AMD Ryzen memory optimisation - The effect of tRC timings

I like it how the test rig specs in the box say "1080 Founders Edition", although the picture of the PC under it shows and ASUS Strix card :p
 
I like it how the test rig specs in the box say "1080 Founders Edition", although the picture of the PC under it shows and ASUS Strix card :p

That is the thing with test beds, hardware changes all the time. I can't take pics with every hardware configuration.
 
That is the thing with test beds, hardware changes all the time. I can't take pics with every hardware configuration.

No no, I understand dude. I was just pointing it out, not to put it in your face or anything though.
 
For whatever it's worth, my memory/motherboard does not seem to exhibit this issue. My XMP 2.0 values populated and booted without any fussing, and the tCR value is accurate.

MB: ASRock AB350M Pro4, BIOS v2.5 (AGESA 1.0.0.6)
Ram: Corsair LPX 2x8GB 3200MHz C16 (CMK16GX4M2C3200C16S)
Timings: 16-18-18-36-54-1 (tCAS-tRC-tRP-tRAS-tCS-tCR)
https://valid.x86.fr/te3x7z

A new update was just released today for AGESA 1.0.0.6a, so I'll need to see how that impacts things, but in this one case at least, things appear to not wonky.
 
For whatever it's worth, my memory/motherboard does not seem to exhibit this issue. My XMP 2.0 values populated and booted without any fussing, and the tCR value is accurate.

MB: ASRock AB350M Pro4, BIOS v2.5 (AGESA 1.0.0.6)
Ram: Corsair LPX 2x8GB 3200MHz C16 (CMK16GX4M2C3200C16S)
Timings: 16-18-18-36-54-1 (tCAS-tRC-tRP-tRAS-tCS-tCR)
https://valid.x86.fr/te3x7z

A new update was just released today for AGESA 1.0.0.6a, so I'll need to see how that impacts things, but in this one case at least, things appear to not wonky.

Well that is certainly a positive thing to hear, one of the main points for this is to raise awareness and get a potential fix in future BIOS' if possible. I do remember the Taichi having the issue at launch, so this is likely a recent development.

Hopefully other manufacturers will soon follow suit. It is memory performance that is simply sitting on the table.
 
Great article. Just finished reading it.

It's quite amazing how such a small increase in tRC values makes to the performance. I'm pretty surprised tbh. If AMD keeps this up Ryzen is just going to get faster and faster. Just makes me wonder how Ryzen 2 is going to be with all the kinks worked out with 1.
 
You have made a great start. I posted a note about finding improvements in latency to improve the Ryzen gaming issues here back when when Tom was doing the x370 motherboard reviews. I have just seen this article.

For your part 2 article, You might like to also try tras 28, tRRDs 4, tRRDl 5 and tFAW 16. If unstable, increase the timing setting by one and retest.

You should see a significant latency reduction.

The latency improvements do not magically make GPUs faster for higher frame rates. The lower latency helps reduce memory access queuing time with the R7 16 threaded memory loads when it is competing with the GPU for access to system memory. 12 threads from a 1600x do not generate as many memory requests as the 16 threads from an R7 so the R5 1600 cores are more efficiently utilized producing draw calls for the GPU.

As part of your investigations, take a note of the before and after CPU and GPU utilization during each benchmark. You should see an increase in both CPU and GPU utilization in the tests with lower latency
 
For some odd reason, the latest Asus beta bios set mine down to 42 using stock timings even on XMP settings.

So does that mean it's time for a Ryzen re-retest video :)
 
Last edited:
Back
Top