Why is AMD not part of OpenPOWER?

AMD is a member of RISC-V, and a member of OpenCAPI, yet not a member of OpenPOWER.

If there is a concern about being open to lawsuits about stealing CPU designs, then I would expect that AMD would certainly not be a member of the RISC foundation, which also deals with CPU designs. Also if this were a reasonable fear, why would Nvidia, which also makes CPUs (Tegra) be a member of OpenPOWER?

AMD is a member of OpenCAPI, an interface originally designed for POWER if I am not mistaken. It just seems like a weird omission to not be a member of OpenPOWER, given the other organisations AMD is a member of. Thoughts or information anyone?

1 Like

Because AMD designs RISC chips. The K5 was a RISC chip that could translate instructions FASTER than the intel chip of that year. They have designs and they know their shit, thats why.

They have been involved in POWER based systems and make a lot of technology specifically for those platforms. IE the Talos 2, Sony RSX Chip, etc.

maybe because AMD Does not want to design systems based off IBM’s POWER architecture as that is the main point of the OpenPOWER group in relation to Companies like AMD. I mean Intel is also not a member of OpenPower it just does not make sense for Processor manufacturers to get involved in industry consortiums/groups that they do not have a product in.

4 Likes

It’s not the point of OpenPOWER in relation to Nvidia though. AMD might do well to make their Radeon Instinct cards work well with OpenPOWER.

Interesting, and its core was actually Berkeley RISC derived, not just RISC in style. So you are thinking that RISC-V designs might be used for AMD’s x86 cores? Since POWER does not have Berkeley RISC heritage its designs would be of less value to AMD.

Hmm, since RISC-V is BSD licensed, does this mean that AMD chips might have a license attribution for RISC-V somewhere in the packaging?

How is Nvidia doing anything specifically for Talos II, Raptor Computing only offers AMD graphics cards for it; also the Sony RSX predates the formation of OpenPOWER and current consoles are using either ARM or x86.

1 Like

Go look at the talos thread tk was freaking out about it for a whole week.

Actually I think its more towards PA-RISC with the separate CCX’s in FX and now Ryzen with how cores and process lineage was handled. I want FX so damn bad because running code like how the processor was designed… Mathematically FX 8 cores should be retardedly powerful.

Also the rsx was used long after the ps3 in CELL and POWER based servers. Till 2013 I believe.

I mean at that point you are just talking about writing drivers for the OS that runs on IBM POWER processors and not needing to be part of the consortium/group to do that as the OS’s that run on those systems have their own dev teams and groups.

Nvidia is attacking the machine learning and AI market and needs to build a depth of knowledge on high end processors so it is conceivable that they are joining groups like OpenPOWER to gain the knowledge they need to build high power processors not just graphics cards.

They already do?

OpenPower is a consortium around IBM server and HPC technology. Its IBM´s turf revolving around the power architecture. Eveyone on the consortium are there cause the are involved in producing the Power Systems. Take the term ¨Open¨ with a grain of salt.

Nvidia is there because they need to cooperate with IBM to produce machines with integrated Nvisia Tesla cards (currently p80 and p100) with NvLinks for the HPC market. AMD has no such product. The professional Radeon parts are just simple PCI-E cards. No specific cooperation and specific design required as long as there is a PCI-E free slot on a machine whether power or x86 based. Also the Radeon parts are not really suitable for the HPC market at this point anyway. They are more a workstation part.

Nvidia has drivers for POWER, but they are not made with Talos II in mind, that’s why Raptor sells AMD cards only. I didn’t know there were Cell servers, that’s kind of interesting.

Yup sony sold CELL and XELL till around 2013.

If that is the only reason Nvidia is there, then my question is why doesn’t AMD want the such a product? A “ROCLink”, if you will, would allow much better bandwidth to Radeon Instinct cards. If NVLink is superfluous due to OpenCAPI, then why is Nvidia staying on OpenPOWER?

If not, why doesn’t AMD work on a “ROCLink” to match performance?

From what I see, this would be just as beneficial to AMD.

1 Like

The difference between the two is NV wants GPU’s and AMD wants CPU’s.

Because Nvidia Telsa and NvLink technology has a good decade of development behind it and Nvidia has 10 times more R&D resources to make such a thing. If AMD had the technology and resources to develop sth similar they would…They do not up until now and do not target the specific market. The capital needed for such technology is huge. Very few companies are capable of it.

Also Nvidia is not staying on open power. Nvidia it part of all other OEMs consortiums that produce Telsa based products.

Could you expand on that? That’s very interesting,

Nvidia Telsa card products are provided by all OEMs and EMC companies that provide HPC products. Dell, HP, IBM etc all have p100 based server and HPC systems in their catalog. Nvidia is involved with all of them to provide support to make those systems and make them work reliably.

EDIT: simply put Nvidia sells their stuff to everyone that want it and helps them make the final system…Thats how all companies do it with their products.

Does AMD not have something similar with Radeon Instinct/MI Series?

They do but the ecosystem is brand new. I do not think these products are even available yet for actual purchase. I am quite sure that deals are already been made with the EMCs and other companies but i doubt there will be actual system available within this year. If anything these will come up next year summer at the earliest.

Nvidia on the other hand is part of the market for years. That is why they are part of every consortium…

1 Like

They could make licensed POWER cores just as they have done with ARM. It would at the least give them more options.

As a licensee of ARM cores and a member of RISC-V, not being a member of OpenPOWER seems like a strange omission to me. In addition to the possible Radeon Instinct related benefits.

I would lose my shit if I could get that. That being said, PowerPC is still around and is basically the consumer version of POWER now. VIA and FreeScale Silicon distribute that now though, rather than IBM.

Its a difference of Risc Vs PPC / PWR.

And yes I know POWER and PPC is RISC based in family, though now the tech is different enough from the originator that I am pretty sure its its own thing now.