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

[PuckPoltergeist]

I don't know about ROCm, but Mesa only needs llvm-libs, and you can definitely install those in parallel. I currently have 10 and 11 installed (though my package for 11 does not seem to have as many files as the one for 10, so maybe it's not that easy).

And ROCm use it's own version of llvm. If an application is using mesa (for OpenGL/Vulkan) and ROCm (for OpenCL) like Blender does, both versions of llvm are linked in. This doesn't work and the application crash at start.

[geearf]

And ROCm use it's own version of llvm. If an application is using mesa (for OpenGL/Vulkan) and ROCm (for OpenCL) like Blender does, both versions of llvm are linked in. This doesn't work and the application crash at start.

That used to be a problem with RPCS3 but not anymore, I am not sure what they did to workaround that, but it seems feasible.

[xception]

That used to be a problem with RPCS3 but not anymore, I am not sure what they did to workaround that, but it seems feasible.

Actually it's still a problem for me (using blender), if it works for you please let me know as much information as possible to get it working for me as well... like ROCm version, mesa version, llvm libs and versions

[geearf]

Actually it's still a problem for me (using blender), if it works for you please let me know as much information as possible to get it working for me as well... like ROCm version, mesa version, llvm libs and versions

That is not what I meant.
RPCS3 has its own llvm that used to not work alongside Mesa's llvm, but after the RPCS3 devs did some changes, I forgot which, it worked fine.
The previous workaround was to use RPCS3 in appimage and it worked fine with whatever llvm Mesa used, maybe that is feasible for Blender as well? Not sure.

[pal666]

If you don't use compute, you don't know what people are talking about and your opinion does not matter.

my opinion doesn't matter only for people who do use compute(that's why i made it clear from the start). but most of people just need to run browser and games and for them my opinion is much better than your

[pal666]

1.) if you are developing GPGPU code there is nothing else but CUDA

only for brainwashed nvidiots. didn't you read on these forums that amd focuses compute resources on vega because they have large gpgpu vega customer?

[pal666]

Vulkan doesn't do compute (for now).

as a matter of fact, it does.(which doesn't mean it can't be improved)

And unlike Vulkan, or DX12, webgpu is supported by everyone including Apple, and therefore will be standard in browsers. It's the first api that we're likely to see implemented by everyone.

unlike vulkan it's wrapper api, it requires underlying implementation(for example vulkan). "everyone" should include gpu vendors

[jrch2k8]

only for brainwashed nvidiots. didn't you read on these forums that amd focuses compute resources on vega because they have large gpgpu vega customer?

Well, first we have to see what the hell they mean by large and even if it is true in certain specific workloads every new compute app going into market this days is CUDA based, even software like blender and davinci work best with CUDA and i think davinci is kinda silently letting OpenCL rot. Not taking into account any AI stack this days is CUDA only or have very bare bone OpenCL implementation if at all.

Caveat: I dont use nVidia, my RX570(fully FOSS stack) is working beautifully and i love it but i'm not delusional enough to not realize AMD position in compute is a bloody mess at best

[boxerab]

I have been using ROCm OpenCL stack on my RX470 for years, and I am very happy with the results.