Radeon ROCm 3.5 Released With New Features But Still No Navi Support

[ms178]

It is not like ROCm have been targeted for consumer market at all. It so much P.I.T.A. to install it on any recent distro that most people who need compute just use nvidia instead.
And that on top of very narrow hardware support list. Will probably see usable Mesa implementation of OpenCL sooner than any good changes in ROCm.

I got at least the OpenCL part (which is the part most consumers are very much interested in) running on openSUSE Tumbleweed thanks to a blog post which mentioned all the needed packages and how to overcome other installation hurdles.

Let's see how performant the Mesa solution will be and if that turns out to be practical in the end. By the way, I am still waiting for OpenCL 1.2 support on my R600 notebook chip (and that is 9 years old hardware).

[oleid]

And they still don't have a build of amdgpu pro (aka Catalyst, aka fglrx, although it's no longer using the fglrx kernel module) for Ubuntu 20.04.

I guess the only thing amdgpu-pro might be good for nowadays is OpenCL. And that only barely. Its OpenGL implementation is hopelessly slow, the oss vulkan drivers are probably at least as fast as the proprietary implementation nowadays.

[oleid]

I got at least the OpenCL part (which is the part most consumers are very much interested in) running on openSUSE Tumbleweed thanks to a blog post which mentioned all the needed packages and how to overcome other installation hurdles.

Let's see how performant the Mesa solution will be and if that turns out to be practical in the end. By the way, I am still waiting for OpenCL 1.2 support on my R600 notebook chip (and that is 9 years old hardware).

Out of curiosity : what are you doing with OpenCL? I never really found any app which uses it. And nowadays, you can do your compute programming using vulkan as well.

[bug77]

Michael is so funny when he keeps saying/writing "surprisingly". You'd think he only recently started looking at AMD software.

[_r00t-]

One of my biggest mistakes in my life that i bought Radeon RX 5600 XT..... pitty

[bug77]

One of my biggest mistakes in my life that i bought Radeon RX 5600 XT..... pitty

It's a good card. Your biggest mistake is probably not doing your homework before buying

[bridgman]

It is not like ROCm have been targeted for consumer market at all. It so much P.I.T.A. to install it on any recent distro that most people who need compute just use nvidia instead.

What problems are you having installing on a recent distro ? We split the packages and dependency hierarchy to make it easy to install ROCm userspace on the distro's existing kernel and drivers, and have covered that in the install instructions for almost a year.

And that on top of very narrow hardware support list. Will probably see usable Mesa implementation of OpenCL sooner than any good changes in ROCm.

I'll double-check, but I thought the HW support list was all Polaris & Vega dGPUs with Navi in the pipe, ie pretty much every dGPU we have been selling in the last few years. I think it was even working on the KabyLake G parts (which had an AMD dGPU). We do need to get Navi support finished but getting the transition from HCC to hip-clang done (part of 3.5) will make that easier.

We have been making some changes to APU support (using GPUVM-based dGPU code paths rather than the original APU-specific ATC/IOMMUv2 paths) in order to let APUs be a better development platform for dGPUs and you should see that code start to fetch up soon as well.

[Laughing1]

Out of curiosity : what are you doing with OpenCL? I never really found any app which uses it. And nowadays, you can do your compute programming using vulkan as well.

https://en.wikipedia.org/wiki/List_of_OpenCL_applications

[TemplarGR]

One of my biggest mistakes in my life that i bought Radeon RX 5600 XT..... pitty

Give it to me then

[ms178]

Out of curiosity : what are you doing with OpenCL? I never really found any app which uses it. And nowadays, you can do your compute programming using vulkan as well.

Apart from clinfo, Luxmark and LibreOffice I haven't used it that much on my older Linux laptop to be honest. On my more performant hardware on Windows I use OpenCL often in video editing and image manipulation software and it is usually much faster than CPU-only solutions.