6000 36-36-36-96 EXPO subtimings

[GrimReaper]

Hi,

I have Gigabyte X670E Aorus Master with latest F8a BIOS.
I also have the XMP version of 6000 36-36-36-96.
But I tried Karhu and it errors in first minute.

These are the XMP default subtimings:
tRAS - 96
tRC - 132
tWR - 90
tRFC1 - 884
tRFC2 - 480
tRFCSb - 390
tRTP - 12
tRRD_L - 15
tRRD_S - 8
tFAW - 40
tWTR - 30
tWTRS - 8

Can you share the EXPO subtimings?
Do I need to up some motherboard voltages? Now I have only DRAM at 1.35V and everything else default/Auto.

[GrimReaper]

I just remembered board has its own Auto settings.

tRC - 146
tRTP - 23
tFAW - 32 (actually lower than XMP)
The rest are the same as XMP.

Karhu 10min and going.

[GSKILL TECH]

Do you have memory in slots two and four away from the CPU? Have you tried each module to see if they can be stable at DDR5-6000 on their own? Or do you know if a lower DRAM Frequency such as DDR5-5600 or 5800 can be more stable? The results should provide more insight to where the limitation may be.

[GrimReaper]

Slots 2 and 4, yes. I just presumed it is the same as Intel, didn't care to look at the manual.

It is stable now, Karhu 30min+, then I stopped.

These are the subtimings/voltages from SPD Setup in BIOS:
Frequency - 6000
tCL - 36
tRCD/tRP - 36
tRAS - 96
Vdd - 1.35
Vddq - 1.35
Vpp - 1.80
Vimc - 1.35
tRC - 132
tWR - 90
tRFC - 884
tRFC2 - 480

Interesting that tRTP and tFAW are not even in XMP, yet board sets them different than Auto if I select XMP and reboot.

Here I found EXPO timings for 6000 CL30:
https://www.geeknetic.es/Review/2428...30-Review.html

Looks like pretty much standard timings.
So I would say XMP and EXPO are exactly the same for 6000 CL36, if even 6000 CL30 EXPO is the same as 6000 CL36 XMP (apart from first 3 primaries of course).

I don't think it matters much which one is the culprit, but I would say tRTP 12. Board simply sets it too low once you select XMP.
I would test this, but BIOS is so slow I am happy to be in Windows again. I don't think original BIOS was this slow, so I have to wait for new BIOS before venturing into lowering subtimings.

[GrimReaper]

It's not too low tRC and tRTP. What I think happened yesterday, I was lowering some subtimings (no luck though below XMP, only BSODs), then when I loaded XMP immediately after that, I think what happened is the RAM was still too hot or whatever, and it errored. Once settled (at least 5-10min I guess), it worked fine again.
I have some vague memory about this from DDR4 overclocking when I had the same problem. You need to let them cool down.

I already lost all interest in lowering subtimings, it's a lot of hassle. tRC 120 BSODs, same with tWR 70. So it will be mostly down to tRFC and tFAW.

[GrimReaper]

Stability is pretty much the same on 4800 CL40 as on 6000 CL36, both 10-40min in stress test.

I think RAM is overheating. One proof is that this morning and several before, everything worked fine. BIOS was not sluggish, no BSODs.

But going from XMP to 4800 CL40 (with Load Optimized Defaults) makes even more problems, two times it was like this:
first boot - BSOD before Win or at login screen
second boot - stuttering sound
third boot - BSOD before Win or at login screen
Fourth boot usually works, but one time I still had sound problems.

BIOS for Aorus Master X670E is best the latest - F8a. XMP works without problems.
I tried F7 yesterday and it is the same as F6 that came with the board. 4800 CL40 works, XMP 6000 CL36 doesn't boot, it requires CLEAR CMOS. Then I tried also 5200 CL36 and it boots, but hangs on black screen after restart, so another CLEAR CMOS. And I flashed F8a again.

So before RMAing RAM, I could try placing 120mm fan on top of the case, pushing air towards RAM.
One stick is also a good idea, but that doesn't prevent me RMAing RAM.

[emissary42]

So before RMAing RAM, I could try placing 120mm fan on top of the case, pushing air towards RAM.

That I would always test if you are suspecting you might have temperature issues. It is also a fairly simple and quick thing to do.

[GrimReaper]

Left stick was bad, Karhu errored in 1min, right one works now Karhu 1h 40min, Coverage 6700%.
That was a good advice.

Fan didn't help much, before average 51°C, with fan 49°C. It's hard to mount fan to cool RAM anyway. On DDR4 I even bought the fan holder (like a stick you can curve), fixed it in 5.25" drive cage (the stick, fan was close to right RAM stick). Before that I just put it in drive cage, but then it is too far away.
Now I put it where there is place for AIO or exhaust fans on top of the case (be quiet 802, before Corsair 600T), but I made it intake. But 120mm positions right above RAM, pushing air more into CPU heatsink than into RAM, so I would need 140mm to actually cool RAM.

This one I will RMA. But in the meantime I will buy the 6000 30-38-38. I wanted that one anyway, so this is a good thing for me.

[GrimReaper]

I got Neo 6000 30-38-38 yesterday. Here right stick is bad. I always arrange them by serial number, lower number in 2nd slot, that way I don't have trouble knowing which is which.

But it's not bad meaning it doesn't work, it just isn't stable on 6000. So far it seems 5600 is stable. I have to check the 6000 CL36 kit's bad stick again, if 5600 is stable.

I don't even have to say I am very disappointed in G.Skill's QC. I am 100% sure they don't check every kit for stability on rated speed. Because my 2 bad sticks error out between 1min and 2min.
DDR4 was way better, 6 out of 6 kits were problem free, 3 kits are in use to this day.

[GrimReaper]

I have to test further, but so far (25min+ on the worse stick) it seems 1.36V will be enough for Neo kit to be stable. Even the better stick actually BSOD-ed after 1h on 1.35V, so there is not much difference between both sticks.

It's probably same for 6000 CL36 worse stick, one stick simply needs more voltage. But there the better stick works 1.31V and that will already be 0.05V difference. I'm not happy with that, but I don't think it's worth RMA for that, I can get similar thing again.

As far as my little experience with different boards, I did go from Z490 to Z590 and noticed it needed +0.01-0.02V. And especially with X570 which needed +0.07-0.08V more than Z590 and tRCDRD 16 instead of 14.
So board certainly can make a difference. But still, G.Skill should stop making such tight bins that board difference even comes into play. It was same with 4000 16-16-16-36 1.40V, was stable only on 1.40V and not on 1.39V.
Before these 3 kits 0.05V leeway was nothing out of the ordinary. For example 3600 14-15-15 1.45V worked on 1.37V. This is better for manufacturer (way less RMA) and for the customer (way less trouble).