AMD
The {hardware} that has enabled and pushed the current explosion of generative AI merchandise has primarily been manufactured by Nvidia. That is thanks partially to its dominant place within the client and workstation graphics markets, plus the well-established CUDA libraries and its AI-accelerating Tensor cores.
Though Microsoft makes use of “tens of hundreds” of Nvidia GPUs in its Azure knowledge facilities, Bloomberg stories that the corporate is additionally working with AMD to enhance the AI capabilities of its GPUs. Though extraordinarily gentle on particulars, the report says Microsoft is contributing “assist” and “engineering sources” to AMD to enhance its merchandise’ efficiency in AI workloads.
Bloomberg additionally says that AMD is engaged on a Microsoft-developed AI accelerator named “Athena,” which The Info reported particulars on final month. A Microsoft spokesperson, nevertheless, flatly denied that AMD was concerned in Athena’s improvement.
The upside for Microsoft is {that a} extra aggressive marketplace for AI-accelerating {hardware} can save on server prices as the corporate appears to be like to construct AI options into increasingly of its merchandise. For instance, Microsoft is reportedly engaged on a non-public model of ChatGPT to supply to companies that might price “10 occasions” as a lot to run because the common model of ChatGPT; if the server {hardware} required to run these generative AI fashions is cheaper, Microsoft may decrease its costs, its prices, or each to make these merchandise extra interesting.
The 2 firms have kind-of-sort-of collaborated earlier than; “Floor Version” Ryzen processors have been utilized in some older variations of Microsoft’s Floor PCs, although not in any of the newest Floor gadgets. Just like the SQ-series chips from Qualcomm that Microsoft makes use of within the Arm variations of the Floor Professional, the Floor Version Ryzen processors weren’t a lot totally different from common Ryzen processors—they’re the identical chips AMD offered to different producers however with Microsoft-assisted optimizations to the “firmware, drivers, and software program stack.”
Though it isn’t consultant of how every firm’s server-grade GPUs carry out, current exams from Tom’s {Hardware} illustrate how AMD’s GPU architectures wrestle with AI workloads. In testing with the Secure Diffusion picture generator, AMD’s present Radeon RX 7900 XTX flagship carried out slower than Nvidia’s RTX 4090, 4080, and 4070 Ti, though the RX 7900 XTX is quicker than the RTX 4080 in most video games. Even the entry-level RTX 3050 managed to beat all of AMD’s previous-generation RX 6000-series playing cards (software program optimizations may assist these playing cards carry out higher, however the truth that the software program was Nvidia-optimized by default says quite a bit concerning the present state of issues).
