Talking to AMD’s David McAfee about whether we will see more products utilize the UMA approach, he said that there’s going to be a lot of focus on these architectures moving forward. AMD’s first-gen MAX offerings offered up to 128 GB memory and could dedicate up to 112 GB of system RAM to the GPU, & the same is the case with NVIDIA’s RTX Spark, which dynamically allocates memory to the GPU and CPU, based on the workloads. These dynamic and unified memory architectures are suitable for a broad range of applications.
AMD believes that UMA will help shape its next-gen architectures, products, and roadmap. They are already working on the upcoming Ryzen AI MAX 400 series, which offers up to 192 GB of memory and can dedicate up to 160 GB of memory to the GPU, enabling support for 300B+ AI LLMs.
[Roundtable Jounralists] Q - Okay, and maybe just one more question. Yep. Could we ever see a UMA Ryzen gaming CPU, for example? Or something like, I’ll add to it, can we see something like a Strix Halo with either 3D V-Cache or a on-package memory design, maybe?
[David McAfee - AMD] A - I have no idea, so I’ll start with that. But I do think that with Strix Halo, with NVIDIA entering the space as well, there’s going to be a lot of focus on UMA systems, there’s going to be a lot of focus on identifying the right architecture for what these UMA systems can do, as well as enhancements to, let’s call it, the existing state-of-the-art that exists in these platforms.
And I don’t know which direction that goes over the next couple of years, but I think that it opens a world of possibilities, because I think this is a totally new workload, this is a totally new computing space that we’re solving for here, and I think it will shape lots of things around our product choices, our roadmap products, our deployment options, all of those things.
AMD also endorses NVIDIA’s RTX Spark as it is an advancement to the UMA architecture. Interestingly, David also mentions high-performance desktops when talking about UMA.