ZLUDA v2 ... opensource CUDA driver for Intel products.. AMD.. what about you? hehe

Was watching “Hot News” channel on youtube today and they mentioned ZLUDA… a CUDA wrapper that INtel is trying to use for their integrated GPU’s…

Running CUDA code on non-Nvidia is being worked on by INTEL? nice… Thoughts? (outside of lawsuit and closure… though it is open source… hmmm)

https://www.phoronix.com/scan.php?page=news_item&px=ZLUDA-v2-Released

LoL the impossible is possible :open_mouth:

1 Like

OTOY had something like this a few years ago for AMD graphic cards.

It is interpereter or emulator. Can not works fast.

One of many interesting and original open-source projects to be started in 2020 was ZLUDA, an open-spurce drop-in CUDA implementation for Intel graphics. ZLUDA - developed independent of Intel and NVIDIA - is built atop Intel’s oneAPI Level Zero interface (hence the name, ZLUDA) and allows for unmodified CUDA applications to run on Intel UHD/Xe Graphics hardware with near-native performance. Well, that’s the goal at least but with the initial ZLUDA release were a number of support limitations.

I tested it w Blender. It shows up as a Cuda device but it cant render anything given its implemented instruction set is probably not a match for Cycles.

1 Like

I propose to name it BarraCUDA.

4 Likes

Priceless response :laughing:

1 Like

Just got updated publicly for AMD. I tried to render Barbershop with CUDA on my 7900 xtx and it was just slightly slower than HIP.

1 Like

Hope they’ll fund the project again

with HIP RT enabled?

wow, thats impressive
You would have expect to any kind of translation layer to add a little overhead, and even in their examples there are cases where its faster than native OpenCL

edit: Oh, they are also working on OptiX support

1 Like

In Barbershop though, HIP RT is faster for about 5 seconds.

Also, the lights si significantly more pronounced with HIP RT (left) than with ZLUDA (right).


Edit.: I rendered the scene in CPU only mode and it looked like HIP RT, not like ZLUDA. I take it CPU always gets the same visual results, unlike GPU based solution, correct? Meaning HIP gets much closer to proper visual representation of CPU based path tracing, it seems.

Blender scores were following. Note that benchmark app verification process crashed when I tried to upload ZLUDA results. So pics only.

ZLUDA

HIP RT

4 Likes