MagicBeanDev/Longshot

Watch

Fork 98

Game: [CRITICAL] GPU hang + Windows forced reset in Hyrule Castle Town at night - RTX 5050 8GB #668

New issue

Open

opened 2026-11-15 03:00:39 +00:00 by guilhermervalle · 3 comments

guilhermervalle commented

2026-11-15 03:00:39 +00:00

Longshot version

v3.0.0-rc.3

Channel

Release candidate

What kind of problem

Crash, freeze or black screen

What happened

System hard-crashed (Kernel-Power 41) with forced Windows reset while entering Hyrule Castle Town at night.

The last entries in longshot_log.txt before the crash show the game exceeded its own VRAM budget:

"vk: over the driver's video-memory allowance (7.1 of 6.9 GB): DLAA may stutter here; Quality or Balanced uses less"

Combined with massive shader compilation stall:

11,310 pipelines compiled in 6994ms (~7 second freeze)
VRAM compaction ran 17 times trying to free memory - insufficient
FPS collapsed from 100+ to 23.2 before hang

Windows Event Viewer confirms:

Kernel-Power 41 (Critical) at 23:30:06 - forced reset
EventLog 6008 at 23:30:16 - unexpected shutdown confirmed
NO nvlddmkm event was generated - the hang was so severe TDR (Timeout Detection and Recovery) didn't fire, indicating a hard freeze rather than recoverable driver hang

Hardware context (important): this PC has superior specs to Switch 2 in every category - RTX 5050 (~15 TFLOPS vs ~3), 40GB total memory (8GB VRAM + 32GB RAM vs 12GB unified). The only Switch 2 advantage is unified memory addressing, which is where the recomp is clearly not handling PC memory architecture well.

8GB VRAM is the industry baseline (RTX 3060/4060/5050 - most common NVIDIA cards). If the port can't run on 8GB cards at advertised settings, this limits adoption significantly.

Suggested fixes:

Respect "Limite de VRAM" as HARD limit (currently advisory - game went to 7.1GB despite 6GB config)
Auto quality fallback (DLAA → Quality 67%) when VRAM > 85% of limit
Pre-compile Castle Town shaders during loading, not on-demand (7s freeze is unacceptable)
BC7 compression on 4K ASTC textures during build
Per-GPU tier profiles - auto-detect 8GB cards and set conservative defaults

Issue appears reproducible (just entered the area, didn't do anything exotic), but I'm hesitant to crash my PC again without first applying workaround (DLAA → Quality 67%). Can test if requested.

Where in the game

Hyrule Castle Town (Market Town), at night, walking toward the central fountain. Had just arrived from Hyrule Field with child Link. Was pursuing the "lost dog" sidequest.

How often

Once so far

Graphics card and driver version

NVIDIA GeForce RTX 5050 8GB, driver 616.56

CPU, RAM and operating system

AMD Ryzen 5 8600G, 32 GB DDR5 5600MHz, Windows 11 Pro

Changes from the defaults

Insert menu settings at crash:

Resolução interna: DLAA (Native 1920x1080)
Geração de quadros: 2x
Limite de VRAM: 6 GB
NVIDIA Reflex: Ligado + Boost
Tela cheia: Ligado
Perfil: Modo TV
DLSS model: Transformer (K)
Filtragem anisotrópica: 16x
Game language (at build time): Portuguese (BR)
Build options: NVIDIA DLSS enabled, "Prepare shaders now" enabled during build

No longshot_env.txt. No overlays, ReShade, or frame-gen injectors hooked in.

longshot_log.txt

See attached longshot_log.txt in longshot_log.zip long (1.3MB) above.

Key excerpts from the crash:

Screenshots or video

Before you send it

The problem is still there on the newest release or nightly.
I searched the open issues for it.
I'm not attaching game files, keys, the game's executable, its .bin caches or saves.

### Longshot version v3.0.0-rc.3 ### Channel Release candidate ### What kind of problem Crash, freeze or black screen ### What happened System hard-crashed (Kernel-Power 41) with forced Windows reset while entering Hyrule Castle Town at night. The last entries in longshot_log.txt before the crash show the game exceeded its own VRAM budget: "vk: over the driver's video-memory allowance (7.1 of 6.9 GB): DLAA may stutter here; Quality or Balanced uses less" Combined with massive shader compilation stall: - 11,310 pipelines compiled in 6994ms (~7 second freeze) - VRAM compaction ran 17 times trying to free memory - insufficient - FPS collapsed from 100+ to 23.2 before hang Windows Event Viewer confirms: - Kernel-Power 41 (Critical) at 23:30:06 - forced reset - EventLog 6008 at 23:30:16 - unexpected shutdown confirmed - NO nvlddmkm event was generated - the hang was so severe TDR (Timeout Detection and Recovery) didn't fire, indicating a hard freeze rather than recoverable driver hang Hardware context (important): this PC has superior specs to Switch 2 in every category - RTX 5050 (~15 TFLOPS vs ~3), 40GB total memory (8GB VRAM + 32GB RAM vs 12GB unified). The only Switch 2 advantage is unified memory addressing, which is where the recomp is clearly not handling PC memory architecture well. 8GB VRAM is the industry baseline (RTX 3060/4060/5050 - most common NVIDIA cards). If the port can't run on 8GB cards at advertised settings, this limits adoption significantly. Suggested fixes: 1. Respect "Limite de VRAM" as HARD limit (currently advisory - game went to 7.1GB despite 6GB config) 2. Auto quality fallback (DLAA → Quality 67%) when VRAM > 85% of limit 3. Pre-compile Castle Town shaders during loading, not on-demand (7s freeze is unacceptable) 4. BC7 compression on 4K ASTC textures during build 5. Per-GPU tier profiles - auto-detect 8GB cards and set conservative defaults Issue appears reproducible (just entered the area, didn't do anything exotic), but I'm hesitant to crash my PC again without first applying workaround (DLAA → Quality 67%). Can test if requested. ### Where in the game Hyrule Castle Town (Market Town), at night, walking toward the central fountain. Had just arrived from Hyrule Field with child Link. Was pursuing the "lost dog" sidequest. ### How often Once so far ### Graphics card and driver version NVIDIA GeForce RTX 5050 8GB, driver 616.56 ### CPU, RAM and operating system AMD Ryzen 5 8600G, 32 GB DDR5 5600MHz, Windows 11 Pro ### Changes from the defaults Insert menu settings at crash: - Resolução interna: DLAA (Native 1920x1080) - Geração de quadros: 2x - Limite de VRAM: 6 GB - NVIDIA Reflex: Ligado + Boost - Tela cheia: Ligado - Perfil: Modo TV - DLSS model: Transformer (K) - Filtragem anisotrópica: 16x - Game language (at build time): Portuguese (BR) - Build options: NVIDIA DLSS enabled, "Prepare shaders now" enabled during build No longshot_env.txt. No overlays, ReShade, or frame-gen injectors hooked in. ### longshot_log.txt See attached longshot_log.txt in longshot_log.zip long (1.3MB) above. Key excerpts from the crash: ### Screenshots or video ![image](/attachments/0e273b15-4a8d-4d19-a449-2646b5ce62b4) ![image](/attachments/ffda79f3-c35c-4bb9-9909-72f9a48f9f70) ![image](/attachments/e33c3cd7-54d4-4d2f-b5d1-14b5733e9a85) ![image](/attachments/19526e00-3c37-4e9a-a727-a2c406f0e9f5) ### Before you send it - [x] The problem is still there on the newest release or nightly. - [x] I searched the open issues for it. - [x] I'm not attaching game files, keys, the game's executable, its .bin caches or saves.

image.png

272 KiB

image.png

354 KiB

image.png

357 KiB

image.png

383 KiB

longshot_log.zip

1.2 MiB

guilhermervalle added the

kind

bug

label

2026-11-15 03:00:39 +00:00

Zapiltar referenced this issue

2026-11-15 06:31:19 +00:00

WIP: perf(vk): the 8 GB cards' VRAM budget step mirrors 2 GB of the game's pools, not 1 #724

HumanBean referenced this issue

2026-11-15 07:47:38 +00:00

perf(vk): the VRAM budgets set by the card, not a menu row or a preset; 8 GB cards' mirrors 2 GB #735

HumanBean referenced this issue from a commit

2026-11-15 07:56:56 +00:00

perf(vk): the VRAM budgets set by the card, not a menu row or a preset; 8 GB cards' mirrors 2 GB (#735)

malon referenced this issue

2026-11-15 15:11:03 +00:00

Game: Keeps on crashing in death mountain #817

BeanIssues commented

2026-11-15 16:27:36 +00:00

Collaborator

From tboogie77 on Discord:

Additional report: black screen with audio/gameplay continuing on entering Castle Town, ROG Ally Z1 Extreme; Frame Generation off, VRAM setting 3.6 GB. Discord longshot_log.txt attachment was too large for IssueBot to read, so its contents have not been verified.

From [tboogie77 on Discord](https://discord.com/channels/1569849200047427645/1571562199007105064/1571562199007105064): Additional report: black screen with audio/gameplay continuing on entering Castle Town, ROG Ally Z1 Extreme; Frame Generation off, VRAM setting 3.6 GB. Discord longshot_log.txt attachment was too large for IssueBot to read, so its contents have not been verified.

BeanIssues commented

2026-11-15 16:48:57 +00:00

Collaborator

From tboogie77 on Discord:

Update from tboogie77: the black screen with audio/gameplay continuing still occurs on nightly-189-aa95d35f on a ROG Ally Z1 Extreme, with Frame Generation off and VRAM set to 3.6 GB. The attached log is for nightly-189; it reports 2048x1152 output, graphics scale 1, and 59.9–60 fps in sampled frames. Full log is attached in Discord.

From [tboogie77 on Discord](https://discord.com/channels/1569849200047427645/1571562199007105064/1571567462602768506): Update from tboogie77: the black screen with audio/gameplay continuing still occurs on nightly-189-aa95d35f on a ROG Ally Z1 Extreme, with Frame Generation off and VRAM set to 3.6 GB. The attached log is for nightly-189; it reports 2048x1152 output, graphics scale 1, and 59.9–60 fps in sampled frames. Full log is attached in Discord.

BeanIssues referenced this issue

2026-11-15 16:56:52 +00:00

Game: #837

BeanIssues commented

2026-11-15 17:05:04 +00:00

Collaborator

From tboogie77 on Discord:

Additional version detail from tboogie77: the original black screen occurred on Longshot v3.0.0-rc.4 (Frame Generation off, VRAM setting 3.6 GB); it persisted on nightly-189-aa95d35f. The nightly-189 log is attached in Discord; it includes vk: submit 63809 failed (-2) and subsequent submit failures.

From [tboogie77 on Discord](https://discord.com/channels/1569849200047427645/1571562199007105064/1571562199007105064): Additional version detail from tboogie77: the original black screen occurred on Longshot v3.0.0-rc.4 (Frame Generation off, VRAM setting 3.6 GB); it persisted on nightly-189-aa95d35f. The nightly-189 log is attached in Discord; it includes `vk: submit 63809 failed (-2)` and subsequent submit failures.