RAM Overclocking on Clevo

I'd start by testing limits with stock VCCSA and VCCIO, no need to push additional voltages if you can avoid it.

It always begins with the nvram reset, so any prepopulated values are wiped. On first boot the only thing you change is the profile to "custom" and keep whatever timings the machine set at 1.2v. Let it train, boot into windows and do a shutdown from OS. Then go up a multi. Should it fails at a higher multi, raise the voltage to 1.25v

This kit would train going from nvram reset directly to 3600 at 1.2/1.25v, giving it more voltage would make it fail. Only way to get it to train with higher voltage was to tighten the timings, which seems illogical since you normally loosen up the timings to try and find the 'speed limit'.

As for training, give it 5-15min to train and boot loop, if it fails > shutdown and clear nvram. Keep notes of multipliers and voltages you test. ( i have 3 pages of notes from trying different combos, before hitting something that actually would boot)
I found that my particular chip either trains instantly or after 3-5 reboots, after that it would infinitely boot loop.

Oh and remember to run the sticks in s/n pairs with dimm1+dimm4 and dimm2+dimm3

 

Well, there's one mistake glaring I made. I was using the XMP voltage setting from the start of 1.35v. I was running them in s/n pairs then single sticks but I know some runs were not slot specific except favoring slot 1 for single sticks, but I'll go back in next week and start fresh with a complete kit training session as outlined above. I always keep notes in a notebook when testing, so I'll go back over the pages to fine tune it and get ready for round 2. I just need my laptop up and running while my desktop goes through breakdown and testing hell.

 

Still got to keep it in mind

 

Does anyone knows the default values for VCCSA and VCCIO?

 

Should tell you when you default, something like 1.1v and 0.95v IIRC.

 

That's exactly it with a 1.2v and 1.5v cap.

 

This is where I am at the moment:



Not had too much time to play.

 

I just want to set them to the default value, since not overriding those voltages can lead them to spike when overclocking the RAM.
Although I think I can't push my Balistix kit beyond their rated spec of 32000Mhz. My IMC must be weak or the mobo, one of the two, since even running them at default values is hard :|

 

On my 9750H system nothing above 3100Mhz is stable, doesn't matter the voltage, timings or if they run 3600Mhz with other CPUs, the CPU internal memory controler just won't play ball.

@Mr. Fox can run 3200Mhz on an almost identical board with just another CPU sample.

Yep, usually push a stick into TM5 error territory and then swapp it around slots to categorize them. Then do the same with each stick and put the best sticks into the weakest slots.

That is a very impressive result for those sticks!

These are my final daily driver settings of Sammy B-die. Took a while to get them solid, but now passed 7:30 hours on Memtest86+ and TM5.

Spoiler

[\SPOILER]

 

Impressive. Can you tell what VCCIO, VCCSA and memory voltage you are using?

 

Very nice @Prema I'm going to have to play some more.

 

They are all at stock (BIOS defaults for VCC and 1.35v for the RAM).
I really only ever overvolt anything when benching...

 

All the volts all the time but nice results for stock voltages.

 

What performance are you getting in aida64 cache bench brother?

 

Had a little more time to play, I also doubled my refresh cycle count. Testing for stability.

 

My system is disassembled and the heatsink is baking as per @ole!!! and
your guidance...

 

It will prolong the life of LM application for sure. Perhaps we will see nickel plated heatsinks with time.

 

I reasonably easily tweaked my PowerSpec-1530/PB50DF2's 2x16GB SK Hynix (crap...?) 3200MHz 22-22-22-52 to 19-19-19-43. This was my first time tuning RAM so I got a little too excited and short-sightedly changed too many settings at once and she would no longer POST. That's how this game is played though... no biggie.


My problem is, my Clevo can apparently only clear the CMOS if I remove the battery. Fn + (D or G) + Power has no effect on the CMOS.

So, any pointers for making my life easier while tuning ram? Maybe some other key + Power combo? Maybe some jumper I could rig up so I don't need to keep my keyboard detached (to remove battery) while inevitably pushing my RAM too far?

 

On the extra heatsink I have lying around, I didn’t bake it because I didn’t want the foam padding on the fin stacks and heatpipes to fall off, but I spread LM into the copper and let it sit for a few weeks. Same with the tall BartX IHS. Next weekend when I install my i7-8086K and the IHS, the LM should hopefully be fully absorbed into the copper.

I’ll also take the opportunity to swap my Corsair Vengeance B-die sticks into slots 1/4 to see if I can squeeze some more OC out of them!

 

I have to push my IO and SA voltage to 1.2v on each for the above to make it boot happily every time but it's being reliable so far.

 

Hi@all,

could you please help me regarding XMP Profiles ?

I found two XMP Profiles Posted on NBR specially for the P570wm.
One is for HyperX impact 2133mhz
The other is ripjaws 2133

I can’t find exactly that sticks anywhere to buy. Could i use the HyperX Impact Profile on any hyperX impact stick even higher MHz ones with different CL?
How about the Ripjaws ?

i currently have 1600 hyperX running 2133mhz stable. I’d like to go 2400mhz with thight timings.

please help me.

Thank you

 

No luck with slots 1+4 on my system. XMP-3000 wouldn’t even train, let alone my custom settings. Looks like slots 3+4 still work best here.

 

It is interesting and shows how much variation there is in system to system.

 

I'm currently @ 3466 CL17 1.25v TM5 0 errors at stock settings on this Corsair 3800 CL18 kit stock VCC. 1.20v spits out errors left and right.

I'll push for 3600 next and see where it taps out with this 9900ks and then do it all over with the 9900k p0 if it's still here and see where it goes.

 

3466 seems to be the limit again with the 9900ks and 3800 kit just like the 8086k and the 3600 kit I had before except the 3800 kit is hitting the same timings and speed at 1.25v vs 1.45v.

The 3800 kit now will boot right away @ 3466 1.25v which is a plus.

Ready to give the 9900k p0 a full run next....it's ready w/ the re-re-lidded BartX 4.3 and a lovely blob of Nanogrease.....

 

My new BIOS (thanks @Prema has moved my sweet spot a bit so I'll have to play with my secondary timings again.

 

From what I remember b-die is temperature sensitive, which is why I began running MemTestPro to 400% to see if stability held after hours of testing. Just something to keep in mind, but at 1.25v I doubt it will be a problem, since even at 1.45v the stick remain quite cool.

Also check the performance scaling, even if you hit frequency wall, you can tighten these kits up quite well. Looking forward to the P0 numbers!

 

They will hardly be warm at 1.25v.

 

4-6 hours memtest will show

 

Eh still at 1.25v.

 

I let them run overnight memtest from the usb bootable version and no problems at 3466 17-17-17. From a fresh nvram reset they go from stock 2400 to 3466 1.25v quickly now.

They will not boot past 3000 in numbered order pairs 24+25 (slot 1 / 4) and 26+27 slot 2/3) no matter how long i clear and retrain.

They hit 3466 slot 1 = 26, slot 2 = 24, slot 3 = 27, slot 4 = 25.

9900k stalls at 3600 but based on the previous stalls at 3466 with the Corsair 3600 kit on both my 8086k and 8700k I was not surprised. I might try some other configurations, but 4 CPUs and two memory kits the writing might be on the wall for the system board and 3466 being its limit.

For now I’ll focus on tightening up the timings.

I forgot how much warmer the 9900k p0 runs in here versus the 9900ks r0 even with only a shim modded 9900ks vs BartX 9900 p0.

 

Yeah once you find what insta trains, makes everything much faster for tuning.
That's pretty much how I run mine as well, since you can either do 1+2 & 3+4 or 1+4 & 2+3 do get dual channel to work. I had the opposite where R0 runs hotter than my R0, but that could be do to silicon quality.