Lisa Su Reaffirms Commitment To Improving AMD ROCm Support, Engaging The Community

[phoronix]

Phoronix: Lisa Su Reaffirms Commitment To Improving AMD ROCm Support, Engaging The Community

AMD CEO Dr. Lisa Su has reaffirmed the company's commitment to the open-source ROCm compute stack and working with the community and ultimately improving their software support...

https://www.phoronix.com/news/Lisa-Su-ROCm-Commitment

[CochainComplex]

This is good news!

[raystriker]

AMD are already around a year behind the green team in their hardware availability cadence. For them to have *any* chance at competing with Nvidia seriously, they need to have ROCm work properly on each and every RDNA/XDNA/CDNA device under the sun at launch. They have no choice. Intel was asleep at the wheel, but Jenson was not.

[Danny3]

Yeah right!
When will it be fully open source and installed by default like Mesa?

[NeoMorpheus]

Thats good to hear!

All Hail Our Holy Lady Dr. Lisa Su! 😁

[peterdk]

Great, I have some faith In Lisa, so I hope the compute thing will get better for Linux.

Personally I would love to buy a AMD GPU, but the blender HW Raytracing acceleration isn't there yet and Nvidia is really good at it. But hope that the software side will get some serious attention now.

[brucethemoose]

George Hotz's video on his falling out with AMD was... freewheeling, to say the least. See for yourself:

https://m.youtube.com/watch?v=Mr0rWJhv9jU

It actually made me skeptical of tinygrad. It seemed like a cool project at first, and I really don't know anything about George, but what advantage is it going to have over big, corporate backed ML languages/compilers already targeting AMD? Or other VC backed community projects, like llama.cpp, that are already operating and ahead of the curve with seemingly more grounded devs?

[CTown]

George Hotz's video on his falling out with AMD was... freewheeling, to say the least. See for yourself:

https://m.youtube.com/watch?v=Mr0rWJhv9jU

It actually made me skeptical of tinygrad. It seemed like a cool project at first, and I really don't know anything about George, but what advantage is it going to have over big, corporate backed ML languages/compilers already targeting AMD? Or other VC backed community projects, like llama.cpp, that are already operating and ahead of the curve with seemingly more grounded devs?

He must have some strong faith in Rocm. Geroge Hotz has always liked for things to be open (he is the famous PS3 hacker afterall) and probably has an affinity for AMD. AMD must step up their game quickly. As the article tells us, Rocm's support for consumer hardware is a joke.

[Mahboi]

I typically dislike attacking character, but I've seen the Hotz meltdown on Youtube and it's the most childish thing I've seen in a long while.
Screaming for 10 good minutes and announcing that he was dropping AMD for good because the answer of his AMD contact was unacceptable to him...

This kind of childish tantrum was "justified" by a kernel panic caused by the AMD software, which, when he asked his AMD contact, said that he'd "go around the engineering team to capture mindshare about the problem". He screamed all this time as if it was a problem of "morality" and that "kernel panics are unacceptable"...

Someone should tell this child that there's probably a kernel panic every other day with kernel devs, and that fixing them, when you have not one but many, like AMD probably does, isn't about morality but about resources. Whatever problem he found was probably one of many, and saying "I'll capture mindshare about {this one}" is not a cause for throwing a tantrum like this.

Rolling back on your statement that you were giving up on AMD just a week after also reinforces the idea that he's immature.
Being immature doesn't make him a bad coder or anything, but running a company is an extremely long and painful process, especially for startups. After seeing him scream for 10 mins, claiming that he is dropping AMD, only to roll back a week later, all I can think is that he isn't going to be able to take in the load of launching an actual startup.

[M@yeulC]

ROCm was a pain to install. There is a PKGBUILD, but no stable releases, complex dependencies, and a lot of code to download. Code often didn't compile when building the package.

I ended up successfully installing it for my R9 Fury (gfx803) some time ago, but there has been some breakage since, and IIRC compute kernels just segfault. I've seen a few repositories trying to fix this, but I threw the towel before successfully fixing it.

How can I take AMD seriously? If I haven't been able to make their stack work on my consumer card after days of fiddling, I certainly won't recommend their stack at work. Consumer products are an entry point to the pro market. A lot of pytorch, etc, contributors only have access to consumer cards.

There is a serious lack of regression testing and interest in consumer hardware with the ROCm team. They should look at what Espressif built with the community. SDKs and runtime should be as easy to install as possible, and work on any consumer hardware (even if slowly).