AMD Will Bring Smart Access Memory Support to Intel, Nvidia Hardware
When AMD announced its Smart Access Memory, it sounded as if the company had finally designed a method of allowing Ryzen CPUs and GPUs to specifically work together in order to deliver higher performance than either could achieve together. Our performance tests confirmed that SAM worked fairly well, but it hasn’t been clear if the future would be restricted to AMD-AMD CPU/GPU configurations or not.
Thanks to a recent PCWorld interview, we have an answer. According to AMD, it has people on the Ryzen team working to get SAM working on Nvidia GPUs, while there are people on the Radeon team working with Intel to get the feature functional with Intel CPUs and chipsets. If AMD is comfortable making this kind of announcement, it implies that there’s reciprocity in these arrangements, meaning we’ll see cross-platform, cross-vendor support, though we haven’t heard anything about Nvidia/Intel cooperation. It only makes sense for the two companies to work together, however, since the alternative amounts to giving AMD a free performance advantage.
This confirms that SAM isn’t an AMD-specific technology as such, though AMD has done the work of enabling the feature before anyone else did. Resizable BAR Capability (that’s the PCIe specification-name for SAM) was initially baked into the PCIe 2.0 standard in 2008 before being modified in revisions to PCIe 3.0 in 2016. Microsoft added support for the feature with Windows 10 when it introduced Windows Display Driver Model 2.0, but evidently, no GPU vendor supported it until now.
If this were an AMD-specific technology, one might suspect that the company had to design Zen 3 and/or RDNA2 to use it. The fact that support can apparently be extended to Intel and Nvidia hardware implies the feature either wasn’t viewed as being worth the trouble or that the companies in question weren’t aware it could deliver a real uplift in performance until someone actually tested it. The latter would be rather droll.
According to AMD, there’s some work required to support the feature appropriately, implying we may not see it enabled immediately on Intel and Nvidia platforms. It’ll be interesting to see what kind of performance we see other platforms and hardware pick up from enabling this capability — Intel might benefit more than AMD (or vice-versa) and AMD GPUs might benefit more than Nvidia cards or the reverse.