Possible edge case issue with grass rendering and RTX 4090

john doe shared this bug 56 days ago
Need More Information

High grass densities in Space Engineers appear to trigger DXGI_ERROR_DEVICE_REMOVED on RTX 4090 GPUs. Despite ample headroom (cool temperatures, moderate power draw, and unused VRAM), the Ada Lovelace architecture can quickly process massive amounts of geometry, revealing potential race conditions or buffer mismanagement within the VRage engine. When Windows TDR (code 141) detects a stall, it resets the GPU. Notably, disabling grass density prevents these stalls, indicating that aspects of the grass rendering pipeline—especially instancing, LOD transitions, or resource updates—cause short GPU hangs not observed on less powerful hardware.

Since setting grass density to zero bypasses the issue, this effectively sidesteps the problematic rendering path for geometry loading, averting the stall and subsequent crash.

A recurring theme in user and community support threads is that such errors disproportionately affect newer, high-throughput GPUs—particularly those based on Ada Lovelace (AD102, AD103, etc.) or potentially late Ampere cards under heavy loads.

Moreover, the AD102/3/4/6/7 series differs in how they schedule work across Streaming Multiprocessors (SMs), especially in parallel graphics and compute tasks. These architectural nuances likely contribute to the problem.

All of this leads me to believe that the root cause of this specific edge case could involve:

  1. Race conditions
  2. Buffer mismanagement

The high-level logic chain is:

“More grass geometry” → “More complex resource usage” → “Increased risk of race/buffer flaws” → “Device removal” and TDR events

Although significant anecdotal evidence (e.g., support threads, Reddit, Steam forums) suggests that faster, modern GPUs are more prone to this crash, non-hardware factors or general engine edge cases may also play a role.


Whatever the exact cause, it merits further investigation if time and resources allow. I understand the team’s capacity may be limited due to ongoing Space Engineers 2 development. Still, given Space Engineers’ enduring popularity and its importance to Keen Software House, resolving this or releasing any stability improvements for high-end GPUs would be greatly appreciated by the community. Although I would be elated to see an update on this particular issue, I don't expect you guys to post updates here. Regardless, thank you for your hard work, and best of luck!

Replies (4)

photo
1

If anyone is having a similar problem, try setting your grass density to 0 and see how it works for you.


What do I mean by similar?

  • Game CTD with message "your gpu may be failing or overheating"
  • In SE logs, you see DXGI_ERROR_DEVICE_REMOVED
  • In SE logs, you see 0x887A0005
  • Confirmed that your driver is EITHER stable version. OR
  • Confirmed that your driver is the most recent stable version
  • Checked frequently during your play time that your GPU hotspots are within nominal range(For my 4090, hotspots~43c, junction~40c
  • In windows reliability monitor, you see a event logged at the precise time your game CTD with EITHER
  • LiveKernelEventCode 141 OR
  • LiveKernelEventCode 117(because failed reset attempt by TDR can also show this)

photo
1

I tried turning off Low Latency Mode in Nvidia control panel for this game, and that seemed to allow me to have grasses on

photo
1

Hello Engineer,

thank you for reaching our forum with this problem and for provided information.

Do I understand it correctly, that it is now actually working for you without any further problems?

I made sure to write down the ideas for possible workarounds that you wrote here, to try to help other players in the future.

Please let me know if there is still any more help needed from our side.

Thank you in advance.

Kind Regards

Keen Software House: QA Department

photo
1

It is technically "working".

If you define "working" as in playable but not being able to use one of the native features, sure.

However, if you define "working" as in playable with all the intended functions of the game enabled, then no it is not working.

photo
1

Bonjour j'ai peut être la solution qui ne nécessite aucune contrepartie au niveau des graphismes

il suffit de rajouter une clé registre TdrDelay : 10

  1. Accédez simplement à votre registre : Démarrer --> Exécuter --> Tapez "Regedit".
  2. Accédez à "HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\GraphicsDrivers " et créez une clé de type "DWORD (32 bits)" et nommez " TdrDelay " avec une valeur de * 10 . * comme valeur décimale.
    ça évite que Windows redémarre le pilote graphique ou quelque chose comme ca
    ca l'air de marcher pour le moment mais si mes crash revienne je le signalerai en attendant j'espère que ca pourra aider d'autre personne

photo
1

ne marche pas. ce que j'ai remarquer c'est que quand je joue a space.e la Vram de ma carte graphique ne se libère pas elle monte en continu j'jusqu'au point de rupture qui mène a ce code d'erreur, le jeu doit surement accumuler les texture et ne les décharge pas ce qui fait que quand les graphisme sont élevé ca crash beaucoup plus vite

photo
photo
1

Hello Engineer,

since we are now investigating this issue further internally, do you think it would be possible to share us a game log from the session where you experienced this problem, please?

You can access your log files by typing %appdata% into your Windows search bar and you will be redirected to the hidden Roaming folder. After that just follow: \Roaming\SpaceEngineers.

It will help us in further investigation.

Thank you.

Kind Regards

Keen Software House: QA Department

Leave a Comment
 
Attach a file