When AMD announced its Ryzen 5000 CPUs, it introduced a feature it dubbed Smart Access Memory, known more generally across the industry as Resizable BAR. Resizable BAR allows a CPU to access more than 256MB of GPU memory at any given time. The feature can boost game performance on Ryzen 5000 CPUs by between 3-7 percent on average based on our tests at 1080p, 1440p, and 4K. Nvidia claims up to 10 percent improvements for Ampere GPUs.
Initially, SAM was going to be a Ryzen 5000 feature and required a 500-series motherboard from the B550 or X570 families to use. Now, AMD has announced it’ll be bringing the feature to its Ryzen 3000 CPUs as well. A 500-series motherboard will still be required. Note that the Ryzen 3000 APUs, which technically use the Zen+ architecture, are not included here.
Formal support will still require a GPU from AMD’s RDNA2 family or an Nvidia Ampere card. In Nvidia’s case, it also requires a VBIOS update (all RDNA2 GPUs support the feature out of the box). Presumably, motherboard vendor UEFIs will also need to be updated to enable the feature on Ryzen 3000 CPUs. Intel support will be available on Z490 motherboards and upcoming 500-series products for 10th and 11th Gen CPUs. Z390 has apparently been supported as well by some manufacturers, but that’ll be OEM by OEM.
This is a canny move on AMD’s part. Ryzen 5000 chips have been in very short supply these last months, making it harder than usual for the company’s fans to actually buy its hardware. Extending this small boost downward into the Ryzen 3000 family won’t change anybody’s life, but it’s a nice gesture to people who were looking for upgrades this year and may not have gotten the hardware they wanted.
The reason we won’t see Resizable BAR/SAM support added across the spectrum of current PC hardware is that UEFI/BIOS updates and GPU BIOS updates are apparently both required. Motherboard vendors and GPU manufacturers aren’t going to revisit the idea of adding these features to older cards and card families.
Hunting for Performance in the Proverbial Couch Cushions
Companies are getting increasingly creative in the places they look for additional performance. Nvidia’s DLSS feature leverages the cloud and AI/ML training to provide superior visual quality at a lower base resolution. DLSS 2.0 is a substantial improvement on 1.0, and while the feature isn’t perfect, it’s evolving nicely.
In AMD’s case, it’s got a rumored response to DLSS coming soon and we’ve recently seen the effectiveness of slapping a large L3 on top of a GPU, as well as the introduction of features like ReBAR/SAM. Intel has plans to integrate hybrid low-power CPU cores into its products, starting with Alder Lake later this year. Features like Variable Rate Shading have been introduced (if not yet popularized) as another way of diverting more GPU horsepower to the areas that need it most.
I suspect the next few years will see a lot of mud tossed at these proverbial walls as the industry continues to move away from the idea that lithography will provide additional performance improvements, and towards a model that prizes a multi-disciplinary approach to semiconductor performance improvement. Tightening the linkages between hardware and software and squeezing out inefficiencies is how companies are pushing performance forward these days. Clock jumps still count — witness the increase on AMD’s Radeon 6700 XT — but they’re increasingly just one tool in the toolbox.