i tested in Forever Winter which blows past 8gb vram in ashen mesa and they very much did improve it drastically. I would tank to 15fps before but now can at least maintain a stuttery 60fps
just briefly tested spider-man. still seems to behave a little weirdly but is improved. i let vram reach 8.0gib with very high textures, but at this point firefox wasn’t able to open and initialise correctly
It’s definitely improved, specifically the overall system stability/feel while vram is filled, but personally I am still getting kingdom come deliverance 2 filling up vram and then the output from the game freezes on a frame, I can still hear it in the background and interact with menus and hear it responding, but the actual display dies and it never comes back, I’m starting to wonder if my specific issue was partly this but partly also just an issue with dx12/vkd3d, but hey any attention nvidia is paying to this area is still good
That was the reason I created the GreenBoost kernel modules: to orchestrate memory between T1 GPU VRAM and T2 system RAM while keeping all tensor computations on the tensor cores (with no CPU spillover).
Today I’ve just released v2.8 after deleting NVidia name from the project (since this is not an official NVidia project), due law I needed to do the change
Last Friday got a bicycle accident, I had not in mind release the new version yet, hence vulkan layer + greenboost proton is at an early beta stage
I think it would be interesting now that it just came out, so just food for thought, as they’re still in an immature state, at least to mention the new improved VRAM management dmem cgroup patches for Mesa. If there are any plans for NVIDIA to support or implement something similar.
What exactly someone thought was wrong with @opisalwaysgreat’s question? Seemed perfectly legit to me: some update from NV how things are going regarding this issue would be most appreciated.
CC: @ahuillet
It was rude. I’m surprised it was up for as long as it was. I’ve had comments removed / hidden here for far less.