4000mhz RAM tuning with Ryzen 5600x

[t0aster]

I've learned a little today about upgrading ram to higher clock speed models and I wanted to share what I have seen so far In my testing.

I run an Asus TUF gaming X570 plus with a Ryzen 5600X. The CPU is in PBO-mode and hits a max of 4.85 the way its currently set up.

To start off, I ordered a 16GB set of Team Group Nighthawk RGB XMP rated at 4000mhz CL 18-20-20-44 1.35v because I wanted to see what sort of improvement I could get over the set of Corsair Vengeance RGB 3200 CL16 I was using. The Corsair was highly recommended when I got it and it has ran rock solid at the XMP rating and I was able to boost it up to 3466 CL15 which gave a good improvement in tests.

So my best benchmark for the Corsair was 3288


So I installed the Nighthawk and Turned on DOCP. It looked to be working correctly setting the timing to 4000 and the timings to the 18-20-20-44 and 1.35 I expected. But when I ran a benchmark this was the result.


So I was really disappointed when I saw that it was actually performing slower than the Corsair.

I went through the Bios again and everything was how it was supposed to be as far as I could tell.
So I did some reading on 4000mhz ram and looked for what I was missing. The important thing I found out is that ram runs better when the infinity fabric (Fclock) timing is 1:1 with the memory clock (the actual memory clock not the number used, so 3800 ram is 1800, 3200 = 1200 and so on). And I read that the sweet spot for Ryzen cpu's is around 3800mhz with the Fclock at 1900.

So I took a reading with taiphoon and put it in the memory calculator and took pictures of the safe settings at different speeds from 3200 up to 3800 (It wouldn't give me timings for anything over 4000mhz) and I started out with the recommended 3200 cl14 timings and I filled in all the sub timings also. And when I booted it up the first benchmark I got was much better.


So since that went so well and I saw a big increase in the other Benchmarks I ran I decided to get right to it and see how it did using the "safe" settings for 3800mhz 16-18-19-18-36.
Now here is where I want to mention one thing I learned. Memory calculator lists almost all the timings in the same order you input into the bios of an Asus TUF x570. But the 2nd two "tRCDWR" and tRCDRD are switched. So be careful there. I now wonder if that could have hindered some of my past attempts at getting higher timings. Because although it did boot to windows and I was able to get an amazing benchmark after about 20min my system reset. That was when I noticed that the order was slightly different. Then I changed them to where they belonged and had things running pretty stable. Then I ran a benchmark without any overclock on the CPU and got this result.


I learned before that my system is stable with PBO up to 4.95mhz if I set all the settings in my bios to match the recommended settings on page 1 of Ryzen calculator and match the settings for best dram stability on the power settings page. So I set all those to the recommended values then used the manual PBO settings I have been using (google 5600x ideal settings) and ran this for my best OC so far.


So now we are talking a 400 point increase on Memory mark alone. Other benchmarks were also highly improved. At this point I am gonna stay where I am for now. I don't know why the Ryzen calculator doesn't give settings for over 3800mhz but I could probably find ones that work through trial and error. But since I have read the sweet spot is supposedly right at 3800mhz for ryzen systems I am gonna assume that's pretty close to Ideal for this set.

I hope this helps some people who are in the same place I was. I know if I didn't read what I did and the ram I just got actually slowed me down I would have probably sent the kit back for a refund and tried a diff set. And prob had the same result :P

[David (PassMark)]

Thanks for the info.

It is probably worth noting however that the memory test is fairly sensitive to the memory speed. Other benchmarks (and many real life applications) aren't nearly as sensitive. Meaning a 10% speed improvement in the RAM might only add 2 or 3% to most applications. So while it is a fun experiment, the improvement is likely too small for most people to notice in daily use.

[t0aster]

Since this post was written, I have learned much more about ram overclocking. And David, you are right about that, It doesn't give really large user end performance Increases. Most users will do fine with XMP settings. And when not done correctly can actually cause large performance hits.
That being said. Latency performance is the most important thing to measure. Bandwidth rates are usually tied to the clock speed of the kit. I found that if you want to get the fastest possible RAM the way to go (on the ryzen platform) is to get a kit rated over 4000Mhz with a cl18 or better. I use Tforce Xtreem -TEAMGROUP-UD4-4500 and Gskill Ripjaw F4-4000C15-8GVK.
Both kits can run at 3800Mhz-cl14 for me and I am currently using 2x8GB sticks of Ripjaw and 2x8GB TeamGroup together because they work together so well. I don't know if there are any kits able to run lower than cl14 at 3800. But any top end B-die kits should do really well. Kits rated below 3800 may not be able to overclock as low.

Now the 3800Mhz limit is pretty much on point. Some ryzen cpu's are able to run the Infinity Fabric + Memory clocks higher than that, but you need to check for WHEA errors. I could set the clocks to 4200Mhz on a 5600x I have and get ultra low latency and close to max bandwidth speed and the system would appear stable. But I found whea errors would start to appear in hwinfo over and over. I haven't looked into the reason they happen, but I am guessing it's because the I-fab is at such a high setting. Past bios updates have raised the I-fab limits, and while it is possible future bios updates may be able to raise the stable speeds even higher, I think It's more likely that the 3800Mhz (give or take) limit is the best we are gonna get.

My 5600x is able to run a higher I-fab+mclk than my 5900x. And from the charts I have looked at, the 5600x chips typically run with I-fab set to speeds over 2000Mhz and some users don't notice any problems except for the thousands of whea errors. After finding the lowest stable (500% 1usmus-v3-TM5) timings at 3800Mhz and 4200Mhz. - I am planning to compare the performance difference of each after overclocking the CPU. Only because I am interested in knowing if there is any performance penalty due to the whea errors or not. The RAM benchmarks show a decent improvement in latency and bandwidth. So the plan was to find the best performing overclock at 3800Mhz since that is usually achievable without a problem. Then see if it is possible to get more performance with RAM @ 4200Mhz with the Ryzen 5600x.

I haven't gotten around to it yet because I have been testing out my 5900x in my spare time. Right now there are so many different opinions on what the best way to get the most from the new Ryzen 3 chips. I am pretty happy with the results I've gotten with beta CTR 2.1. I paid for the early access for the last couple months and It really is an awesome piece of software. I've managed to get an all time high of 44012 in the passmark CPU test (which may be higher than the highest reported score for 5900x) which I couldn't submit and a submitted score of 43,395 https://www.passmark.com/baselines/V...d=137690043799 using a corsair h150i 3 fan AIO, without using crazy voltage levels or playing with the curve optimizer. Since the passmark benchmark tests single thread and multi core performance tests and covers many different performance points (and since this is the passmark forum) I'm not gonna mention other benchmarks (but they are high).

I did have a chance to try CTR 2.1 with a 4200Mhz RAM + 2100 Infinity fabric(whea errors) but the results were inconsistent. And I am more interested in the 5900x for now since it's basically 2 5600x's. But when I do get back to the 5600 I'm gonna look into what effect the whea errors have on performance. But my guess is that the errors mean the cpu needs to correct some error repeatedly (hurting performance) and any performance gain from the higher RAM clock will be negated.

Below are some test results.



Memory latency test, Tforce 4500cl18 Gskill ripjaw 4000cl15 mixed 2x8G and 2x8GB
I expected the Tforce to hold the gskill back a little but the t-force keeps right up in the 4x8 configuration
I like to use premium B-Die kits rated much higher. Just keep in mind the XMP profile will be set too high for most ryzen cpu's
Just for convenience I purchased typhoon burner to change the xmp profile.



I couldn't post the result due to a previous 5% difference. Even after buying the full version. The ideal
method for overclocking Ryzen 3 is still up for debate.