A couple of days ago I upgraded to a 7950X3D and 2x32GB DDR5-6000 on an MSI X670E Carbon motherboard. This is my first time touching an AM5 system. I’ve noticed the SOC voltage was automatically set to 1.35V when XMP is enabled, which seems very high to me (on my 5950X I had it at 1.05V and with XMP disabled the 7950X3D runs at 1.05V as well).
Is 1.35V SOC voltage normal and safe for Zen 4? Since this seemed very high to me I manually set it back down to 1.1V, which appears to be completely stable. If 1.1V is working fine, is there any possible reason to have it as high as 1.35? The RAM is running at 1.4V as per XMP, if that matters.
Good to know that it’s safe. I’ve tried to searching for information on what this voltage was supposed to be on AM5 but found very little, so I was unsure. If I try to manually set it to 1.35V, the entry turns red in the UEFI so I thought it was pretty high. So far 1.1V has been stable.
As for the 3rd slot, yes I am actually using it for a 10G network card, so I can have that in while the GPU still gets full bandwidth (as unnecessary as it may be in practice). I mostly wanted a board with at least 3 slots (and all of them with at least 4 lanes) because in the future it’s likely to migrate into my home server as I upgrade my desktop again and the PCIe slots are a lot more useful there.
The board so far has been generally working well and I got the features I wanted like the PCIe slots, the 6 SATA ports, the full complement of audio jacks, the debug LED, sufficient internal USB headers and a thermistor connection which I actually use to measure the temperature on the radiator. There are some quirks, but nothing I would consider a huge problem at this point:
Booting is slow, it takes about 30s at least until I get any video output, I think this may be memory training. This is a huge pain when tweaking things, rebooting to change a setting takes a long time.
There’s a memory-related setting called ‘High Efficiency mode’ which is supposed to improve performance, but turning it on makes the system BSOD when loading Windows every single time. I’m unsure whether this could be caused by me lowering SoC voltage, will have to test with it at the default 1.35V too.
The CPU is also locked to 150W power consumption with PBO enabled, no matter what I set PPT, TDC and EDC to, it just won’t go over. It’s not hitting the temperature limit either, as it only runs at about 80C in full load. There should be headroom left but it just won’t use it. I’m unsure if this is a motherboard bug or whether the 7950X3D is hard-locked due to the cache.
Setting the clock offset also doesn’t do anything, the CPU boosts up to 5750MHz at stock and I’ve been unable to get it any higher with PBO. I can see in Ryzen Master that the clock offset is applied and the limit changes (to 5950), but the CPU never actually boosts higher than stock.
Practically speaking, PBO is only useful for Curve Optimizer. Like I said, I don’t know how much of this is intended because the CPU has V-cache and how much is the result of UEFI bugs.
Despite these potential issues, compared to my AM4 Aorus Master this board is very well behaved, on the Aorus I could hardly touch PBO settings without causing instability and returning them to stock did not help, only clearing CMOS did. That was a very severe bug, the Carbon so far doesn’t seem to have any crippling problems.
thanks for your feedback, the MSI Carbon is one of three boards I have on my list if I go to AM5.
As far as I know the 7950X3D is hard-locked due to the cache, if your system is stable, I would leave the VSOC voltage as it is, more is not necessary better, on an ASUS Strix Gaming I had to set a negative offset for VSOC to get the system stable with 4 DIMMs.
I am on the MSI x670e Carbon, too, but with 7950x
SOC voltage, I manually set it to 1.1v running DDR5-6000 2x32GB rock stable. I tried with 1.05v. It could boot into Windows, but not stable with 1.05v.
As said, the bios is kind of buggy. “Memory Context Restore” is broken; Longer boot time than any other brands; “High Efficiency Mode” doesn’t work. However, it is pretty good to run as a server/workstation. As a server, you just don’t reboot that often.
Right now, the only thing really bugs me is that IOMMU group on the 3rd slot is broken. All virtual functions created by SR-IOV are in the same IOMMU group, which makes it impossible to pass those virtual functions to VM. Fortunately, IOMMU group works properly on the 2nd slot.
Yeah this seems to be the case, I’ve seen another thread from another user here who had the same behavior on an Asus board, so I’m quite certain this is intended behavior with this CPU. If there are no PBO woes, then I can say I’m really quite pleased with the board so far.
That’s exactly what I’m running as well, it’s been perfectly stable for me too.
Are you sure this is the case? Like I said above, I get about 30s on a normal boot, which I do think is long, however I’ve been swapping notes with a buddy who bought a 7950X3D system at the same time as me, except he has an Asus X670E TUF motherboard. From what he told me, he experiences the same ~30s black screen at boot.
I haven’t messed around with IOMMU and passthrough on this board yet. If the primary PCIe slots, the ones which are connected to the CPU, work fine in this regard it’s probably good enough for me. The 3rd is probably not working like the others because it hangs off the chipset rather than going straight to the CPU.
yes, I don’t know a board where PCIe slots have separate IOMMU groups over the PCH, if you need three slots with separate IOMMU groups, then with AM5 only MSI ACE and Godlike and maybe the ASUS ProArt B650 creator remains.
With the ProArt it looks like you have 4 CPU PCIe lanes in the third slot when M.2_3 is not used.
It only takes you 30 seconds to boot? That is quick.
It takes 80 seconds to boot my system. In the windows task manager, “Last Bios Time” is also over 80 seconds, which confirms the slow boot.
Is there any bios option you can think of that makes it boot faster?
I was on old z87 intel system. I ran sr-iov on the last pcie-slot, which is from the chipset. There was no issue to passthrough virtual functions. I believe it depends on motherboard manufacturers to implement iommu group. They just don’t bother to put extra code to separate them.
why should they, I’m not sure if more than 0.1% of DIY users know what SR-IOV is or want to use that on a consumer board.
The fact that you have this option in your bios does not mean that it has been tested.
Yeah same here, I actually have another MSI board, an older X470 Gaming Plus Max in my home server and it has IOMMU group separation that’s like that as well. Everything going straight to the CPU is in its own group, the 3rd PCIe x16 slot along with all the x1 slots and a bunch of integrated peripherals are in the same group.
Just about, I’m not using that PC right now to check the BIOS time but it’s definitely around 30s. I suppose I should clarify, it’s about 30s for a “normal” boot where I just turn on the PC. If I change any BIOS setting at all it takes very long, definitely over a minute. Fortunately I don’t intend to do that after I have all the settings dialed in so it’s not TOO bad, though 30s still seems long to me.
Unfortunately not, it behaves this way even with almost completely default settings, for instance just disabling TPM and turning on XMP. I doubt either of those does anything to the boot time, everything else was default when I first booted the board and it acted the same as described above.
I’ve actually looked at my settings and played with BIOS options some more, this time actually looking at boot time and it turns out what I initially said was wrong since I misremembered the settings I was running and didn’t check before posting.
In order to get quick booting the Memory Context Restore option must be enabled manually, with it enabled I get just under 20s BIOS time as reported by Windows. The default Auto value on the setting results in about 60s boot time and changing settings in the BIOS results in over 90s boot time, only for the next boot after saving changes.