Last patch made me start Out of memory crashing every ~90+ minutes

You can nuke the bundled_ shader_cache.hans and play without it to see if it makes any difference, the game is capable of running shaderless hansless just fine :slight_smile:

However i saw 0 improvements playing like that, 0 impact on the existing issues and i don’t think shaders or the PSO files corruption has any impact on performance outside busted visuals when playing on High textures!

Furthermore deleting shader cache like that will very likely corrupt the PSO library and will take few session for it to rebuild properly, which again doesn’t matter performance wise or for people that play on Medium, but it bears mentioning that deleting the shaders causes its own set of problems

You mean stop the game re-creating that file? What is best way of doing that. Editing file somewhere? I’d like to give that a go out of curiosity.

X:\Steam\steamapps\common\Warhammer 40,000 DARKTIDE\shader_cache

temporarily move the bundled_shader_cache.hans away from the folder.

In the:

X:\Users\XXXX\AppData\Roaming\Fatshark\Darktide

remove shader_cache.hans

The game should start with error telling you that shader creator is busted, just click on Ok and it will continue without creating a new shader_cache.hans!

Or

X:\Steam\steamapps\common\Warhammer 40,000 DARKTIDE\shader_cache

temporarily move the bundled_shader_cache.hans and darktide_shader_cache_builder away from the folder.

In the:

X:\Users\XXXX\AppData\Roaming\Fatshark\Darktide

remove shader_cache.hans

Two ways to do it, both result in shader_cache.hans not being recreated but the end results might differ a little. It seems that if both bundled_shader_cache.hans and darktide_shader_cache_builder are moved, the game just doesn’t seems to notice at all on my end, like if the shaders never existed at all and there wasnt any need for them!

*Strongly suggesting to backup both PSO files from the AppData folder, not critical but it will spare you an extra session rebuilding them

Btw i’m at the end of my rope for real with this game, there wasn’t a goddamn patch for weeks and suddenly this BS gets just conjured out of nowhere :confused:

That is strange. The game didn’t need it, just done a quick mission test. No bugs/errors or anything and ran smooth.

Yes it is strange, nonetheless it tells us something about the efficacy of clearing shaders and the placebo effect that goes with it.

Maybe it does fallback on DXcache…

i turned down my texture settings from high to medium like someone suggested in this or another post, the issue is completly gone now :smiley:

Yaaay

I have 32GB of RAM and have never seen a out of memory error.

It seems to me that people who see this are either playing with mods, or fools who keep disabling the page file.

yo its a gpu memory issue

Or try to play with high textures on an 8 GB card…

There is some logging at the end of the console log that some people believe is evidence of a massive GPU memory leak but it’s not. The game is not releasing memory just prior to quitting which is totally normal because the OS will take care of that as soon as the process ends.

It’s great that the devs found a leak and are fixing it but there’s zero evidence as to what type or the severity.

I really hope they do a dev post on it, I’m very curious.

Unbelievable…

In the NVIDIA CUDA ecosystem, cudaMalloc is the standard runtime API used to allocate memory directly on the GPU device

In CUDA, memory allocated using cudaMalloc is not automatically managed and remains reserved on the device until you explicitly release it or the application terminates.

The CUDA runtime automatically wipes the context and frees all device allocations when the owner CPU process terminates

You are literally arguing that there isn’t evidence of GPU memory leak because there is a textobook evidence of a GPU memory leak!

00:26:08.197 warning: [Profiler] [main] Stall 5040 ms on frame 265227: PatchedResourcePackage::flush 5040 ms PatchedResourcePackage::finish_unloading 5040 ms PatchedResourcePackage::start_unloading_resources 5040 ms cleanup 5040 ms 00:26:08.319 warning: [Profiler] [renderer] Stall 4800 ms on frame 265227: ThreadPool::do_work_while_waiting_for_event 4800 ms

The OS released, the Malloc is done and its now waiting for cudaMalloc to do the work.

One more time, THE OS RELEASED and its waiting for the GPU do to as well, the log literally tells you that. So tell me again how is the OS taking care of that when you can see its waiting for the separate vendor memory controller for release, huh?

Please just stop being this clueless

The game is not using cudaMalloc… That’s not even related. And your quotes dosn’t even support you. It clearly says it will be released when the application terminates.

Again, you don’t know what you are talking about.

Wtf? This has nothing to do with the game, cudaMalloc is the GPU memory controller and you wont be doing anything without it

cudaMalloc is a core function in the CUDA Runtime API used to allocate linear memory on the GPU

Did you just said that Darktide doesn’t use cudaMalloc? You don’t have a clue do you?

By a cuda runtime, are you missing that intentionally just to troll?

The CUDA runtime automatically wipes the context and frees all device allocations when the owner CPU process terminates

The cuda is doing the release on the GPU, using the cudaMalloc controller after getting instruction from Malloc!

To get your data from the CPU to the GPU, you must first use malloc for the host, then cudaMalloc for the device

You have two separate memory controllers bridged by a cudaMemcpy bridge to communicate with each other.

Are you still gonna insist that its the OS that releases the GPU memory even thou its two separate controllers that need a bridge to communicate?

And when you see the OS being done releasing (Malloc) and waiting for a separate GPU vendor controller stall to finish (cudaMalloc), how is that not an evidence of GPU memory leak? When it literally tells you it stall on renderer and profiler?

Once again, stop being clueless and atleast google it, it will help you avoid saying clown stuff like this:

The game is not using cudaMalloc

I mean just google the stupidity you just uttered and tell me again how i don’t know what i’m talking about, you troll clown!

LOL, you clearly don’t even know what the CUDA Runtime API is or what it’s used for.

I bought a brand new PC… 4080 and 9800X3D

Still getting Out of Memory crashes, even after this weeks patch T_T

Did not work for me