• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

DRAM overclock complete Noob dumb question

That latency is way too high.

100% stable Ryzen 1600AF with Hynix RAM <65ns

stable Q.png
 
Fellas ... broke into the 60s today ... high 60s, but still :D

Lorry, I am halfway through reading your original thread on Ryzen. Following in your footsteps.


69ns.jpg
 
Last edited:
Fellas ... broke into the 60s today ... high 60s, but still :D


View attachment 147490


I've read a lot of people having issues with memory overclocking on asrock Taichi x370/470 boards and even buildzoid comments on it being an issue. I'm guessing it's part of what's going on.
 
This is on ECO mode with 3750Mhz CPU. I have a ways to go.
 
This is on ECO mode with 3750Mhz CPU. I have a ways to go.
Did you try these settings on nominal speeds? If you like going eco it’s fine, but for testing purposes and finding the best possible DRAM settings the CPU has to clock higher IMO. Otherwise most people, including me, don’t have a clue how it’s suppose to perform in such a low clock. Just a thought!
I know also that having these CPUs clock and boost all over the place when are on auto can affect actual DRAM performance results. But now it’s the low speed of CPU that may confuse us and not know if these results are actually good or not.
 
Last edited:
I keep forgetting he runs his CPU way lower than neccessary.
 
Thanks for your feedback as always, Gentlemen. A couple of questions ...

1) When you "boost" is that only the PBO and Auto Overclocking options in UEFI, or are there other settings to adjust? There is an SOC setting under CPU, and another stand alone SOC on the ASRock Tweaker set of values. I stopped using the Advanced/AMD Overclocking set of DRAM values, because I suspect ASRock has tied some memory training demonry into their settings at boot;

2) I am using TPU's own MemTest64 to test for memory errors, along with the version included in the Ryzen DRAM Calc. After a half hour on MemTest64 it has completed 22 loops through the memory with no errors. Why the continuous looping? It's already gone through the available memory pool 22 times, wouldn't a physically bad RAM have failed at one? Is it a matter of fatigue, or heat build-up, or just trying to establish real-world use with the memory testing?;

2a) What is your opinion of the Karhu memory test out of Finland? I heard it is *very* fast at showing memory errors.

My testing at ECO is because my primary use on the machine is crunching BOINC numbers 24x7 and it gives a 30C degree drop in temperatures 100% cores 100% cycles. Once I wring all I can at this level of power, I will juice it back to 105W and see what else I can get.

Really ... how do I "boost" as you are telling me?
Aida64 cachemem9.png

That latency is way too high.

100% stable Ryzen 1600AF with Hynix RAM <65ns

I am running 4x8Gb on 3950X.
Check the throughputs on Aida64

Ryzen plus doesn't really apply here it scales differently with memory with a good kit you can get into the mid 50ns range

oxrufiioxo ... it intrigues me your R/W and Copy numbers are reversed from mine.
Makes me wonder if I can learn which memory readings are involved in those actions and tackle the timings that way.
Your latency is impressive, I don't see how you can run 4000 at 1:1 and notice there is no North Bridge frequency in your Aida64 screenshot.
Are you running 1:2 at C2T? How many banks of DRAM and what brand? What DRAM voltage does it take to get there?

Lorry, Zach ... never mind my boost question above, I've found your thread on trying to understand Ryzen boost.
 
Last edited:
OK I am in the neighborhood now.
Zach ... just for you, I threw some boost at it in the second run. :)

Latest TPU ECO.jpgLatest TPU boost.jpg
 
OK I am in the neighborhood now.
Zach ... just for you, I threw some boost at it in the second run. :)

View attachment 147789View attachment 147790

Nice 66ns is a pretty normal result imo.... I personally haven't noticed anything in testing below 68ns making a difference in gaming maybe when the 3080 ti comes out people will see some gains.
 
Btw, Ryzen calc is a one for one match for Asus ROG bios in regards to memory fields. In other words you don't have to hunt and peck like with other bios'.
 
Do I proceed by selecting, one by one, separate Primary timing values in order :: Reboot :: Test ?
Or, is there a full set of values (all Primary, eight or ten Secondary) I should try all at once?
Do this:
part-0-system-startup.jpgpart-1-easy-tune.jpgpart-2-tuning-some-timings.jpgpart-3-tuning-cad-bus.jpgpart-4-additional-voltages.jpg

My memory OC:
CL14.png

Good luck!
 
Dont show please your ZEN+ settings and results because they are completely irrelevant for ZEN2. Different line, different architecture and memory sub-system structure, different settings and results.
Although those steps up there, are relevant and may apply to any Ryzen as a procedure, but not as actual values
 
Last edited:
Last edited:
Nice :)
But, boost?

Auto Overclocking and 100 Boost override. Isn't that part of it?

Now that’s nice! I think you are at a very good level now. And as @oxrufiioxo said, hardly you can see any benefit from higher bandwidth or lower latency that those.

Thank you, could not have gotten here without following the mega-threads you and Lorry left behind

I am going to carry this UEFI in my pocket for everyday use. The screen flings new windows up a light speed, I am not much of a gamer, but it runs cool, steady and fast crunching numbers.

With this DRAM though, there is probably substantially more to wring out for the simple satisfaction of doing it. I have not figured out how to punch up the CPU power to highest levels ... with four banks filled and closing in on the limit of FCLK::UCLK::MCLK feels like I am driving a cruise liner with an outboard engine. I really want to run after the 3800Mhz edge.

Btw, Ryzen calc is a one for one match for Asus ROG bios in regards to memory fields. In other words you don't have to hunt and peck like with other bios'.

Yeah, I am pretty jealous about that. ASRock has me jumping back and forth and translating a bunch of it, made me finally get a three-ring binder and punch holes in my notes and references.
I ran an ASUS M5A99X EVO board for my Phenom 9850 Black Edition and really had a good experience. But with four memory filled, I wanted a t-top board and heard ASRock VRMs can take a beating.
 
Last edited:
Dont show please your ZEN+ settings and results because they are completely irrelevant for ZEN2. Different line, different architecture and memory sub-system structure, different settings and results.
Although those steps up there, are relevant and may apply to any Ryzen as a procedure, but not as actual values

I reread topic title and OP first post twice. Where is i can know that this thread about Zen 2 only? Btw, Zen 2 has better memory subsystem. And any other Zen results will work at Zen 2.
 
No harm, no foul R2! Interesting to see any Ryzen as well.
 
3533cl16-19-16-32 at 1.39V stable on a ryzen 5 2400G with safe settings, except iGPU became unstable, everything else was rock solid. Micron E-die 2x8 sr at Asus Tuf B450m gaming plus. Also undervolted cpu 38mv.

Got 3733cl16-20-15-30 530 tRFC at 1.32V 100% stable with ryzen 3600, Gigabyte B450m DS3H and Micron E-die 2x8 sr. Also undervolted cpu 42mv.

Aida64 with latter setting.
 

Attachments

  • IMG_20200326_095927.jpg
    IMG_20200326_095927.jpg
    5.8 MB · Views: 165
Last edited:
3733 cl16 here (from 3200 cl16) 16-16-16--16-32 at 1.45V
 
Back
Top