Z590 Tomahawk wifi auto voltages are dangerous!!

M

mnemonic

Member
Joined
Dec 26, 2021
Messages
120
I have noticed only after a few weeks but this is not good for cpu i think.
This has been noticed with an i7-11700K without overclocking @ stock clocks.
It concerns the VCCSA and VCCIO2 voltages, when left to auto like the board comes out of the box the voltages are way out of spec.
VCCSA at auto gives 1.597V , when setting it manually the red values start beyond 1.2 V , that is 0.4V overdriven.
VCCIO2 at auto gives 1.376V , when setting it manually the red values start beyond 1.15V , that is 0.2V overdriven.

Why does MSI do this ? It is good to degrade the cpu very fast which is not what anyone wants besides maybe Intel who would like extra sales.

When having them set at their correct values of 1.2V and 1.15V there is no negative effect,
in fact with the higher auto voltages the system was getting data corruption which has now ended by setting them manually to the max safe values.

This has to be addressed ASAP before CPU's all over get killed with these high out of spec auto voltages.

Other values at auto seem to be in spec, but maybe the cpu cores are getting also more than they actually need to run, this would require more testing which i'm not looking to do. Going up to over 1.4 Volts is not needed for running at stock speeds without overclocking either. The system could be way more power efficient than it is.
 
Sort by date Sort by votes
How does the CPU VCCSA, VCCIO voltage being identified? Is it from BIOS or third party software like HWiNFO?
Could you post screenshot showing the voltage value?
WIth default BIOS settings, the CPU VCCSA, VCCIO voltage should work around 0.94~1V.
Could you load BIOS default with F6 key and save settings to check the CPU VCCSA, VCCIO value?
 
Upvote 0 Downvote
both in bios and hwinfo they get displayed. Default bios settings put them in auto and that's when they get too high. I just put both VCCIO2 and VCCSA to manual at their max safe value in bios.
the set value in bios will show red (in manual mode) when going over the 1.2 vccsa and 1.15 vccio2. Bios defaults being auto gave me the high values. Vccio (1.044V) is in the correct range as are all other values except the vccsa and vccio2. When set manual to 1.2 and 1.15 the values reported are 1.196 vccsa and 1.144 vccio which are much safer compared to their level in auto mode.
 
Upvote 0 Downvote
Yes, i talked about this several times before, like here, here and here.

I can only deduct that for Rocket Lake, the acceptable values have increased. Remember, 11th gen was a completely new architecture for the first time in six years. The age-old maximum values for SA and IO voltages (which there is also IO2 now) don't seem to apply anymore for Rocket Lake in the same way. If you look at the values that the board manufacturers themselves let the BIOS set, once you enable XMP, they can go up to 1.6V for SA and IO2 on Auto, as you observed. This is not something you would've seen in previous CPU generations, so Rocket Lake surely can take a bit more there.

For example, i'm now using a new RAM kit on Z590 + 11th gen, 2x 16GB DDR4-3600 CL15-15-15 (G.Skill with Samsung B-Die). I need 1.24V SA and 1.32V IO2 for that, below that it will fail POST. With my previous 2x 8GB DDR4-3600 16-16-16 (also G.Skill B-Die), i "only" needed 1.22V for each. It highly depends on your RAM and its frequency and timings how much you need. But the Auto values were over 1.4V for both. So the voltages increased with Rocket Lake in general.

I think they kept the point of the values turning red in the BIOS from previous generations like Z390/Z490, but for Rocket Lake the point of danger moved up a bit. Certainly, MSI sets excessive values there, and it's clear that they are not necessary. This drives up power consumption of the "uncore" part of the CPU in all load situations including idle. But i think other mainboard manufacturers also do something similar.

I don't know what a good solution could be, because if MSI would lower the auto voltages for XMP, then they'd basically have to redo all their RAM compatibility testing, propably even their memory training code, because there is bound to be some crazy high-end RAM that may not be fully stable anymore at lower SA/IO2 voltages. Right now they leave huge headroom in these voltages, and i don't know by how much they would be willing to reduce it without putting in question their entire RAM QVL.

Factually, i agree with you. Those voltages can be set crazy high by the BIOS, and it's not good. Maybe the 11th gen CPUs can handle it, but i don't like it.
 
Last edited:
Upvote 0 Downvote
According to Intel the specs for RL voltages are not above 1.2 and 1.15V anyway. I'd attach the official intel , pdf's , were it not that this site claims the size is too big.
Nice sabotaging everywhere. So only the smaller files get accepted, but i can guarantee that Intel says the voltages mentioned are what they should be.
Also setting them to spec manually does not have any negative effects. Like i said also, less data corruption and win 11 not destroying itself that way.
So the voltages are way too high, this does not depend on the ram. I run 4x8 8GB 3600Mhz corsair vengeance, The board bios just overdrives the voltages WAY too high when it is not needed.
If it doesn't work with those voltages as you had then the ram or cpu isn't good enough imo, b-die or not. My ram is identical but two sticks are samsung and the other 2 hynix. Good ram and it works flawless at Intels spec'd voltages, so it is not needed at all for RL to go any higher.
 

Attachments

  • Intel 11 gen cpu-3.pdf
    390.7 KB · Views: 232
  • Intel 500 series chipset-3.pdf
    1 MB · Views: 1,978
Upvote 0 Downvote
Those are just documents listing errata... instead of attaching, what happened to good old linking? But i'm not talking about theoretical specification documents. If we only go by the official CPU specs, then we would have DDR4-3200 maximum etc. I'm talking about voltages used in practice by the board makers, and what voltages are necessary to achieve stability. This has risen considerably with Rocket Lake, no matter what the documents state or what you and i like. Because as i said, i'm fully with you here, they are very high if you let the BIOS do what it wants.

I said that with my 2x 8GB DDR4-3600 B-Die with highly manually optimized timings, i "only" needed 1.22V for VCCSA/VCCIO2 on 11th gen / Z590.
You know what i needed for the same kit on 9th gen / Z390? VCCSA 1.05V and VCCIO 0.95V. Those are the default values for non-XMP, which also worked at XMP speed.
Even the Z390 BIOS liked to increase those values considerably, but not as high as Z590. And it was much the same with earlier platforms. I remember on 6th gen / Z170 with XMP, the VCCSA/IO voltages were pushed so high on Auto that i had graphics errors when using the internal graphics. When i lowered the voltages manually, it was all ok again.

Trust me, the higher frequency the RAM and the tighter you optimize the timings, the more voltage you do need. And optimizing the RAM timings manually (not only primary timings, but secondary/tertiary) can be as stressful for the memory system as a couple 100 MHz higher RAM frequency. But again, i agree, you don't need nearly as much voltage as the board makers want to apply. And i'm not singling out MSI here, because if you read here for example, it's an industry-wide phenomenon with 11th gen / Z590. So first we have to get some wide recognition that the Auto values are too excessive. Then the board makers could maybe agree and lower them a little. Even then, they will never lower them too much (for example to the Intel spec), because they don't particularly care about power draw or putting more stress on parts of the CPU, they only care about lots of stability headroom so they get less complaints for their support.
 
Last edited:
Upvote 0 Downvote
I'd have to find the links again since i downloaded the documents, but the electrical spec section does mention it all.
I'll try to find a way to insert the info here. But give me some time. RL may be more tolerant to higher voltages but the Intel info says it is not needed to go this high.
And i notice very well it is not needed at all, since the data corruption has ended when both vccsa and vccio2 are set within the intel specs. Since RL is also new and less adopted due to social media influencers crying all over how bad RL is according to them just because it has 14Nm again....there is not so much people actually running them to get all their experiences with it.
 
Upvote 0 Downvote
citay said:
If we only go by the official CPU specs, then we would have DDR4-3200 maximum etc.
Click to expand...

Try to run 64 or 128GB at 3600 and you will understand why Intel (and AMD) states up to 3200 for all the top CPUs.
As I told you before, it cannot be any "sweet spot" at 3600MHz for memory.
Simply because that "sweet spot" is true for 2x8GB and 2x16GB only!
:biggrin:
 
Upvote 0 Downvote
Well there is also the point, that with 4x8GB 3600Mhz system would not get past boot on first boot after build or after bios update. Just a simple reset and it did pass without a hitch.
Some oddities but nothing alarming. The 4x8GB 3600MHz xmp even does run in gear 1 mode, so that ain't bad at all. And it does so whether the voltages are in- or outside of spec.
Therefore i really don't see why those voltages at auto were selected to be that high out of spec.
 
Upvote 0 Downvote
mnemonic said:
The 4x8GB 3600MHz xmp even does run in gear 1 mode, so that ain't bad at all.
Click to expand...

Yes, 32GB at 3600MHz in Gear 1 with a good CPU is doable.
But you will always need higher voltages for 4x8GB vs 2x8GB.
Otherwise you'll get into all kind of stability issues.
 
Upvote 0 Downvote
I already mentioned that i was getting data corruption when the voltages where out of spec at auto.
Since i lowered those two voltages the data corruption seems to have vanished.
Don't know what else to make of it.
 
Upvote 0 Downvote
As you can read from the Reddit post i linked, this happens on other brands' boards as well. So they all must've established in their early BIOS testing that this is some kind of "sensible" value to set it to on Auto with certain RAM configurations (of course it's not really, but as i said, they don't care so much about factors such as power draw etc.). It is good that you want to report it as a bug or a mistake, but it seems to be intentional, so you'd probably have to change their mind in how they approach these voltages.
 
Upvote 0 Downvote
Back
Top