AMD Publishes Video To Explain The Radeon Open Compute Stack (ROCm)

[darkbasic]

ROCm doesn't need a video, it needs more **polishing**.

[polarathene]

The video looks like they had all the right ingredients, except for whomever put the visuals/animations together.. that was open-source/voluntary grade, this was something paid for by AMD right? It's not engaging and really difficult to maintain interest until the end(even though it's only a little over 2 mins), the graphics could have helped avoid that if done well, seems like they might have been handled in parallel to voice over recording with a rough sync up on some parts and the looping graphics that aren't serving much value.

If it's meant to attract customers / interest, I'd be surprised if that works out vs some slides/presentation or landing page.

ROCm is a terrible name, marketing-wise. It just doesn't roll.

Rock'em, rolls fine.. but still might not work as well, also the lowercase m makes it a little annoying to type. Brightside is it should work fine as a keyword as long as it's not auto-corrected as a typo (eg on Google that sometimes results in different search results of what it thinks you meant).

Any sign of them supporting all their cards with this? I think I've missed out on 4 generations of "compute support" with AMD, it worked partially (though badly) with Clover on my 6770M, my M395X is just about EOL without ever having compute support, the Ryzen 3400G doesn't work with it and I'm pretty sure the 5700M in the latest laptops is NAVI and not supported either

Navi is coming iirc, but their main priority atm is focused on the bigger customers of ROCm who aren't using Navi GPUs just yet? (as in they're needing support for large volume of Vega GPUs) Something like that... Older GPUs aren't likely to see support, Bridgman mentioned that in Phoronix forums in the past month or so, one of the older ones almost made the cut, but lacked some spec which'd require more work to support (which'd be hard to justify prioritizing), while everything else before lacks hardware requirements to work with ROCm properly.

Hopefully, once Navi gets support, the next GPUs due out this year(Big Navi? I don't really follow AMD GPU news much), will not have as big of a lag for receiving ROCm support. I assume individuals with their newest GPUs on Linux wanting ROCm is a small market, enterprise demands would probably have more influence on the ROCm roadmap?

[polarathene]

bridgman mentioned that there will be Navi support for ROCm. But why bother, it's unusable for consumer/enduser purposes.

Unstable in what way? (nvidia user here, patiently waiting on decent AMD product with good Linux support and compute capabilities, without the string of issues that each GPU release from AMD seems to be plagued with for 1-2 years).

It's a bit of a dilemma with AMD GPUs, they often seem to have gotchas that take quite a while to fix (not that I'm complaining, I know we're minorities, so I understand), by the time they're in a good state, if you're in the market for a new GPU, it's usually something newer(especially when certain hardware features/support comes into play). Those new Renoir APUs for laptops seem attractive, but I remember the 2200G/2400G, both had problems, with one taking 18 months from release to have a stable experience(on rolling distro release packages). Other AMD stuff I recall driver issues with GPU resets that'd take quite a while to see resolved, features from hardware that weren't usable for a long time(maybe different story with proprietary driver).

Whatever issue the latest GPUs are having, hopefully they'll be resolved and if Big Navi isn't as different as Vega to Navi hardware was, then maybe a smooth release will happen for once and the only ones that might complain would be Ubuntu LTS users?(or similar)

[sp82]

Still Blender (OpenCL2.0+) not working and now also Luxrender(OpenCL1.2) not working anymore after the last release of rocm. This thing is so broken and not conformant that only libraries and software written by AMD works on it. So I suppose its open but its not friendly.

[vegabook]

Still Blender (OpenCL2.0+) not working and now also Luxrender(OpenCL1.2) not working anymore after the last release of rocm. This thing is so broken and not conformant that only libraries and software written by AMD works on it. So I suppose its open but its not friendly.

Probably only way efficiently to use cross-platform compute anytime soon will be WebGPU and its "native" equivalent wgpu. I'm not bothering with ROCm at all. It's an utter mess.

[vein]

I do not understand why people have sop much problems with these libraries. I have been using them for OpenCL and now OpenMP with my Fiji card for at least...2 years on Arch and AUR. I have had relatively few problems. Few hicks when upgrading to new versions, that have sorted themselves out pretty quickly.

Still Blender (OpenCL2.0+) not working and now also Luxrender(OpenCL1.2) not working anymore after the last release of rocm. This thing is so broken and not conformant that only libraries and software written by AMD works on it. So I suppose its open but its not friendly.

Hm...strange, I got no problems with Luxrenderer. Haven't tried blender.

[pal666]

Sadly on the GPU Compute side CUDA is literally the only option.

proprietary shit is never an option

I'm guessing atm Radeon Group don't see much chance to eat market share here from nVidia so they basically kinda glue togheter ROCm to have something marketing wise but not much else.

i'm guessing you are clueless

[pal666]

(nvidia user here, patiently waiting on decent AMD product with good Linux support and compute capabilities, without the string of issues that each GPU release from AMD seems to be plagued with for 1-2 years).

i'm not aware of any issues with my rx580(but i don't use compute)

Whatever issue the latest GPUs are having, hopefully they'll be resolved and if Big Navi isn't as different as Vega to Navi hardware was, then maybe a smooth release will happen for once and the only ones that might complain would be Ubuntu LTS users?(or similar)

indeed. but as usual, i'll wait for field reports before buying

[peterdk]

I did not know that Tenserflow had AMD support. Is it any good? Now using NVIDIA cards, because they work fine.

[pal666]

Probably only way efficiently to use cross-platform compute anytime soon will be WebGPU

it's a wrapper, which still requires underlying implementation. efficient and cross-platform compute is called vulkan