4x with `present ahead` on, and exactly 7 submissions sent but not finished (`the timeline at 897, 904 sent` / `78, 85 sent`). The device fault is empty and the driver never reports the device lost, so there was no TDR. The queue is waiting on something, not running a shader that hung. - Neither log has a `vk: fsr3 inputs:` or `vk: fsr3fi: ... makes` line, so FSR 3.1 frame generation (#1103, new in rc.6) never ran before the hang. It is not the cause. - In #1391 the game started behind another window: `hid: game window not in front (foreground window: ... "Chrome_WidgetWin_1")`. In #1429, starting from the launcher or its shortcut hangs and starting `longshot.exe` directly works. #1376 is the same hang about 1 to 20 s after Alt+Tab on an RX 9070 XT, same driver, at 2x. - Before the hang the 7600 XT ran at `120 presents, 8.7 fps`, with `present 29.93` ms a frame in the replay. That is about one 100 ms wait per frame-generation set. The display thread waits up to 100 ms for the set's first copy (vk_framegen.cpp `display_main`), and a copy only finishes once its swapchain image is given back. **The cause, as far as the code shows** `present_image` (gpu/vk/vk_replay.cpp) submits the copy into the swapchain image on the replay's queue, waiting on the acquire's semaphore (`waits[0].semaphore = f.acquired`). AMD's Windows driver returns from `vkAcquireNextImageKHR` at once and leaves the wait to that semaphore. While the fullscreen window is behind another one, the image comes back late or not at all. When it doesn't come back, the GPU-side wait blocks every later submission of the game's work, and the GPU hang watch ends the game 30 s later. Frame generation 4x with present ahead makes this more likely: a set holds 4 acquired images at once. Also, present_image has only 3 frame slots, so the set's 4th acquire first waits, with no time limit, for the set's own first copy to finish, before any of the set has been presented. **Status:** the agent on #1376 has a local fix for exactly this wait (acquire with a fence, waited on the CPU, and the image held when it doesn't come back in 100 ms), not opened as a PR yet. That fix should cover this issue and #1429 too. Its Test Farm run on Windows AMD should include starting from the launcher with Frame generation 4x through the intro. Not opening a second PR for the same wait. **For the reporter**, to confirm on rc.7: put `RX_VK_PRESENT_AHEAD=0` in `longshot_env.txt` (next to `longshot_options.ini`), or set Frame generation Off, and start the game with nothing else in front of it. 🤖 Generated with [Claude Code](https://claude.com/claude-code)
Sign in to join this conversation.
No milestone
No project
No assignees
2 participants
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#1391
No description provided.