Radeon ROCm 4.1 Released - Still Without RDNA GPU Support

[skeevy420]

My bad. I forgot it. APUs moving toward RDNA if the roadmaps are correct.
CDNA is more like a successor to GCN and RDNA developed more independently.

For all intents and purposes, CDNA is basically headless RDNA; both are the successor to GCN.

And there are RDNA APUs in the game consoles. I'm sure we'll see consumer ones eventually.

[raun0]

GCN is Vega/VII and earlier AMDGPU. *DNA are the GPUs released after that. RDNA is for gaming, general purpose and CDNA is for compute.

Pretty much all their APUs are GCN and I wouldn't claim those are for super computers and servers.

My bad. I forgot it. APUs moving toward RDNA if the roadmaps are correct.
CDNA is more like a successor to GCN and RDNA developed more independently.

[andre30correia]

no opencl with new amd is that? well I buy nvidia or intel in the end of year, it's ridicullous at least they can help clover to have some gpgpu support out of box or simple to use, even if its bad

[Spacefish]

The TR4 IO Die is different, it has much more IO and can talk to more chiplets.. Question is, is the AM4 IO Die a scaled down version of the TR4 IO Die, the other way arround, or are they completly different chips?

[chithanh]

The IO hub is actually pretty much the same between Zen2 and Zen3.

I think this is only true for the desktop parts (Matisse/Vermeer) but not for Milan as the previous comment said. The Milan I/O chiplet is mostly the same but there are differences in the PSP and memory controllers:

https://twitter.com/Locuza_/status/1371952669900865539

[Spacefish]

IO Die
The IO Die was probably one of the first chips developed for the platform surrounding the AM4 socket / Zen in general.. AMD was linked up with Global Foundries back then and did not have a large R&D budget, as they had very uncompetitive CPU offerings.
It can be used as a PCH (x570 chipset) and or as the IO Die to link the chiplets / offer all the integrated USB/Ethernet,SATA PCIe connectivity.
This chips is developed for GF 12nm and still produced there. And probably the design did not change much during Zen -> Zen 3, as there is no reason for that / R&D is expensive for little gains.
Furthermore this chip has do do multiple PCIe4 Lanes, which require high clockrates and that requires a lot of power.

Speculation: Maybe they did skip on some fine grain clock gating oportunities as this chip is probably high risk / complex already anyway. Just think of how many protocols / functions it has to implement.

Compare that to Renoir, which does not use that chip, it uses much less power, but only does PCIe3 and is probably a newer design and only some blocks are copied over on the logic level from the IO Die design..

Furthermore, get into AMD shoes: They have much larger demand than they can deliver products, they are limited by the 7nm capacity they booked with TSMC some years ago.. Why would the move the existing chip to a 7nm process or another TSMC process, if that means cutting down output, this would make no sense from a business standpoint.

I guess the Chip is replaced with the new Platform (New Socket + DDR5 / PCIe5) as a large part of that chip has to be redesigned then... Maybe they even move to a monolithic die on the desktop then for up to 8 or 12 core designs with included graphics.. But this is just wishfull thinking.. But on 5nm the chip could be pretty small even with 8 cores and integrated I/O die... Depends on the complexity and space requirements of then Zen 4 Cores, which we don´t know and probably wont see for another 6-12 Months.

If you wan´t good energy efficiency on the desktop today, get a renoir based APU for AM4 or wait for cezanne.
New Platform might be different, but no one knows..

RDNA Support in ROCm
It´s a pitty that they don´t support it.. But i do get that they focus on existing customers / HPC market with CDNA.. Still believe it´s a mistake that will materialize in the long run to not allocate more resources to it, to get a minimal RDNA Support such that academia / students can board the platform early on.

[skeevy420]

Yes, I was looking at it yesterday night (EU timezone)
It is fairly old, as the packages are old.
After the experience of the latest 4++ days I am tempted to put together my on pkg

With regards with pacman command: AMD expressly states to remove old packages before installing ... (on ubuntu/centos/redhat)

So do they, especially with the locally compiled packages, but some of those packages have been updated within the past three hours so give it some time or give them a hand.

But in their instructions are the remove commands between updates and the notice to compile in a clean chroot when applicable.

[Grinness]

Have you tried the Arch4edu repos with their rocm-arch packages?

You shouldn't need to do more than "sudo pacman -S rocm-dev rocm-utils rocm-libs" if bridgman is correct about how it should just work w/o the dkms package with newer kernels.

I have a Polaris but no clue on how to run any ROCm stuff cause I've never done that so I can't help much more than that without being point towards some things to try with a tutorial.

Yes, I was looking at it yesterday night (EU timezone)
It is fairly old, as the packages are old.
After the experience of the latest 4++ days I am tempted to put together my on pkg

With regards with pacman command: AMD expressly states to remove old packages before installing ... (on ubuntu/centos/redhat)

[skeevy420]

Bridgman,

is there an effort to simplify/streamline the adoption of ROCm for end users?
Conceptually ROCm/HIP is fascinating and very promising, but it does not compare with the simplicity of installation (and 3rd party support/implementation) of nvidia.

Years ago I used to do cuda on nvidia,
Now I am on polaris (arch linux), i just spent 4 days (probably more, between AMD website, git repos, Arch AUR) to try to get pytorch, torchvisiom, torchtext to run.
Tried from native rocm on arch, to dockers. An ugly mess!

Finally yesterday night by creating my own pkgbuild (and a lot of kicking and screaming) got all instaleld
... to find out that the gpu computing stalls, no errrors, no dmesg -- just trying a jupyter notebook tutorial from torchtext

How do I debug?
How do i ask for help? ( I assume first thing will be: arch is not supported, followed by oh, polaris is not officially supported)

On top of that ROCm 4.1 is out and I have to recompile EVERITHING ....

Come on. ...

Have you tried the Arch4edu repos with their rocm-arch packages?

You shouldn't need to do more than "sudo pacman -S rocm-dev rocm-utils rocm-libs" if bridgman is correct about how it should just work w/o the dkms package with newer kernels.

I have a Polaris but no clue on how to run any ROCm stuff cause I've never done that so I can't help much more than that without being point towards some things to try with a tutorial.

[skeevy420]

For information AMD has two architectural lines GCN and RDNA.

RDNA is for people whom has empty energy drink cans on their desk and the main color lights flashing all over the place.

GCN is for super computers and servers.

GCN is Vega/VII and earlier AMDGPU. *DNA are the GPUs released after that. RDNA is for gaming, general purpose and CDNA is for compute.

Pretty much all their APUs are GCN and I wouldn't claim those are for super computers and servers.