AMD To Open Source Micro Engine Scheduler Firmware For Radeon GPUs 23
AMD plans to document and open source its Micro Engine Scheduler (MES) firmware for GPUs, giving users more control over Radeon graphics cards. From a report: It's part of a larger effort AMD confirmed earlier this week about making its GPUs more open source at both a software level in respect to the ROCm stack for GPU programming and a hardware level. Details were scarce with this initial announcement, and the only concrete thing it introduced was a GitHub tracker.
However, yesterday AMD divulged more details, specifying that one of the things it would be making open source was the MES firmware for Radeon GPUs. AMD says it will be publishing documentation for MES around the end of May, and will then release the source code some time afterward. For one George Hotz and his startup, Tiny Corp, this is great news. Throughout March, Hotz had agitated for AMD to make MES open source in order to fix issues he was experiencing with his RX 7900 XTX-powered AI server box. He had talked several times to AMD representatives, and even the company's CEO, Lisa Su.
However, yesterday AMD divulged more details, specifying that one of the things it would be making open source was the MES firmware for Radeon GPUs. AMD says it will be publishing documentation for MES around the end of May, and will then release the source code some time afterward. For one George Hotz and his startup, Tiny Corp, this is great news. Throughout March, Hotz had agitated for AMD to make MES open source in order to fix issues he was experiencing with his RX 7900 XTX-powered AI server box. He had talked several times to AMD representatives, and even the company's CEO, Lisa Su.
What language... (Score:4, Interesting)
... is GPU firmware written in? Some kind of assembler, C or something highly specialised like VHDL or Verilog?
Re: (Score:2)
Re: (Score:3)
Newton is the deadliest son of a bitch in space!
If he was on a square meter he would be Pascal.
Re: (Score:1)
Re: (Score:3)
Re: (Score:1)
I'm guessing it's C++, sigh 80% of the code being macros.... but don't worrry somebody will reprogram it in Rust sooner or later.
Re: (Score:2)
Probably C and assembly.
OT cloudflare sucks (Score:1)
Re:Yet more crying by people (Score:5, Informative)
ROCm supports the 7900XTX. AMD could have removed their consumer cards from the support list, but instead they open sourced some of the firmware.
The only downside I see here is that demand for the aging 7900XTX will stay higher.
Re: (Score:2)
The only downside I see here is that demand for the aging 7900XTX will stay higher.
And that down side is temporary. Part of what sells Nvidia cards is that consumers know there will be continuing software support. They are still supporting the 10xx cards, after all, and only dropped the earlier cards very recently. AMD needs to give their customers the same confidence, and the only way to do that is to support older cards.
Re: (Score:2)
Pretty sure AMD does support older cards. Not the ancient VLIW ones but at least as far back Vega.
Re: (Score:2)
I don't care what it was designed to do, I care about what it can do.
Re: (Score:1)
Thanks for the unwitting update! What we can tell from this inappropriately angry little rant is that you're a Nvidia corporate stooge and they're running scared today over this news.
Why do the work when you can get others to do it (Score:3)
Re:Why do the work when you can get others to do i (Score:4, Insightful)
Why should CUDA have a competitor? just add support for it.
https://github.com/vosen/ZLUDA [github.com]
Re: (Score:2)
Why should CUDA have a competitor?
Because open standards are nice.
Why should OpenCL have a competitor? You could point out that CUDA performs better, but that probably wouldn't be true if Nvidia had put that effort into OpenCL instead. So the obvious answer is: OpenCL should have a competitor because that makes Nvidia more money. Hooray.
Re: (Score:2)
I'd prefer OpenCL too. But CUDA is where we're at right now. Thanks to Oracle vs Google, we know APIs aren't copyrightable, to extending that to ABIs isn't such a stretch, so assimilating CUDA into an open something common and open should be doable.
What I'm definitely against is each vendor rolling its own: AMD ROCm, Intel OneAPI, nvidia CUDA.
I'm old enough to fuzzily remember the Direct3D, OpenGL, Glide mess of the 90s. Of which 2 survived, and we got Vulkan in the end...
Hard to imagine no perf issues... (Score:4)
There is a lot of work that goes into getting a CUDA app to run quickly, even when you know the underlying hardware limitations. Now you're trying to pretend you're an NVIDIA card, but really are an AMD card?
No, I don't know enough of the internals to be 100% sure, and yeah... it's great to be able to get something to run at all. I'd just be surprised if this turned out to be the final solution to accelerator interoperability.
And if it did start eating into profits they'd add DRM like hardware checks somewhere. NVIDIA has been spending a LOT of time and money on creating it, and they're nothing if not fiscally prudent and protective of their IP. Just like how Intel killed off any dream of an x86 compatible CPU from NVIDIA (was the original target for Denver, once upon a time).
Re: (Score:2)
> Now you're trying to pretend you're an NVIDIA card, but really are an AMD card?
Graphics drivers do that to games quite often, why not with compute?
Re: (Score:2)
They've had a hard time trying to come up with their own competition to CUDA. Seems like their just trying to outsource it.
They should have done a long time ago. AMD, and ATI before them, has always been bad at software. I love AMD CPUs, but their chipset driver package has often been wonky, and I have regretted every AMD GPU purchase ever because of the software. These days the Linux OSS driver is very good, IF you have a properly supported card, meaning not a spanking new one.
Release a 36gb VRAM card priced like a 4090 or cheaper and people may be inspired enough to do so.
AMD cards have been generally inferior until recently and yet the OSS driver has still been developed. Now they are only really inferior from a closed s