In case you didn’t know, AMD recently detailed three new RDNA 2 features: RAGE Mode, Smart Access Memory and Infinity Cache. The SAM or Smart Access Memory is a new feature that allows AMD CPUs to access all the VRAM in a Radeon RX 6000-series GPU.
AMD’s solution was only supposed to work with its latest Ryzen 5000-series CPU and RX 6000-series GPU lineup, with compatible B550/X570 motherboards. So many were concerned that this feature would only work within AMD’s ecosystem, but now it appears AMD is opening the compatibility on both Intel and Nvidia-based platforms.
In an interview with PCWorld AMD has stated that its Radeon group is working with INTEL to get this feature to work with RX 6000-series GPUs, and also on Intel’s latest compatible CPUs and motherboards.
It was also mentioned that AMD’s Ryzen group is working with Nvidia to get the new Smart Access Memory/SAM feature to work with NVIDIA GeForce GPUs. Thus, we can expect SAM to work on Nvidia GPUs as well.
AMD also stated in the interview that this feature isn’t just a simple “toggle option” which you can just easily switch on/off. Instead, it will require development and proper optimization to get max performance gains from the feature.
Recently, Nvidia also confirmed that the company is working with Intel to get a similar feature, to work on their own GeForce GPUs. NVIDIA stated that they are also working on their own SAM or Smart Access Memory feature similar to what AMD has enabled on their RDNA 2 GPU lineup.
Nvidia said this SAM feature will be enabled on all the GeForce RTX 30-series Ampere GPUs via future software and driver updates, and it will be compatible with both AMD and Intel processors. Thus, Nvidia GPUs are also going to leverage a similar feature which might give them an upper hand in Games as well.
From NVIDIA, re:SAM: “The capability for resizable BAR is part of the PCI Express spec. NVIDIA hardware supports this functionality and will enable it on Ampere GPUs through future software updates. We have it working internally and are seeing similar performance results."
— GamersNexus (@GamersNexus) November 12, 2020
Nvidia said this feature is easy to implement since the “resizable BAR” is actually a part of the PCI-Express specifications, and NVIDIA’s existing hardware fully supports this functionality.
As per this PDF document dating back to 2008, “This optional ECN adds a capability for Functions with BARs to report various options for sizes of their memory mapped resources that will operate properly. Also added is an ability for software to program the size to configure the BAR to.”
So every PCIe compatible device can enable it with a driver update via the software.
This new technology feature will not require a PCIe Gen 4-compatible platform as it will be supported by PCIe Gen 3 systems as well. BAR basically defines how much discrete GPU memory space can be mapped. Modern PCs are typically limited to 256 MB of mapped memory.
It is typical today for a discrete graphics processing unit (GPU) to have only a small portion of its frame buffer exposed over the PCI bus. For compatibility with 32bit OSes, discrete GPUs typically claim a 256MB I/O region for their frame buffers and this is how typical firmware configures them.
A GPU, supporting resizable BAR, must ensure that it can keep the display up and showing a static image during the reprogramming of the BAR.
This feature is rather important for graphics hardware, because the PCI BARs are usually limited to 256MB while on modern cards you can easily find 4GB or more VRAM. The end result is that only a fraction of that VRAM is CPU accessible, causing a whole bunch of workarounds in the driver stack for that hardware.
AMD says that with SAM they can access all of the GPU memory, thus removing any bottlenecks. This will also allow for faster performance. In conventional Windows-based PC systems, processors can only access a fraction of graphics memory (VRAM) at once, limiting system performance.
With AMD Smart Access Memory, the data channel gets expanded to harness the full potential of GPU memory, utilizing the bandwidth of PCI Express to remove the bottlenecks and increase performance.
To explain this feature in more details, AMD’s new Smart Access Memory feature will boost the overall gaming performance by optimizing the data transfer between the CPU and the GPU.
What this basically means is that the Radeon RX 6000 GPUs can now work and operate in tandem with AMD’s Ryzen 5000-series processors, provided you use a 500-series compatible motherboard, through this new Smart Access Memory feature.
Smart Access Memory aims at optimizing both the GPU and CPU to offer the best possible performance when they operate in tandem.
According to the concept, once you enable the Smart Memory Access feature in the RX 6000 card’s VBIOS and the motherboard BIOS, both the CPU and GPU will gain full access to each other’s memory. Doing this will maximize data transfer and performance between the CPU and the GPU’s VRAM.
Smart Access Memory helps boost performance by enabling faster data transfer speeds between the CPU and GPU. And by pairing this efficient data transfer smart access feature with the new AMD 128MB Infinity Cache feature might help boost the throughput between the CPU and the GPU as well.
The new Infinity Cache takes advantage of the GPU’s data paths to maximize performance while minimizing the data movement and power within the GPU itself.
According to AMD, Infinity Cache delivers a 10% increase in power efficiency and it also doubles the bandwidth (almost 117% increase), all at a lower power than traditional memory.
Infinity Cache is based on the Zen CPU’s L3 cache design. Infinity Cache boosts performance-per-clock scaling as frequency increases largely because the GPU is now less constrained by external memory bandwidth limits. The Infinity Cache also boosts ray tracing performance, as more of the data set is kept closer to the compute units to feed.
Most importantly, AMD says that individual game developers will have to optimize games for this new Smart Memory Access feature. As such, it could take six to twelve months before we can see this new tech in games.
Lastly, AMD’s Radeon RX 6000 GPUs will also support the DirectStorage API, which might help reduce the game load times. It remains to be seen how much of a performance boost this SAM feature gives on both AMD and Nvidia’s ecosystem.
Stay tuned for more!