Intel’s ex-executive VP and founder of Mahira AI, Raja Koduri, has an interesting insight on the world of AI & Data Center stating that the current success of the ecosystem depends on GPUs designed for the PC segment.
Accessibility To PC Gaming GPUs Has Driven AI & Data Center Business Success, According To Raja Koduri
The adoption of consumer-centric GPUs, or so-called gaming GPUs, is pretty high on a global level since not only are they easily accessible, but they are much cheaper in pricing compared to counterparts such as workstation ones.
They are more suitable for individuals to work on. Another neglected aspect of such GPUs is their contribution to the developer community, as the easy accessibility means that every other person, whether he is from any corner of the world, could visit his nearest retailer to get an AMD Radeon or NVIDIA GeForce GPU for work, but current-generation stacks are being framed in a way actually to hinder this aspect.
The same idea is reiterated by Raja Koduri, who, in a post on X, believes that tech giants such as AMD and Intel might need to rethink their approach to consumer GPUs moving ahead. He says that PC developers see this type of tech as essential for their work, and based on how stacks such as AMD’s ROCm and Intel’s SYCL are being developed to put PC GPUs on the sidelines, which means that such developers are missing out a lot. He believes that NVIDIA and AMD are in a much better state than Intel, and ironically, this has hindered the developer community’s adoption of Intel’s consumer GPUs since the same developers would love to have the best of both worlds (strong gaming & AI capabilities).
Yes. Developers on PC GPUs are the key enablers to DC GPU success. So all the dev tools need to work flawlessly on PC GPUs. Currently this is largely true with Geforce. Radeons definitely got better these past 6 months and they are showing increased commitment to PC developer .… https://t.co/9r1uhzksZg
— Raja Koduri (@RajaXg) February 18, 2024
Raja Koduri says that the AI ecosystem has been the sole cause of this since it is evident that GPU manufacturers are focusing on AI accelerators in every way, which means that existing and upcoming resources are more inclined towards catering to their desired audience rather than generic ones. I mean, sure, there are solutions such as the recently-surfaced ZLUDA, which allows leveraging NVIDIA’s CUDA libraries on the ROCm stack, but when you focus on an individual level, it is apparent that modern-day stacks aren’t that “open-source” when it comes to their performance across all sorts of GPUs.
This really captures my sentiment with @IntelAI. Please, please, please put some effort marketing the @IntelGraphics consumer products to ML enthusiasts—they are the people who will help build up your datacenter market share.https://t.co/07oSmGtCe8
— Eric Hallahan (@EricHallahan) February 16, 2024
NVIDIA recently opened up TensorRT-LLM support on its consumer GPUs while AMD has also opened up ROCm support for a certain range of its Radeon GPUs.
Well, for an ordinary gamer, this certainly isn’t something alarming and appreciable since having modern AI capabilities and software stacks to back them up is going to be a necessity in the upcoming AI PC era, but developers could potentially have to rethink their decision to use a consumer GPU moving ahead unless manufacturers change the way the software ecosystem is progressing.