> Some of the newly added features may come as a surprise to those of you who keep a close track of ZLUDA development. Most of them were previously explicitly outside ZLUDA's roadmap. There has been a change of plans. ZLUDA development is no longer commercially funded, so it's back to being my weekend project. This means that the priority is no longer what makes commercial sense, but what I find the most entertaining. That's why the sudden addition of textures, PhysX and better Windows support.
Outside of the time commitment that full-time support would offer, I generally think focusing on amusement is a good life strategy.
Agreed. I think the real value of Zluda to the community is going to lie on more niche applications. The ones that are more commercially interesting are likely to have other solutions developed for them.
The 32 bit PhysX support is particularly interesting as Nvidia had wanted to remove support for that particular combo on the 5000 series. They eventually flipped back to supporting it due to the pushback, but there was a short period where people were wondering if the same kind of 3rd party translation layer would be needed for Nvidia's own GPUs in this specific scenario.
Iirc he was kind-of funded by AMD for a while to break the CUDA LLM moat (the mention of change of direction at the bottom of the article), no idea if it was legal issues with him being paid by them or why it changed, but..
1: Much initial research and early OSS was CUDA focused due to libaries and momentum, but with more Mac users,etc kernels/frameworks are more portable so with stuff like Unsloth so the CUDA moat is falling there (If it's Vulkan or more propietary backends doesn't matter, Vulkan or OpenCL are probably fairly equal in usefulness since Vulkan support would probably require custom extensions for full performance).
2: As alluded above, ZLuda seems to have focused a bit on ROCm API's for AMD (I think I've seen mentions of other library supports but it's been up and down iirc since he was employed by AMD). The main benefit compared to Vulkan iirc is that ROCm allows for some CUDA-like feature that makes kernel emulation so much less troublesome (or even feasible at all?).
I appreciate the focus on amusement in:
> Some of the newly added features may come as a surprise to those of you who keep a close track of ZLUDA development. Most of them were previously explicitly outside ZLUDA's roadmap. There has been a change of plans. ZLUDA development is no longer commercially funded, so it's back to being my weekend project. This means that the priority is no longer what makes commercial sense, but what I find the most entertaining. That's why the sudden addition of textures, PhysX and better Windows support.
Outside of the time commitment that full-time support would offer, I generally think focusing on amusement is a good life strategy.
Agreed. I think the real value of Zluda to the community is going to lie on more niche applications. The ones that are more commercially interesting are likely to have other solutions developed for them.
The 32 bit PhysX support is particularly interesting as Nvidia had wanted to remove support for that particular combo on the 5000 series. They eventually flipped back to supporting it due to the pushback, but there was a short period where people were wondering if the same kind of 3rd party translation layer would be needed for Nvidia's own GPUs in this specific scenario.
Has anyone tried this for LLMs and how does it compare to Vulkan?
Iirc he was kind-of funded by AMD for a while to break the CUDA LLM moat (the mention of change of direction at the bottom of the article), no idea if it was legal issues with him being paid by them or why it changed, but..
1: Much initial research and early OSS was CUDA focused due to libaries and momentum, but with more Mac users,etc kernels/frameworks are more portable so with stuff like Unsloth so the CUDA moat is falling there (If it's Vulkan or more propietary backends doesn't matter, Vulkan or OpenCL are probably fairly equal in usefulness since Vulkan support would probably require custom extensions for full performance).
2: As alluded above, ZLuda seems to have focused a bit on ROCm API's for AMD (I think I've seen mentions of other library supports but it's been up and down iirc since he was employed by AMD). The main benefit compared to Vulkan iirc is that ROCm allows for some CUDA-like feature that makes kernel emulation so much less troublesome (or even feasible at all?).
Do other models - VLMs, Diffusion, etc., also work fine with Vulcan?
Useful context https://www.tomshardware.com/pc-components/gpu-drivers/cuda-...
Nice pun
Złuda means "mirage"/"illusion"/"figment of imagination"
Even funnier that CUDA in polish means “miracles”