What are safe 24/7 IO and SA voltages for coffee lake?

[nicholas70]

MSI Z390 MAG tomahawk
i7 9700k
rtx 2060
hxi 850w
corsair vengeance 3200mzh 16gb

So to make a long story really short I recently discovered my motherboard isn't supplying correct voltage to either my RAM or cpu or both. I know this because I've tested known good RAM and cpu and still get errors at factory settings running OCCT. Intel Processor diags will also eventually fail on IMC test on stock settings. I'm not running any OC but if I downclock the RAM slightly and disable turboboost the tests pass. So this brings me to my main question what are safe voltages for the VCCSA and VCCIO or SA/IO? I've been researching this for hours now but it seems everyone has different answers to this question and Intel's spec sheet seems to imply that IO voltage over .96(the default) is out of spec and would void the cpu warranty as well as risk degrading/damaging the cpu. While other sources I've found disagree on the matter and indicate safe voltages might be as high as 1.3 or as low as 1.15, so what is it already? Any help I could get here would be great as I kinda think RMAing the board could be a real pain, and I'm considering just limping along with this one until I can save up for a new better mobo but don't want to damage of void my cpu warranty.

 

[trents]

Max safe 24/7 system agent voltages are generally considered to be about 1.3 I think for last several Intel CPU generations.

 

[nicholas70]

Good to know. I will probably start tinkering with the voltages again when I get home and will start out by just changing the SA and RAM voltages slightly and hope that does the trick as what I'm reading seems to indicate the IO voltage is the one that you don't want to have to mess with if avoidable. I also notice that my board pushes a lot of vcore by default up to 1.31 on factory settings during heavy load. Am I correct to assume lowering that shouldn't cause memory/imc related errors?

And again thanks for your help on this.

 

[wingman99]

Lowering the core voltage can have L1-L2 cache errors, there in the cores.

 

[Alaric]

Max safe 24/7 system agent voltages are generally considered to be about 1.3 I think for last several Intel CPU generations.

If that includes mine, I maintain a fair memory OC at 1.20v for SA and IO.

 

[nicholas70]

Lowering the core voltage can have L1-L2 cache errors, there in the cores.

Guess that means I'll have to address the memory/imc error before I try lowering vcore to try to bring my temps down a bit.

 

[Woomack]

Core and cache voltage in Coffee Lake is the same - always marked as CPU voltage.
Cache clock has to be 300MHz below CPU clock to avoid errors.

The typical lowest voltage at which CPU/cache is still stable is about 1.1V (up to about 4.2GHz). How high voltage is required depends on the CPU and its frequency so that you have to test yourself.

Maximum safe voltage for IO/SA was not really confirmed. Typical max, which higher motherboard series set is 1.35V IO and 1.40V SA. Most motherboards won't set more than 1.35V so it suggests that something between 1.35-1.40V is maximum safe. However, you won't need more than ~1.3V to set memory at 4000-4266 (lower capacity, 4-8GB modules). Typical at 3200 will be up to 1.15V. Auto settings are usually higher than required.

 

[nicholas70]

With the previous cpu I had, before I RMA'd it, it appeared the cpu min stable vcore was around 1.22. Granted it is a bit hard to say for sure due to the memory/imc errors I was getting, but that was the voltage at which it wouldn't crash the system and would run for hours before hitting what I'm quite sure was the imc/memory error. As for which setting is actually off that is causing the error I'm not totally sure as I haven't troubleshot the problem enough yet to know. All I know is it seems to be between RAM voltage, SA voltage, and IO voltage. I suspect that the stock voltages the bios and software are reading are not what is going into the chip and are falling just short enough of those voltages to cause the error. Part of me really wishes I hadn't got the mobo I did as after I bought it I noticed several sources on the web talking about power delivery problems with the board. None of the reviews mentioned this issue specifically, but it does seem that MSI cut corners on this budget board that doesn't make it the good value it would appear at first glance. Right now I have a feeling that perhaps even if I RMA the board the issue will remain as a minor inherent defect with this model mobo, and I probably won't even bother trying an RMA unless they offer and advance RMA option. I just happen to be OCD enough about these minor errors that I'm unwilling to live with them as some people might given the system seems stable otherwise.

Edit add: I did try xmp at one point and it failed rather quick with what I think was higher io voltage then 1.15 or at least that is what it said in the bios, but as I said before I suspect what I'm reading in bios and in the software is not what the system is actually getting.

 

[Woomack]

Run all at auto and memory with disabled XMP profile so 2133/2400 with all timings at auto. If it crashes the same way then is something else.

You can also check manual settings at lower CPU clock (to be sure it's not the CPU) like 4.5GHz 1.2V, LLC 1-2 steps below max (no idea what settings are on Z390 MSI), IO 1.20V, SA 1.25V, memory at XMP or manually 3200 16-18-18 command rate 2N, 1.35V.

 

[nicholas70]

I already know it can't be that the cpu is faulty, or at least it would be really strange to have 2 cpus both with a bad imc. It is also highly unlikely the RAM is faulty as I've tested with two different sets, but the system does indeed error with everything at factory settings even with an updated bios, and this is why I suspect the voltage readings are off. I will say that the system doesn't crash as a result of the errors it is just that occt will stumble on the error if left to run for say 4 to 8 hrs, and that the Intel diagnostics will flag the cpu imc as bad.

 

[Woomack]

I never liked OCCT as it was sometimes showing weird things.
Check memory in Windows Memory Diagnostic ( it's in Windows Administrative Tools in menu start).
Also, can check hardware errors in something like Hwinfo64. It's on the bottom of the table with sensor readings. It shows if there are any errors and adds info about the cause like memory, cache, etc. error.

I'm usually using AIDA64 for memory and IMC load - memory+cache stability test. It loads ~95-98% of available RAM and runs multithreaded tests.

It's hard to tell what is the issue as memory, cache, and IMC are all connected. I highly doubt there is something wrong with IMC as processors have almost the lowest RMA rate from all components.

 

[wingman99]

I already know it can't be that the cpu is faulty, or at least it would be really strange to have 2 cpus both with a bad imc. It is also highly unlikely the RAM is faulty as I've tested with two different sets, but the system does indeed error with everything at factory settings even with an updated bios, and this is why I suspect the voltage readings are off. I will say that the system doesn't crash as a result of the errors it is just that occt will stumble on the error if left to run for say 4 to 8 hrs, and that the Intel diagnostics will flag the cpu imc as bad.

This testing is with BIOS settings at default stock? If that is the case I would run memtest86+ to see if you have defective memory.

 

[nicholas70]

Ok, so it looks like it was the RAM voltage that was causing the errors because after I upped the voltage from 1.2 to 1.25 for the RAM I was able to run a 170 loops of the Intel diagnostic with no errors. I'll probably run it again tonight just to reconfirm it, but I have to say I feel stupid for having RMA'd a cpu for what was likely a minor problem with factory RAM voltage on the mobo.