That is normal, as you step up the number of memory cells voltage drops and you cannot maintain the same settings. That is why they do kits - because they at least run the QA tests.
Personally I would not go over 2667 with 256GB unless I am really ready to do the following:
Extensive tests under load.
Going back to BIOS several times to hammer down the settings
This one is important - power off the computer at least once a day.
No machine that runs continuously should run memory OC. preferable even go for ECC.
I havenāt done hard battery reset, but i have held down the clear cmos button on the back of the io shield. It definitely reset bios, but I will try a hard reset as well.
Stress tested half the 256gb kit, and I was able to pass memtest even with an overclock from 3200mhz to 3600mhz.
Tossed back in the rest of the kit for a total of 8 dimms, memtest will either freeze, stutters hard, generates errors, or shuts down pc.
Tried downclocking to 2666mhz, which according to Wendell should be supportedā¦ But I Still have the same issues. Is hard stuttering/lag during a memory stress test with a 3990x and 256gb ram normal? Is there some voltage or setting I can tweak to help the stuttering and get 256gb at 2666mhz possibly stable?
Gigabyte hasnāt updated the bios in months unfortunately
Itās just a trick for clearing the file cache. On machines with a big pile of RAM they tend to get the entire boot disk into RAM cache eventually. Then if you get a bit flip in RAM that DLL or font file or icon or whatever will be broken until you reboot.
Okay I set everything to defaults, upped soc voltage a bit, canāt ever get memtest to work without freezing but hci memtest can run without pc freezing or shutdown!!
ā¦ However it will produce 1-3errors within the first hour.
Tried downclocking to 2666 mhz and I still get errors.
This must be a motherboard issue at this point for sure right?
Wendell, how did verify stability with the 256gb kits with the 3990x in your video?
On the same Gigabyte board?
Or was this on the Msi board?
I mean untill this part itās a little bit blurry to me, which exact motherboards you have. And what you actually tested etc.
You firstly need to isolate the variables.
That is the proper way of troubleshooting.
I mean if your system with the same Gigabyte motherboard and same 256GB memory,
kit worked totally fine with the 3970X cpu.
Then the problem is likely not with the motherboard.
The variable left in that case is pretty much is the actual cpu.
Especially if you say that the memory test passes with half,
the said memory kit ( 128GB ) on the same board with the 3990X
But like i said to me the story is all a bit vague and blurry up till this point.
So if you could, please put up a sum up of what you actually have in terms of hardware,
and what you already tried and tested.
We firstly need to start isolate things.
Sorry I know itās all a bit confusing because of all the posts and updates.
Okay here is a summary:
Started with 3970x and 8x32gb ram on gigabyte board. Should have ran memory stress test then but didnāt. System worked great to my knowledge.
Decided to buy a 3990x and plop it into the gb board. Also bought a cheap msi board for the 3970x and used a different 8x16gb ram kit for it.
3970x now runs fine.
Problems with the 3990x begin. System shut downs on load above 2933mhz. At 2666mhz, no more pc shutdown, but memtest has occasional errors. But, If I halve the ram to 4x32gb and keep frequency at 3200mhz, it works!
Further testing:
If I oc the 4x32gb to 3600, it will eventually shut down on load after an hour.
If I throw in the old 8x16gb kit into the 3990x, I can run 3200mhz, but 3600mhz will eventually cause shutdown. If I halve this kit to 4x16gb, I can run 3600!
I also back tested my 8x32gb with the 3970x and msi board. Memtest produced a few errors, but no pc shutdown. Again, I can run any of this ram in half mode as 4x32gb on any of my boards with no issues whatsoever.
My conclusions:
Thatās why I began to suspect the lower tier trx40 mobo as not being able to support the 256gb kits on a 3990x that g skill validates only for the zenith extreme. I can see a pattern emerging where higher frequency tends to cause these cheaper boards to have issues. My explanation for why I didnāt have pc shutdown issues before was because the board was less stressed dealing with 64 threads as opposed to 128 threads.
I have a zenith extreme alpha on the way so Iāll know soonā¦
Yeah well my next suggestion kinda was to pop the 3990X,
into the Msi TRX40 pro board.
And then try to run a memory test only with both kits.
To see what that does.
However the vcore vrm of that Msi board is likely,
going to run very hot when stress testing that 3990X.
But you could give it a go for just a aida64 memory test, and monitor the board temps.
But if you already ordered the Asus zenith Extreme 2 Alpha.
Then it would be better to wait for that.
Because the Msi trx40 pro board vrm isnāt the most powerfull out there.
I kinda still think that the cpu might be the actual cull prit.
But yeah it might be worth a try.
Yeah I was going to do what you suggested, but realizing that the vrms are different and knowing that itās such a cheap board it may fail stress testing for the same reasons as the gigabyte board and I may wrongly assume itās the 3990x
I mean if it was the 3990x, it should have failed when I ran the other tests, like 8x16 at 3200mhz, right?
Threadripper is based off epyc, which should work for terabytes of ram, so the variable seems to be the cheap motherboards, which both failed with different cpus with 8x32gb. (gigabyte shutdown with 3990x, and msi errors in memtest with 3970x)
Well yeah the Gigabyte TRX40 Aorus Master is actually a 16 phase vrm design.
With 70A infineon powerstages.
ItĀ“s the same as the Aorus Extreme and the Designare EX board.
Those should basically handle a 3990X just fine.
However the Gigabyte TRX40 Aorus Pro in that regards,
is only a 12 phase with the same 70A Infineon powerstages.
So this board is indeed more suitable up to a 3970X.
The same pretty much counts for those Msi pro boards,
and like the Asrock TRX40 Creator as well.
But the memory part is of course a different story.
Because that is kinda heavely depending on the imc.
The memory vrmās on all TRX40 boards are sufficient really.
Yeah, but Iām learning theres more than just vrm for cpu too. Seems like Different boards can support different memory speeds, due to better construction, less signal losses from ram to cpu, etc. The zenith seems to have the best support.
Matter of wait and see.
But i kinda think that the cpu might be the problem or the imc rather.
Still at least then you have a very good reason to contact,
your seller for rma of the cpu.
You can point out that you tested 3 boards then.
But yeah 256GB memory kits on Threaripper can be fidly.
You are definitelly not the only one struggling with it.
Like i posted above, 3600mhz on 8x32GB Hynix modules,
is likely not gonna happen.
Just wanted to update this thread. I went ahead and ordered the ram kit that Wendell used in his videos, 256gb 3600mhz gskill to go along with my asus zenith extreme alpha. Once both arrived I reinstalled my 3990x from my old board.
Everything seems to be running smoothly now, 256gb at 3600mhz, so far passed 400%hci memtest with no shutdown. Going to let it go to 1000% and call it a day.
Pretty sure in the end it was the gb aorus master mobo not being able to handle 128 threads and too much/too high frequency ram. The reason I say that is because it would power off when running either 8x32gb kit >2600mhz, or 4x16gb > 3200mhz.
My advice if youāre looking to run 256gb of ram and the 3990x. Get the zenith.
*only slight issue Iām having is that reported dram voltage is 1.32v, whereas I set in bios to 1.35v. Which is correct? Iām also experiencing warm boot issues, where if I restart, occasionally it needs to shut down and reboot. Presumably memory training issues. I raised dram voltage to 1.38v, which reports as 1.36v. Hoping that fixes the issue.
I have ths same issue with the ram voltage. Even measured it with a dmm and the motherboard reads right. It just doesnt hit the 1.35v it is set to. Its stable for me at 1.32v though.
