4 stick RAM won't boot on X570 Xtreme at rated speeds

[reservoir_dog]

So I've bought 4x16GB dual-rank sticks of G.Skill 3200 C14-14-14-34 rated. It fails to post at rated speeds.

Tried XMP - no result.

Tried latest mobo bios - no result.

Raising voltage to 1.4v - no result.

Lowering to 3000Mhz worked, but why do I have to use it at 3000 when it's 3200 rated and moreover is in the Mobo QVL list indicating the support for 4 sticks.

CPU - 3950x.

[Jurrunio]

Try set SOC voltage to 1.1V. you're in the realm of bottleneck of the memory controller of the CPU.

[reservoir_dog]

decent advice thanks, 1.1 didn't work , but I'll try higher

[Jurrunio]

13 minutes ago, reservoir_dog said:

decent advice thanks, 1.1 didn't work , but I'll try higher

higher is not always better, memory controller performance starts dropping past 1.2V in all samples I've seen

[reservoir_dog]

On 2/10/2020 at 9:23 PM, Jurrunio said:

higher is not always better, memory controller performance starts dropping past 1.2V in all samples I've seen

I was thinking to maybe get 4 sticks of the newer NEO series F4-3200C14D-32GTZN which are supposedly optimised for AMD. Currently I run and have troubles with a bit older Royal F4-3200C14D-32GTRS, though all the parameters and b-die chips are the same..

[Jurrunio]

11 minutes ago, reservoir_dog said:

I was thinking to maybe get 4 sticks of the newer NEO series F4-3200C14D-32GTZN which are supposedly optimised for AMD. Currently I run and have troubles with a bit older Royal F4-3200C14D-32GTRS, though all the parameters and b-die chips are the same..

what if you raise the memory voltage further? B-die scales with voltage. Make sure you give them some cooling as well

[The_Geek]

Just curious, what kind of performance gains would you see going or trying to go to 3200 from 3000?

What are you using the rig for?

[reservoir_dog]

On 2/12/2020 at 5:20 PM, Jurrunio said:

what if you raise the memory voltage further? B-die scales with voltage. Make sure you give them some cooling as well

So I have been able to hit 3133Mhz with SoC 1.07v and DRAM 1.37v. I was even able to lower the timing to 14-14-14-30 tRC-44, with more time I think I will be able to lower those and also some subtimings even more. But the funny thing is no matter how hard I try it just won't boot with 3200Mhz and CL14 stock XMP timings, tried memory voltage up to 1.45 (don't want to go any further for daily usage) and SoC 1.15. No matter how high the voltages are it wont even post to bios as soon as I set 3200Mhz.

 

I was thinking maybe it has something to do with the XMP itself being a bit old and made originally for intel. Could it be that some single weird subtiming causes it.

 

So hence the question, what is the difference between regular GSkill Royal and GSkill Neo (AMD)? Do those Neo AMD kits have some different hardware or is it only settings that differ? I've seen people post successful results and no problems with those neo kits. Some guys even hit 3600Mhz cl16 on 4x16 sticks. If it's only xmp settings then it would be easy to copy those for my ram. But if Neo AMD kits have some extra hardware it's totally different story. Finding info online on this discrepancy in kits is very hard. Everybody just says "they're optimised", but nobody knows exactly how they are optimised. 

[reservoir_dog]

On 2/12/2020 at 6:17 PM, The_Geek said:

Just curious, what kind of performance gains would you see going or trying to go to 3200 from 3000?

What are you using the rig for?

The gain is negligible, but it's a matter of pure interest and getting what you pay for. This is a rendering/gaming machine.

[Jurrunio]

7 hours ago, reservoir_dog said:

So hence the question, what is the difference between regular GSkill Royal and GSkill Neo (AMD)? Do those Neo AMD kits have some different hardware or is it only settings that differ? I've seen people post successful results and no problems with those neo kits. Some guys even hit 3600Mhz cl16 on 4x16 sticks. If it's only xmp settings then it would be easy to copy those for my ram. But if Neo AMD kits have some extra hardware it's totally different story. Finding info online on this discrepancy in kits is very hard. Everybody just says "they're optimised", but nobody knows exactly how they are optimised. 

Should be only the settings, though it could also affect the sorting because the testing procedure of their memory dies could change to better suit Ryzen's memory controller.

 

then again, unable to go faster but could tighten timings is a clear sign of memory controller or board bottleneck. Board thing happened on the Gigabyte X470 Gaming 5 and Ultra Gaming for example

[reservoir_dog]

18 minutes ago, Jurrunio said:

Should be only the settings, though it could also affect the sorting because the testing procedure of their memory dies could change to better suit Ryzen's memory controller.

 

then again, unable to go faster but could tighten timings is a clear sign of memory controller or board bottleneck. Board thing happened on the Gigabyte X470 Gaming 5 and Ultra Gaming for example

Yep, Aorus isn't really known for good memory support like MSI

[Jurrunio]

9 minutes ago, reservoir_dog said:

Yep, Aorus isn't really known for good memory support like MSI

but the flagships are fine, even on X470 the Gaming 7 isn't bad at all.

[reservoir_dog]

On 2/14/2020 at 9:12 PM, Jurrunio said:

but the flagships are fine, even on X470 the Gaming 7 isn't bad at all.

So I've managed to hit 12-12-12-28-42 at 3133 with relative ease, zero freezes zero glitches, some extra sub-timings from DRAM calculator lowered as well. Kept voltage at 1.45v to have some headroom and not to worry about it just now, SoC on auto and hits around 1.075. System stable under AIDA stress, had no time yet to try full stability suite of memtest86. Temps are just fine at around 42C. I may even try my luck and push the timings further. But frequency wont move above 3133. This definitely looks like some mobo/ mem controller or ram PCB layout issue. 

 

I mean it does look good so far, not sure whether switching these Royal kits for newer AMD optimised Neos would be beneficial. Royals are allegedly better binned to start with.

I may hit 3600 c16 or 3200 c14 right of the box with Neo (which I can't do with present kits), but would that be better than current 3133 c12? Overclocking 3600 to c14 or lower sounds cool.

 

I've read materials that say that generally higher frequency + higher timings is more favourable than lower frequency with tighter timings, for most applications. It's marginal, but it's a matter of interest.

[Jurrunio]

5 hours ago, reservoir_dog said:

I mean it does look good so far, not sure whether switching these Royal kits for newer AMD optimised Neos would be beneficial. Royals are allegedly better binned to start with.

I may hit 3600 c16 or 3200 c14 right of the box with Neo (which I can't do with present kits), but would that be better than current 3133 c12? Overclocking 3600 to c14 or lower sounds cool.

Or you could still very much hit the wall because the memory controller is what actually taking the blame.

 

You could try overclocking with only 2 sticks, with 32GB it should clock much higher if it's a memory controller problem.

 

5 hours ago, reservoir_dog said:

I've read materials that say that generally higher frequency + higher timings is more favourable than lower frequency with tighter timings, for most applications. It's marginal, but it's a matter of interest.

You can claw back some of the performance loss by desyncing FCLK and raise that up till it's unstable. When synchronized FCLK will be half of memory data rate (so 1567 for 3133MHz memory), but Zen 2 CPUs can do anywhere from 1733 to 1933MHz stable depending on silicon lottery.

[reservoir_dog]

@Jurrunio now it gets even more weird. So not a single time a boot was successful with 3200, under any possible combination I tried - 3200 was not achievable, I never went higher, reckoned that to be pointless, guess what, just for fun I set the system to 3266 c14 and it worked.

Looks like my testing is not over. Will also try to decouple FCLK, but frankly not sure whether that's a good idea.

[Jurrunio]

9 minutes ago, reservoir_dog said:

Will also try to decouple FCLK, but frankly not sure whether that's a good idea.

Only decouple when FCLK is stable at least 200MHz higher than if it's coupled.

[reservoir_dog]

@Jurrunio  I've hit 3733 c14-14-14.. that is kinda cool considering it's 4 double sided sticks... but failed to post with 1.45 on memory.. raised to 1.47.. too close to the thresholds, will need to step down a little for daily usage and run some stability suites. Still have no idea what causes boot fail on 3200, seems like some odd bug.

[reservoir_dog]

 

Gave a pretty good explanation of AMD optimized