System: Intel Xeon W7-2495X, ASUS Pro WS W790 ACE, 512GB DDR5 ECC RDIMM, ASUS RTX-3090, Windows 11 Pro workstation.
I am trying to use the Intel XTU software to do some tweaking to my system.
I figured that would be easier than trying to decipher the complex BIOS settings on this Workstation Motherboard.
I am familiar with overclocking consumer boards, but the BIOS settings in the workstation board have me baffled.
XTU won’t run and says that it is “incompatible with Windows VBS (Virtualization Based Security)”.
It says to enable Undervolt Protection (UVP) in the BIOS.
I don’t have any setting by that name in the BIOS.
If I check Windows settings it definitely has VBS running.
For the first issue, you need to disable Core Isolation in Device Security control panel to disable VBS.
Negative voltage (undervolting) is now locked in XTU due to Plundervolt vulnerability, and XTU will refuse to run without UVP enabled. Undervolting can still be done via BIOS, however. (My understanding is that W790 boards have UVP enabled by default and cannot be turned off, as you already have access to full FIVR control in the BIOS).
I am only seeing if I can get 4.0GHz+ all-core without major temp issues.
If I can’t it is no big issue.
I played around with the CPU Core Ratio in the BIOS, and this was the result:
BIOS, AI Tweaker, CPU Core Ratio, Auto __________, CPU-Z Multithread score 16620, 3.3GHz All Core, 48C maximum
BIOS, AI Tweaker, CPU Core Ratio, By Core Usage, CPU-Z Multithread score 24000, 4.8GHz All Core, 94C in only a few seconds
So I guess I would have to undervolt to get the temperature back under control.
But it does stop at 94C and start to throttle down the clocks to around 4.6GHz.
So I might try running this for a bit, I rarely do all-core workloads for extended periods.
I decided to try the 3D software that I develop, since it is highly threaded software.
Generating a 32km x 32km 3D world, it clocks the CPU up to 4.8GHz all core and 95C and then throttles back the clocks a bit.
The world creation is really fast though and just eats through 4GiB of data really fast (16174 ms or 16.174s).
The same world on my i7-6950X X99 64GB system takes 88068 ms or 88.068s (1 m 28.068 s) or 550% slower.
Nice!
FYI the Xeon W7-2495X system is using a Noctua NH-U14S-4677 cooler with dual fans.
EDIT
Argh. Changing the BIOS CPU Core Ratio to By Core Usage also causes my RTX-3090 GPU Power LEDs to start flashing on power on.
The computer has a Corsair HX1200 1200 watt Platinum Power Supply.
I would think that would be enough power, but apparently not…
The Intel Xeon must be drawing a large amount of power on power on.
Even with the IHS size of the Sapphire Rapids, I think it’s still pretty challenging to pull 500W+ on air in summer without hitting TjMax. Though modern CPUs are pretty safe to be run under this temperature.
On my Supermicro X13SWA-TF + w9-3495X, the wall meter reads 350W when the CPU package is reporting 160W. I’m still running Intel Arc, which idles at pretty high power (I ballpark at around 50W) so IPMI and chipset and everything else seems to consumes around 130W to 150W range.
Given these numbers, even if CPU consumes 500W during boot, there should be plenty of room left for 4090… do you have a watt meter to test power usage during boot?
I have a plug-in watt meter somewhere in my tools in the basement. I will have to dig it out. I don’t recall if it has peak reading capture.
The computer is also plugged into an APC 1500 UPS with digital display, and the UPS shows that it is drawing the same amount whether I have the Core Ratio on Auto or By Core Usage.
So the power on current draw must be a really fast transient peak that is just enough to trigger the RTX-3090 power LEDs but too quick to be shown on the UPS display.
The Corsair HX1200 is a really highly rated PSU, so I can’t see anything being the PSU’s fault.
If the CPU was drawing 500 watts, there is another 350 watts free for the GPU, plus 350 watts for everything else.
I wonder if it is just a really sensitive power trigger on the ASUS ROG STRIX RTX-3090 White OC 24GB model.
Google didn’t turn up any information on that.
I am using three separate PCIe power cables to the GPU, and the supplemental PCIe power cable on the motherboard is also plugged in.
EDIT
I connected my digital watt meter into the circuit.
It shows that the computer draws 1.66 amps during boot, which is 200 watts.
But I may have found the difference…
There is a difference in boot up procedure between Auto and By Core Usage modes.
When I turn on the power in Auto mode, it simply spins the fans, lights up the GPU RGB and boots to the BIOS screen and then Windows.
But in By Core Usage mode, it does a two second power on with fans and RGB, then a reset as if BIOS training where the fans and RGB go off again for half a second, then back to a regular boot to BIOS screen etc.
It must be training for CPU core frequency and voltage etc.
It is after the BIOS training power half second drop out that the GPU red lights flash.
So the GPU isn’t liking that the motherboard is toggling power off again during boot, I am assuming for CPU core training.
On my Supermicro board, I only see it do this boot-train-shutdown-boot sequence when I cut off the power from the PSU (turning off or unplug), or when changing settings that explicitly state “must power cycle after changing these settings” (which By Core Usage is one of them). Subsequent boots are fast as long as I don’t cut off the power again.
The ASUS W790 ACE board always does the boot-reset-boot sequence when CPU By Core Usage mode is enabled.
So it must not be saving any settings that it gets from this process, and requires them on every startup.
I would say that I hope they change this in a future BIOS version but I doubt it.
So I would have to put up with the GPU Power LEDs flashing if I use this mode.
The RTX-3090 runs just fine, even with the red power LEDs flashing.
I did a bunch of Googling on this, and those indicator LEDs can also just be disabled with software.
Since they are a “GPU Power Issue” indicator, they are apparently false triggering by the motherboard power cycle, and it is not an actual issue with the PCIe power connections themselves.
So I will just ignore the LEDs for now.
I switched my system back to CPU Auto mode anyway, since I prefer running at the stock 3.3GHz/4.0GHz speeds for lower temps and less power consumption.
I will use the CPU By Core Usage mode only when I have to perform some really computationally intensive workloads where I want the full 4.6GHz/4.8GHz 500 watt performance.
My system with Noctua NH-U14S-4677 cooler will cause thermal throttling anyway at 500 watts, so there is not a lot of use with the 4.8GHz all-core mode.
If anyone is wondering why I don’t just use an AIO to get higher power soaking, it is because a few years ago I had a $6000 system with a water loop, and it leaked and corroded the motherboard and case, and ever since then I have been air cooling only.
This W7-2495X, W790 ACE, 512GB, RTX-3090 system cost me $13,000 so I’m not gambling on that happening again.
You can absolutely disable, I did so while testimg out wendells 790 board
I forgot exactly which setting it wants off though
I think it was virtualization maybe, try looking in CPU settings
Thank you for this, I wasn’t aware that they’ve now allowed runtime undervolting with specific configuration(s). According to FAQ, the feature matrix now looks like this.
Undervolt Protection
Hypervisor (VBS, HVCI)
Runtime (OS) Undervolting
BIOS Undervolting
Runtime Overclocking
BIOS Overclocking
Off
Off
On
On
On
Off
Off
On
Toggling off Core Isolation on Windows should be enough to disable VBS/HVCI (though in some case Memory Integrity needs to be turned off as well).
For W790, though, there’s no option to disable or enable Undervolt Protection as far as I know. (At least, not on Supermicro X13SWA-TF, and on ASUS W790 SAGE-SE manual)