Ryzen 3000 DRAM & IF tuning with ASRock X570 Tai Chi

[Gouhan]

Hello all.
Been quiet over at TheOverclocker for a number of reasons. That said, a lot of it has to do with just how long it takes to test a new platform.
We always want to bring you more than what a regular review can tell you, but that isn't always possible because it literally takes a long time to learn a platform.

That said, this is some testing I was doing between DRAM frequency and Infinity Fabric performance scaling.
What I wanted to find out is just how much do the additional DRAM frequencies help and where exactly is the sweet spot for performance, latency and of course memory bandwidth.

Full editorial or formalized editorial will come later when I do the full board review for the ASRock X570 Tai chi.

Test System
==========
ASRock X570 Tai Chi (AGESA 1.0.0.3 beta Bios 1.41)
AMD Ryzen 5 3600
CORSAIR Dominator Platinum SE Contrast 2x8GB (Single Rank, Samsung B-die)
CORSAIR Force LE 960GB SATA Drive
CORSAIR AX1500i PSU
CORSAIR H100i SE Platinum AIO (Full review coming soon)
EVGA RTX 2080 FTW 3
Windows 10 x64 1903

* As we know AMD on the 3rd gen Ryzen chips allows half rate Infinity Fabric clock. So FCLK = 1/2 MemCLK above 1800MHz (SDR rate).
* I can't confirm this but from other spoken to it seems the maximum frequency for the FCLK is around 1900MHz on an ok to good CPU.
* The significance of being able to do a 1,900MHz FCLK means memory can operate at 1:1 with a 3,800MT/s (1900MHZ) memory speed to match.

Memory Bandwidth

 

[Gouhan]

 

[Gouhan]

More to come in the editorial, but so far it looks as if true to AMD's word 3600MT/s is the sweet spot, but if you can go higher with a 1:1 ratio you're doing even better.
I do suspect the really great CPU samples can go above 1,900MHz but I've no record or knowledge of any CPU being able to do this, time will tell.
It is also worth keeping in mind that not all motherboards tolerate an FCLK above 1,800MHz and will immediately drop you to half rate or not POST at all.

 

[Gouhan]

Update

Test System
==========
ASRock X570 Tai Chi (AGESA 1.0.0.3 Official Bios 1.60)
AMD Ryzen 5 3600X
CORSAIR Dominator Platinum SE Contrast 2x8GB (Single Rank, Samsung B-die)
CORSAIR Force LE 960GB SATA Drive
CORSAIR AX1500i PSU
EK Phoenix X360
EVGA RTX 2080 FTW 3
Windows 10 x64 1903

Highlighted only what has changed in the setup in Bold.

Finally I think I'm well versed with how this board behaves and how to get a fairly optimum performance from it.
While you can go to 4200MT/s memory as stated previously, the DDR 3800 + 1900MHZ FCLK is the best.

It is also worth noting that the CPU-Z voltage is actually correct. This is the practical limit of the CPU in question.
The better CPUs go all the way to 4,400MHz on all cores at this vCore or there abouts.

That said, I do't think it's practical to actually run 1.4v or more for an everyday setup. There's no evidence to suggest this will in anyway hurt the CPU though. Reason for < 1.4v is simply because in the limited sample size of Ryzen CPUs I've tried and what I've observed from others, scaling with voltage is pretty poor after a certain point. A critical point of sorts if you will.
e.g
To make sure this CPU can be stable at 4.4GHz on all cores, it needs 1.475V
As you can see in the screenshot, just 75MHz less needs only 1.375v that's 100mV for 75MHz. Not worth it in the least.

 

[Gouhan]

On a related note. The quickest way I have found to check for stability without having to go through Prime95 etc. try AIDA 64's FP Ray-Trace benchmark. If there's any instability at all, your system will restart near instantly.
The 84'C peak temp CPU package temperature is correct. This test is extraordinarily heavy on the CPU in ways other stress tests or benchmarks do not seem to be with Ryzen 3rd gen.

The 32-Bit iteration is a little less taxing, so it may pass sometimes, but the 64-bit version is the one to run.