RAM Bandwidth scaling with Ryzen 5 2400G and Linux

Intro

For my first article with The Chip Collective I am going to be doing some testing in Linux. I wanted to know how graphical performance scales with RAM speed with my AMD Ryzen 5 2400G and it’s Radeon RX Vega 11.

This was not meant as some super rigorous testing, but rather as a simple benchmark showing the effect of higher RAM overclocks on my system, also I think it’s the only testing of this kind using Linux.

System specifications

CPU: AMD Ryzen 5 2400G + Radeon RX Vega 11

MOBO: Asus B350M-E, bios 5007

RAM: Crucial Ballistix Sport LT Red 2x4GB DDR4-2400

OS: Manjaro Linux with mesa 19.1.6 and kernel 5.2.10-rt.

Setup and Testing Details

I used the Xonotic automated benchmark since it is available for Linux, I already had it installed, and well, its automated.

Ultra-quality preset with motion blur disabled was used for all the testing.

The system was running stock except for the RAM, All the speeds were tested with CL18 timings, but that should not matter for the APU.

My idea was to go from DDR4-3200 to DDR4-2200 down by 200 MHz steps but since the BIOS did not allow some of the frequencies, I tried the closest ones instead.

The test method was to reboot, change RAM speed in BIOS and then verify it was correctly applied and then run the benchmark.

Results

We can see the average FPS goes up by 18.5% while the RAM bandwidth goes up by 50% this means the bandwidth the bandwidth scaling is 26.5%, I was expecting this to be higher, but it seems Raven Ridge isn’t as bandwidth starved as I thought.