Yesterday, Wccftech reported on a reddit thread where user Merich98 noted that he had enabled Smart Access Memory on his motherboard which is equipped with a 1st generation Ryzen CPU. He also posted some performance numbers.
For the uninitiated Smart Access Memory, or SAM is similar to a resizable Base Address Register or BAR. Typically, the CPU only has access to 4GB at a time in chunks, however this allowed it to access the full VRAM buffer that is available on modern GPUs.
AMD originally only announced this to work on the Ryzen 5000 CPUs and 500 series motherboards, but board manufacturers have also started enabling support for this functionality on Intel systems with 400 series motherboards, and now it seems they’re also enabling it on older Ryzen series CPUs with B450, and assumingly X470 motherboards.
Originally, many assumed AMD had left out previous CPUs due to the fact that they had lacked full rate PDEP instructions like their Intel counterparts and current 500 series CPUs, this however seems not to be the case as confirmed on Twitter by Dr. Ian Cuteress of Anandtech, who received clarification directly from AMD.
Now, for the part most of you are likely really interested to know is the fact that this user used a Ryzen 1700 with an ASUS B450-Plus CPU and enabled SAM using the 2409 BIOS which was released on the 7th of this month. This feature was intended to work with the 5000 series CPUs however again is apparently working with a Ryzen 7 1700.
The user tested performance in Doom Eternal and Shadow of The Tomb Raider, they used the aged Polaris based RX 580. They tested at 2560X1080 at Ultra for DOOM Eternal and noted no performance change in Tomb Raider.
Average FPS: +0.839%;
Min. FPS: -6.20%;
Max. FPS: +0.596%;
1% Low: +0.119%;
0.1% Low: -2.365%
The final results for DOOM Eternal, as noted by Wccftech.
The results were…within margin of error, though its a bit hard for me to trust a user to be testing in the right environment to get consistent runs, in DOOM Eternal its quite normal to get inconsistent runs that need to be thrown out depending on where you test due to variation. While this is interesting, I am unconvinced it actually works, and even so the support from both the driver and API is very important to actually see a performance change.
Some reviewers noted that even with the 6800 series cards and 5000 series CPUs that some games performed worse with the setting on, so I’d say even if the setting is allowed for your specific CPU and motherboard it may not actually help gain performance.
Liked it? Take a second to support Alex Stevens on Patreon!