AMD Ryzen memory optimisation - The effect of tRC timings
Published: 14th July 2017 | Source: OC3D Internal testing | Price: |
Today's results showcase what is possible by making a change to a single timing, one change in a list of 26 new parameters that are found in AGESA 188.8.131.52.
These performance results are the minimum that we can achieve from memory tweaking with Ryzen, as there could easily be more memory timings or options that are set to sub-optimal levels.
This is something that can be considered in a subsequent article, as any Ryzen using enthusiast will want to take advantage of every performance tweak that they can sink their teeth into, especially if they are willing to learn or are already proficient inside their motherboard's BIOS.
In our testing, we found that some of the biggest gains from tRC timing tweaks can be found when using fast 3200MHz memory kits, where auto tRC values are furthest from our RAM's XMP/validated values. These performance gains are minor in the grand scheme of things, but they are certainly worth utilising if you want to get the best performance out of your CPU. Even in gaming workloads, we found that this tweak was able to yield some useful gains, provided game performance is CPU limited.
One other thing that needs to be noted is that we do not expect all Ryzen CPUs or memory kits to handle this tweak as easily as our test system, so please make sure you conduct a full suite of memory stress tests after making any performance tweaks to ensure stability. The most important thing here is to access greater performance while maintaining system stability, as nobody wants a system crash during an extended gaming session or an important rendering or encoding task.
For stability, we do not recommend moving tRC values to below your chosen memory kit's XMP/rated timings and users will need to be careful not to set tRC timings below tRP+tRAS as this will cause problems (likely requiring a CMOS rest). If the memory's rated values prove unstable, we recommend increasing your tRC timings until stability can be achieved.
In AIDA 64 we can see a notable boost in memory read/write/copy speeds and in 3DMARK Fire Strike and Time Spy, we see that this translates to a significant increase in both CPU and combined (CPU+GPU) performance metrics. It is clear that this relatively small change to memory timings offers performance gains that are more than worthwhile.
While the performance boost in several applications is minimal in many cases, it still offers a marked improvement with no real downsides after you have confirmed system stability.
With every new CPU platform comes a set of unique quirks and tweaks that can be utilised to boost performance, reminding us of Phenom II X6 Northbridge overclocking and other architecture specific characteristics. We have already seen this in action with Ryzen, where memory clocks are linked to Infinity Fabric speeds, allowing high-speed memory to hasten thread-to-thread communication as well as offer greater memory performance.
Manually adjusting tRC timings have a notable performance impact on the AM4 platform, something that avid PC tweakers should take advantage of if possible. Ryzen is a new platform and there is certainly a lot to explore here in regards to memory optimisation as well as other potential performance improving tweaks.
You can join the discussion on the benefits of manual memory timing tweaks with AMD Ryzen on the OC3D Forums.