Since Update 1.2, the game has performed very poorly, often dropping to 30 FPS on the first subsequent map load, and down to 8-22 FPS upon reaching an in-map zone transition (such as entering a Tomb of the Erased). At this point, the game essentially becomes unplayable, with stuttering of movement and significant problems of controlling the character.
I am playing the game on Debian 12, using a RX 6600 and the open source drivers, which has 8 GB of VRAM. Steam Proton version is 9.0-4.
Using radeontop, it’s possible to see that, as soon as the game loads, it’s using an upwards of 6-8 GB of VRAM. As soon as the next map is loaded, this causes the game to exhaust video memory, resulting in severely degraded performance:
These readings were taken after logging into “The End of Time” and manually walking into the “Monolith of Fate.” Other than walking to the “The Fall of the Outcasts” waypoint, no further actions were taken nor skills used. All settings are at or below Medium, yet per the graphs above, it’s clear the game is suddenly overtaxing the video card.
Note: to clarify, these issues were not present as of 9:30 PM (EDT) on April 15th, before the servers were taken offline for patching. It appears to be exclusive to changes made via Upadte 1.2.
I’m on nobara linux (fedora 41), latest nvidia drivers, rtx 3080 10gb: same results. Vram gets filled very fast (around 30 minutes of gameplay). Just like the OP I had no issues in the past.
Changing graphical settings brings no benefits, I’m waiting for a fix to play, having to restart the game every 20 minutes it’s very frutrating.
Those launch parameter can be used on linux as well but are made for unreal, I’ll try and if it works I’ll post.
I usually create a dxvk.conf file with only 2 lines, I leave it in the same directory of the .exe and proton automatically takes the parameters:
dxgi.maxDeviceMemory=8192
dxgi.maxSharedMemory=8192
In the rare event a vram leak happens, this saves the day; Too bad that with LE proton refuses to load the .conf file for some reason.
Anyway It’s a problem that also affects windows users, one that hopefully gets fixed.
Ok so it’s not just me, performance has tanked recently for me too. VRAM fills and game stutters like crazy. Using nvidia 4070 and 560 drivers, tried various different drivers and versions of proton, nothing seems to help so far.
Same issue here on Mint with RTX3070 (550 drivers). I’ve tried several proton (inc GE), lowering all settings, several launch options. Volatile GPU-Utils shows 100%.
Did some digging into this, and it looks like there was an update to the Mesa drivers (which are the open source ones you get from the Linux kernel) years back (~2021?) which caused excess VRAM allocation to spill over into GTT, the second bar in the last block.
However, this would represent off-device, shared system memory, which runs at a far lower clock (and over PCIe, as a result) than the card’s GDDR5/6 RAM (mine is only 3200MHz, which is a fraction of the 224 GB/s bandwidth of my card).
This could explain the abysmal frame rates if we assume two things based upon what I’ve observed:
Approximately 8000 MB of VRAM is allocated around the time Last Epoch launches, presumably in pre-caching a lot of graphics data and textures.
Provided the first map you load into does not cause an overage, then the immediate next map (followed by each subsequent map transition), will spill over into GTT space (the 569 MB notated in the image above).
Note that now if a buffer is moved to GTT, it will stay in GTT even if VRAM is empty. Besides the low PCIe throughput (only ~13 GB/s vs VRAM with 512 GB/s), you also lose the infinity cache, which only caches VRAM.
If you didn’t use VM_ALWAYS_VALID, you would at least get significantly throttled buffer movements.
That could explain why the game suddenly bogs down once a map transition causes allocations to spill over into GTT space.
It seems the Unity allocator is failing to release unneeded resources between map transitions.
However, since this occurred only after Update 1.2, I would suspect the latter case and the result of a memory leak, since this behavior was not present prior to the patch going live.
For me the 1.2 patch worked fine. This only started happening to me after the first hotfix (1.2.0.x).
CachyOS
Intel(R) Core™ i9-14900K
GeForce RTX 4070 Ti SUPER
64Gb RAM
After looking into the DXVK configuration options, I was able to set it up via the following launch option:
DXVK_CONFIG_FILE="~/.dxvk-conf" %command%
One parameter in particular resulted in unexpected behavior, causing the game to fail to launch completely (it would get stuck before the window could load).
If enabled, pipeline libraries will be freed aggressively in order
save memory and address space. Has no effect if graphics pipeline
libraries are not supported or disabled.
While this kept the VRAM usage below 3 GB allocation, the game was unable to launch completely, presumably because pipelines were being freed after use when they were relied upon later in the program’s execution.
When enabling this configuration on a different game (tested with Timberborn), the game was still able to launch and function as expected.
This leads me to believe that Unity is creating excess pipelines or related artifacts which are not being cleaned up and exhausting video memory.
I managed to solve the FPS drop and stuttering problem by adding the command LD_PRELOAD=“” PROTON_ENABLE_NVAPI=1 PROTON_USE_WINED3D=1 %command% in Steam, but I created another problem. (without PROTON_ENABLE_NVAPI=1 game not start)
Now the game is closing when I enter a map in the monolith or when I open the interface to increase skill points.
Tested with Proton-Experimental, Proton-Ge-9.27 and Native.
I’m using CachyOS
Nvidia driver updated to version 570.144-2.
PS: Apparently adding -force-dxd11 at the end solves the crash problem.
Last Epoch refuses to load my dxvk.conf file, no matter where I place it or what I do.
It never happened to me before, are you sure your .conf file is actually being loaded?
If yes, try this parameters according to your VRAM:
There is a problem anyway: using the low settings the game doesnt even get to 8gb of vram, but after some time it starts to stutter and becomes unplayable. Notice: vram doesn’t get filled, so it must be something else. Windows is also affected by this, according to the official discord.
I got back to the game now with season 2 after a while, and I can confirm there still is either a memory leak present or problems described as above. Playing on Fedora 42, tried several Proton versions from Experimental to GE-27 and the latest cachyos version, after about 1-2 hours of play, the game massively drops in fps to unplayable levels of about 10fps.
I know, my AMD RX480 with 8GB VRAM is… in dire need of an upgrade - but that’s not the point.
Arch Linux with Radeon 7900 GRE using Proton Experimental or “auto” compatibility layer (tried both) - I’m having the same issues described by others above i.e. massive (and permanent) frame drops after an hour or so of gameplay. The only thing that fixes it is to restart the game.
I was watching VRAM usage to see if that was the issue, it started happening each time around 10GB of VRAM usage but that’s nowhere near the 16GB available on my GPU so I don’t know if that’s a factor or not. Nothing else in “radeontop” output suggested any resource getting anywhere near 100% usage.
Weird thing is that when the issue has been triggered, and my character stands still, I get my capped 60 FPS but the moment my character starts moving or performing actions, that’s when FPS drops to ~15. My graphics settings are “Medium” preset at 3440x1440 resolution, vsync disabled.
I posted this image on the official discord as well, I believe that’s what you are talking about:
Note: VRAM is not full (it’s a 3080 10gb), settings are all low to test.
High settings: VRAM fills up, stuttering, unplayable.
Low Settings: VRAM not full but happens what you described.
Hi, yep that definitely sounds like my experience - I’ll install mangohud and check frame time graph like you’re showing in your screenshot when moving vs. not moving. Won’t be able to check until tomorrow unfortunately.
Other things I tried that didn’t work:
Lower resolution/s
Uncapped FPS (game graphics setting)
Proton stable 9.0.4 (previously only Proton Experimental and default compatibility setting)
I tried this and… so far it’s working. Even on Very High Vram doesn’t skyrocket. Didn’t use -force-dx11 (it crashes), I’ll keep testing and see how it goes.
edit after 3 hrs playing: It does work BUT what it does is forcing OpenGL, which insted of using the VRAM uses the RAM.
Under this conditions the vram usage is around 3 gb on Very High settings, the RAM usage goes to around 20gb.
Performances of course are impacted, but it’s a workaround till EHG fixes the issue.
Looks like a new version was pushed out 1.2.0.3, but I can’t find what changed or patch notes.
Still same issue tho.
Also tried the command line options posted above one at a time and all together. No change.
Also experiencing the same issue since the latest season patch. I’m on NVIDIA 1070 with only 8GB of VRAM. For me it helped changing all graphics related settings in le_graphicsmanager.ini[1] to Low from Medium and making sure to run it on 1920x1080 screen, 3440x1440 ran out of memory after a while.
Has some client crashes and one instance where nvidia driver crashed and caused all graphics on system to hang. It was otherwise responsive (not really helpful on a desktop system) though.
It’s something at least…
Sometimes it fails to launch though. It starts and uses 200% CPU (two full cores) forever (well… forever is unconfirmed…) in the background (no GUI).
This error shows up in kernel ring buffer when this happens:
kernel: [drm:nv_drm_gem_alloc_nvkms_memory_ioctl [nvidia_drm]] ERROR [nvidia-drm] [GPU ID 0x00000a00] Failed to allocate NVKMS memory for GEM object
