![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
Man, look at you go! That's tighter than a >insert something inappropriate here relating to nuns<!
Did it bench with any better speeds and latency?
Hmm not helpful, but I'm running a 2x16gb cheapo 2133 hynix ddr4 too, but on X99. This one was able to get to 3200mhz 16-19-19. It could kind of run at cl15, but it wasn't 100% stable.
Well if that's significantly lower than before, then hot damn! (I'd qualify 2ns as definitely-significant, but even 1ns as significant)
Just built my Ryzen system over this past weekend.
Good stuff, have you been able to tighten any timings further? (They are already tight, but why stop there?)
Lord knows Stilt knows way more about hardware than I do, so the differences between my timings and his may end up being of no consequence in terms of latency and/or bandwidth... That being said, you can see mine on page 5, which there are a number of subtimings on his that are a fair bit higher. On the flip side, I have a couple that are similarly a fair bit higher as well, and that may be the difference in why he is able to lower other when I wasn't able.
I have not attempted any further timing adjustments. Not sure I'll even bother trying since the gains likely won't be worth the hassle for me.
I wish I knew more about squeezing out performance on RAM to be honestSome important factors though would be to let us know what model kit you have and then if they're Single or Dual Rank sticks (CPU-Z tells you in the memory tab).
I have a feeling that they're Hynix ICs though, based on those timings (if you are really interested in finding out for certain you can do so with Typhoon Burner, I posted how to in the old Ryzen memory speeds thread if you need assistance finding it). So I'm not specifically sure how much luck you'd have with my suggestions in terms of how well they'll carry over from my Samsung B IC sticks.
That being said, are those timings currently with everything set to Auto, or did you manually input them? (I only ask because they're, curiously, identical to Duron's lol).
If not, that'd be my first request, is a screenshot with auto-timings. That's our baseline, and lets us know where we're starting from. Otherwise I don't know how far you've tweaked them already and one tick more might equal instability
As far as overclocking the RAM... If you haven't tried faster speeds with Auto-Timings, that's my initial suggestion. Bump the SoC voltage and DRAM Voltage a little (see AMD guide linked below for safe limits) and try 3000 or the next speed higher. For 3000 you can probably just try it with those timings you're at currently, but higher speed, definitely try with Auto-timings. But when it comes to changing tthe termination values, I'm not the person to ask
But to at least offer something to start with...
First you'll want to have a memory benchmark so you can determine if you've increased performance. So either with the RAM configured as it is in that screenshot or with auto-timings, get a few passes on that to get a baseline. I used AIDA64's CacheMem test (it's in the Tools drop down menu), since it gives the Read, Write and Latency results for RAM, L1, L2 and L3 caches. So it knocks them all out rather quickly AND lets you easily save a screenshot of it.
I'd then try Disabling the GearDownMode, give it a quick stability test (either MemTest or AIDA64's System Stability Test set only to "Stress Cache" and "Stress System Memory", which I found identifies stability fair well). Then run the benchmark to see if it helped, and how much. If it didn't really help at all, then might be better to leave it on to ensure stability.
Next I'd see if you can try to lower the tRRDS and tRRDL. You could try 4 and 6. Re-test.
I'd try then to drop the tFAW, maybe 28 (mine's at 25). Re-test.
Seeing if you can knock that tWR down as well will help a decent bit. Just whittle it away slowly, but could try 20 to start with. (mine's at 11). R-t!
As I recall the tRDRDSCL and tWRWRSCL don't impact stability much, but can help a tiny bit with latency, an can be dropped fairly low. Mine are at 2 and 2, so maybe try 4 and 4 with yours.
If you can get the tRFC down then it helps with latency mostly (I think, maybe it was bandwidth). I don't know if there's a method to calculating the number though
GENERALLY LEAVE tWCL ALONE. Is finicky and can impact stability. Typical rule of thumb is 1:1 with tCL
tRTP, tRDWR and tWRRD can result in decent gains, but you may only be able to lower tRTP a tiny bit without impacting stability. Could try 10 or 9. (mine's at 9)
Beyond that, you can definitely always try the primary timings. However, until they allow for odd numbered tCL's again, it might not yield much stability since you'd have to drop back to tCL 14 for it to work. However, you can definitely give 16-16-16-36 a shot, or even 14-16-16-36. Those may require increasing the DRAM voltage in order to get working. I mean realistically the subtimings could even benefit from a little bump. I'm not sure what you're running at currently, if it's like most of the GSkill kits, for 3000 it might just be 1.3V, but either way just start with adding a little bit... so like +0.02V (1.32V in this example). But I bought a CL15 kit so for me to get CL14 working is more likely in my eyes, than you having to go to the next even number down which is also 14.
Something always to keep in mind, though, is that a lot of those timings have a 'mate'. They're grouped pretty well in RyzenTimingChecker and so you can see that rather easily. Not always do they require the name initial letters though, like tRDWR and tWRRD are kinda mates, but tRTP plays a roll with them too. So you'll want to pay attention to the amount of distance between their numbers and then try to keep a similar gap in your adjustments. Those three it's a situation of (bottom to top) WRite to ReaD, ReaD to WRite, and Read To Precharge. So if you drop the tRTP down too low, it'll start wanting to do it's task before tRDWR has had a chance to finish.
This was a handy guide from AMD as it explains some of that, as well as a lot of other aspects:
https://community.amd.com/community/gaming/blog/2017/05/25/community-update-4-lets-talk-dram
This, albeit rather old, may help fill in some gaps as the terms still apply (note: not all of these will be available to you, and some may even have different letter combinations in your BIOS, even between RyzenTimingChecker and your BIOS!)
http://www.overclock.net/forum/18051-memory/381699-ram-timings-explained.html
Hopefully that'll help some. And yes, as it may sound, it is all time consuming, since you have to restart, adjust, boot, test, restart, adjust, boot, test... over and over
I think everyone should have access to as much info as possible and we all share it since we're all enthusiasts. Sadly that will never happen as long as there's money to be made
Regardless, I'll do what I can to try and help, even if what I know pales in comparison to what others may know Try the same in HWinfo64. I've seen this happen to others when trying to use HWmonitor.
Not much for me to add that Neapolitan6th hadn't already said, but I'll just add that I hate HWMonitor with a passion and find it to be completely useless due to how buggy it has always been for me.
My method of putting the paste is the same “grain of rice” on the middle of the processor.
For ram,test with memtest86 or memtest 86+ (4 passage)?
i just use memtest, cant remember if its + or not, either works. i only run it for an hour or two them move on to other testing and using it normally.
This doesn't work, it crashes
then i guess stay where you are with speed or you could try turning off xmp and manully set it to the xmp speeds. xmp is technically intels spec and doesnt also play nice with amd.
