F4-3333C16Q-16GRRD can't quite get 100% stable

[porina]

I might be closer to stability with the Ripjaws 4 at 3333! Still at 2 modules for now. With the errors easily reproduced in test 9, I tried messing around with voltages.

So, at XMP timings, 1.35v, I had lots of errors in test 9
Upped ram to 1.37v, the number of errors dropped a lot. Not bad.
Drop ram back to 1.35v again. Then changed SA from 1.288 measured, to 1.05 set. Errors still low! 1.05 is supposed to be the system default when not overclocking.
Ok, drop the IO voltage from 1.272 measured to 0.95 set, which again is the default. No boot.
IO up to 1.00, lots of errors. Same at 1.05.
IO up to 1.10, 1 error after 90 seconds.
IO up to 1.15, more errors again. So it seems around 1.10 is the sweet spot for IO.
So, IO back to 1.10, let's try SA up to 1.10. It took 14 minutes of test 9 before I got an error. So close...
SA at 1.15 now and errors after 2 minutes. So, maybe worse again.

So like other places have said, there is no magic SA and IO voltage, and too high can be as bad as too low. At least I've seen that with IO, and working on seeing what SA does.

Running out of options here. I don't want to be adjusting 0.01 at a time for either IO or SA. Maybe I'll give the ram itself a fraction more.

[kaan]

Thanks for the info about the warning, since both systems report similar issues, I'm inclined to think it's a false positive too, at least I want to

Indeed using "Parallel CPU's" amplifies the issues a lot, it's extremely hard to replicate an issue that shows up on Parallel with a Single test setup, tho with my very initial faulty Corsair's, I managed to do that, but it wasn't easy

Very interesting IO/SA experiences there, wish I could measure IO/SA too, but with my MSI motherboard I can't, they are not reported anywhere, you can only set them and not read them

I will test a couple of similar things now too, out of curiosity, will report if I get any useful results (Going to test 1.3V + 2666mhz, maybe TridentZ's work at 1.2V but need more juice to pass the parallel hammers)

Edit: Also, as a small suggestion, I don't think you can ever make what you are trying to do work, there is probably some major flaw in that BIOS, and what you are doing is probably just suppressing it, but the issue obviously shows up even in the sweet spot - I also just learned these kind of issues are common in V1 motherboard and some manufacturers like Gigabyte move fast to V2 boards and ignore V1 issues, scary stuff

[RogueEric]

After running since my last post at 4.6Ghz on my 6700K @ 1.35V and XMP timings at 3000Mhz on my Trident Z @ 1.35V with total stability, I think I'm satisfied with those results and am now going to try the other direction and see how little voltage I can run and still be stable.

First test at 1.34V after 4 hours of Realbench is in the books:



Temperatures came down a bit as average temps are in the 60's during testing with spikes into the mid 70'S on the CPU.

As a note, I am now running VCCIO and SA at Auto settings in the UEFI.

[kaan]

I've tested 2666mhz + 1.25V + SPD timings with TridentZ, works great, satisfies me (15-15-15 instead of the specced 15-16-16)

Have you checked the hwinfo64 flags with the 4.6ghz OC?

I've tried 1.25V @ 4.4ghz - although it was seemingly performing well, I got the "IA: Electrical Design Point/Other (ICCmax,PL4,SVID,DDR RAPL)" flag from hwinfo64, only shows up under load

I don't know what any of these flags actually mean, yet from past experience, if the flags show up, the system lags in unexpected ways, so I'm back to 40x constant with 1.2V

[porina]

Edit: Also, as a small suggestion, I don't think you can ever make what you are trying to do work, there is probably some major flaw in that BIOS, and what you are doing is probably just suppressing it, but the issue obviously shows up even in the sweet spot

It is entirely possible I wont find stability at 3333, at least not at any sane long term setting. 3100 is stable, so while it is a slight loss in potential, I need to draw a line somewhere and get on with it. I've also posted on overclockers and will see if I get any further tips there but I'm out of ideas.

Part of me is debating getting an X99 + Haswell-E to play with, which is what the ram was designed for in the first place and quad channels should remove a significant bottleneck. But it is a lot of cash just to play! I'm not sure in performance per cost terms it is better than skylake.

[kaan]

Quick but major update: MSI fixed the XMP issue with the 1.43 beta BIOS for Z170I, other beta BIOS updates and production BIOS updates should be near

I can now enable XMP without any issues or any extra observed heat (3000mhz with my TridentZ)

They are very criptic, they promised to tell what the issue is but they didn't yet, however at least the issue seems to be fixed, as I've tested the system relatively thoroughly

Edit: I know you all have different motherboards but since a BIOS update was able to fix my issue, same should apply to you too, not entirely sure about 3333 tho

[porina]

When did they release the firmware? Recently?

On my MSI mobo, the last firmware was released over a month ago. Is there a related thread for this on their forum?

[kaan]

At first they give these bets bioses to people to try them like test rats, so you might find one for your mobo on the MSI forums, they should give you one if you start a thread, assuming there is one available for that mobo

It was released yesterday