Nice!
1.50V can or not?
Try to match this B-die DR kit https://www.gskill.com/specification/165/29...B-Specification

Unfortunately cannot, got error



My kit is only rated at 3600MHz 14-15-15-35 @ 1.45V
So the G.Skill 4266C17D has to be higher bin.

Another tweak. Trying to 'fix' the RTL value because I noticed that upon cold boot, the RTL may change between these 2 values: 45/45 or 45/46
If boots up with RTL 45/45, then passing TM5 memtest. However, if it boots up with RTL 45/46, will fail TM5 memtest with single error on test #2. No other parameters change except this.

Strangely on this mobo, I can't seem to change the RTL value directly. Have to set RTL training to enable under memory training page to get the RTL value to be fixed. And with this, I got a lower RTL value of 40/41 instead.

With this lower RTL value, I need to bump up the RAM voltage a bit more to 1.57V to be error free. Good thing is, I got to shave off 2ns of latency.

Yeah you're right, manual RTL doesn't work on Z590.

So it's a known issue for Z590?

Yes, not a big problem as you can enable that RTL training in algo page like you already done.

Lowest latency I can get:



But not stable though... just for the benchmark number

32GB? try go command rate 2T. probably can go higher. Yours looks like it's limited to the IMC.

I'm running Gear2, so naturally the command rate will be 1T. Never try 2T on Gear2 before, I bet the performance will suffer.

Just found out that on RocketLake CPU, *_dd and *_dr is reversed compared to CometLake.
Since I'm running only 1 stick of RAM per channel, I should be able to drop *_dr = 0

Getting pretty much similar AIDA number compared to previously when *_dd == *_dr ... noticed latency drop just a little bit, perhaps placebo effect

ah right, forgot you're on 11th lol.
maybe up the trfc and loosen the RTL. you dont really gain very significant latency from there. better to tighten the primary first then only push the secondary.

This post has been edited by Bonchi: Jul 7 2021, 11:43 PM

Urghh... unfortunately this mobo doesn't allow me to hand-tweak the RTL/IOL. Not 100% sure if it's mobo problem or 11th gen problem or combination of it.
I can only force the RTL to change by enabling RTL training BIOS setting.

cant hand tweak but i think owikh shown before in some older posts where to set so that you can set a base for the memory to train from. Well you mobo is a pretty good one. such high read lol. Mine only manages read 61K but my write is at 66K (MSI crappy bios).

odd pairing

Unfortunately I didn’t manage to get it to work. Either default or RTL training enabled with lower value.
Any manual value that I put seems to be ignored.

Care to enlighten which one is odd?

well, generally it's better for the bios to train the values anyways. Dont think Asus will retrain like MSI which will ruin all the effort at random.

Well, as explained by owikh before. the trained RTL value is affected by SA/IO/DIMM voltages (too high or too low) and all the timing values... so if not done right, the RTL IOL will be weird. Thus it can be a sign of instability.
you do have some odd settings such as the 4 TRDWR values ideally should be all same, like 12-12-12-12.

i think what ctskl say weird is that although you fulfilled the rule to not be more than 1 apart but it normally looks like this:

CHA D0 40
CHA D1 41

CHB D0 40
CHB D1 41

cant see your IOL to judge further.
Maybe you can copy some of those sifu's values and try. Your mobo is pretty similar to theirs, maybe can even ask for a bios profile. Im pretty sure they are willing to share.

also bdie you can get away with lower SA/IO/DIMM voltages so sometimes having them too high following from your previous DJR settings, it may cause instability. try 1.4v-1.45 for 4266.. also Bdie is quite temp sensitive so the quality of the ram's heatsink matters too (which is why TridentZ does quite well).

This post has been edited by Bonchi: Jul 8 2021, 11:34 AM

The RTL values are from the MRC training algorithm. I just assume it’s optimized and I can’t manually override it anyway.
If I set tRDWR_* to be all 12, I got error in TM5, so I increase some of it.

And you’re right that I’m running higher voltage than usual B-die kit needs. So I’m going to retune everything on lower voltages later.
Also, my RAM is definitely running hot as my thermal probe reads 51C at heatsink area so memory chip will definitely be higher. Not enough space to mount a fan, unless removed side panel and get a fan to blow it directly - not practical

yeah this kind of settings have to find sweet spot, cannot simply brute force. especially by cranking up voltage.

Sometimes your kit will also will not like certain ranges. Like my B-die struggle to get to 4133 17-17-17-37 as that was my initial target. already put voltage to 1.5V but it just cannot post... brute force the training then will be super unstable anyways. But then 4266 17-17-17-37no issues at 1.4V rock stable.

maybe you can try skip 4266 to 4500 to tune.

Now I’m trying similar as yours: 4266 17-17-17-37 @ 1.5V and loosen a bit secondary/tertiary timings.
So far looks doable, need to run all tests to ensure stability. Can’t do lower than 1.5V else will error out easily.

Previously I did try higher speed like 4400 and 4533. 4400 still possible but need to relax the timing quite a lot but 4533 no boot. So I think 4266 is the best bet for long term stability.