this post was submitted on 08 Aug 2024
62 points (94.3% liked)

Linux

5218 readers
15 users here now

A community for everything relating to the linux operating system

Also check out [email protected]

Original icon base courtesy of [email protected] and The GIMP

founded 1 year ago
MODERATORS
[email protected]
[email protected]
[email protected]
62
AMD takes down NVIDIA-CUDA replacement for Intel GPUs. What? (www.youtube.com)
submitted 3 months ago by [email protected] to c/[email protected]
13 comments fedilink hide all child comments
 

The developer wants to continue the project from the codebase before they used the AMD ROCM code.

Previously, it allowed to run CUDA apps on Intel GPUs (integrated I think) and beat OpenCL in performance.

https://github.com/vosen/ZLUDA

you are viewing a single comment's thread
view the rest of the comments
[–] [email protected] 22 points 3 months ago (2 children)

The project started as an independently developed ZLUDA, meant for Intel GPUs. AMD started funding development of the project, expanding its scope to also work on AMD GPUs, with the condition that if AMD abandons it, the code would become open source. This happened a few months ago, but now for some reason AMD backtracked and pulled the open-source code for reasons we don't know. AMD is usually pretty good with this stuff though, so I'm interested to find out what their motives are.

[–] [email protected] 19 points 3 months ago* (last edited 3 months ago)

I'm familiar with the timeline of events, it's just that this post title is intentionally obscure. As for why, it could be due to a recent enforcement by NVIDIA of a specific part in their EULA

Nvidia has banned running CUDA-based software on other hardware platforms using translation layers in its licensing terms listed online since 2021, but the warning previously wasn't included in the documentation placed on a host system during the installation process. This language has been added to the EULA that's included when installing CUDA 11.6 and newer versions.

The restriction appears to be designed to prevent initiatives like ZLUDA, which both Intel and AMD have recently participated, and, perhaps more critically, some Chinese GPU makers from utilizing CUDA code with translation layers.

My guess is that either AMD were directly contacted or preemptively closed off a legal loophole.

E: I was wondering where I recognised this username from

[–] [email protected] 2 points 3 months ago* (last edited 3 months ago) (1 children)

Probably, AMD was only doing open source because Nvidia wasn't. Now that it is no longer as advantageous since Nvidia is open sourcing (still no nvcc !!!), it is less advantageous. It's typical capitalism conservative terminal brain rotting cancer. They don't do the right thing unless it is profitable to do so. There is no big picture or ethics; just criminals and quarterly reports.

[–] [email protected] 13 points 3 months ago (1 children)
[–] [email protected] 1 points 3 months ago (1 children)

What's stopping Windows from banning WINE if this is the case?

[–] [email protected] 1 points 3 months ago* (last edited 3 months ago)

I suppose if zluda interpreting native cuda code on other IHV platforms is "illegal", whereas HIP can legally be used to translate CUDA to ROCm at runtime, I would equate WINE to HIP, as it translates Windows API calls to POSIX for linux and Unix systems.

I'm not sure Microsoft have a leg to stand on with a move like this given their legal troubles in the past around becoming a monopoly. Maybe they don't see as much threat given the low share of linux and unix (including mac) desktops

There's still a bit of friction around actually leveraging WINE on non-Windows systems, with Proton being a relatively recent outlier as it offers you a mostly plug and play experience via steam, lutris and others.

For nvidia, zluda can allow other IHVs with significantly cheaper DC solutions to undercut them with no friction whatsoever.