1 Rookie
•
25 Posts
0
105
March 8th, 2025 08:39
Peculiar problem with Precision 3650 and graphics adapter
I have two Precision 3650 towers here that I am rebuilding. I found a curious problem affecting both systems exhibit identical symptom. The processor is a supported Xeon W-1250 (SRH48).
Using the iGPU Intel processor graphics (via onboard DP port), the system will spontaneously reset/reboot shortly after loading to the desktop. Usually within 90 seconds but has gone up to about 5 minutes. There is no crash dump or logs other than the 'Windows did not shut down properly' dirty bit error that gets reported on reboot. It does not happen early in the installation of Windows but it does happen very late (when Windows is guiding you through SETUP questions)
When a graphics card (dGPU) is installed in the PCI-E x16 (PEG) slot, the system is 100% stable. But wait, there's more! It is 100% stable even when I am actually connected through the integrated graphics. i.e. no display connected to the dGPU. I noticed some of the benchmark apps were still defaulting to the dGPU for some workloads and outputting via the iGPU. So I disabled the dGPU in Device Manager, restarted and set primary display adapter in BIOS to Onboard, booted to Windows, checked the dGPU was still disabled, to ensure the iGPU (Intel processor integrated) would be used for everything.
I ran PCMark, 3DMark, processor stress, memory stress for several hours (cumulative), 100% stable. I confirmed the iGPU was being utilized for ALL workloads by using Task Manager and watching utilization of the various resources such as 3D, Video Decode, Video Encode, or Memory Copy. All showing utilization on the iGPU not the dGPU.
The dGPU I have been testing are both slot powered (under 60W board power), no aux power connector. Running all the benchmarks and stress tests on the (enabled) dGPU also is stable, BTW.
Simple recap:
System with only IGP = spontaneous reset/reboot
System with dGPU inserted into PCI-E x16 (PEG) whether utilized or not = 100% stable via IGP (or dGPU)
Windows 10 (22H2) and 11 (24H2) exhibit same problem. Latest drivers or older drivers, same problem. Latest BIOS or the oldest BIOS I am permitted to revert to several versions ago, same problem. Tried all three BIOS settings for primary display adapter, Auto, Onboard, or dGPU. Multi-display support on or off. PCI-E (PEG) bifurcation Auto or x16. Disabled almost all onboard devices except for LAN and rear panel USB. Same problem. Kernel DMA protection in Windows ON or OFF, same problem. Core Isolation/Memory Integrity OFF. ReBAR OFF (always a safe choice). PCI decode above 4G ON or OFF, same problem.
Any ideas?
tcsenter_d1db75
1 Rookie
•
25 Posts
0
March 18th, 2025 05:29
That's it! I'm making an 'executive decision' - off to ewaste both mobos are going. I think there is some wonky component or VRM on the mobo. What is the interaction that causes it to be suppressed or masked when PEG slot is populated, I don't know. But time to move on.
tcsenter_d1db75
1 Rookie
•
25 Posts
0
March 8th, 2025 08:44
I have another processor coming to test, a different supported model not the same. Won't arrive until probably Thursday March 13th.
tcsenter_d1db75
1 Rookie
•
25 Posts
0
March 13th, 2025 08:33
Installed a supported Rocket Lake i5-11600K to rule-out the Comet Lake Xeon W-1250. Same problem, but even worse! Unlike before, I can't even get it to load Windows SETUP. As soon as it starts loading the OS (WinPE) boot files = reboot! I have reset/cleared BIOS, load UEFI defaults, etc. No change.
When I insert a PCI Express graphics card, keep primary display adapter in BIOS to Onboard or Auto, with monitor connected to the onboard iGPU, everything works. OR when using the PCI Express graphics card, works great.
So it not the CPU. Something must be going on with the chipset/BIOS, low level PCI/PCIE resource configuration or assignment (firmware) in Dell's BIOS code. GAAAAHHHH!
Results from Dell's Diagnostics (BIOS based) Advanced Test image attached (in case anyone was going to ask).
mazzinia_
4 Operator
•
1.4K Posts
1
March 14th, 2025 09:59
Hello,
was following this and I've to say is a very weird issue. And should not be widespread, otherwise there would be countless complaints.
Now, this is me just musing ... this feels an electrical issue. It feels the power distribution is too unstable to drive the integrated gpu once reaching a certain stage ( that's not reached while in bios ) , and that it stabilizes when an external gpu gets added to the fray (forcing wattage to go to the pci-e slot).
Do you have another pci-e card lying around ? not a gpu... anything that draws 10W or whatever, to test what happens if this is installed in the 16x slot , the external gpu is not present, and windows setup is started ?
(that i5 11600k is a 95 W cpu , the xeon w1250 is an 80 W cpu, it's not impossible that they draw a different amount of power during the same operation, in this case the windows setup. and the i5 would draw just a bit more by inferring on the wattage of the package, thus incurring in the issue just a bit earlier )
tcsenter_d1db75
1 Rookie
•
25 Posts
0
March 14th, 2025 11:32
I was considering some voltage regulation/stability issue like the one years ago where some very reputable (and pricey) PSU models could not accommodate very low load on one of the rails during POST and boot process of some newer Intel VRM spec mainboard designs, causing some funky cross-regulation drop out that would make the system reset or reboot. Remember that about 15 years ago?
The PSU is 3 year-old Corsair CX500M unit that runs great in two other systems. But I'm going to try a different unit.
mazzinia_
4 Operator
•
1.4K Posts
0
March 14th, 2025 13:08
@tcsenter_d1db75 to be honest no, i don't remember that issue , but I've experienced slight issues developed by one or two psu over the years that were visible only thanks to a kvm
tcsenter_d1db75
1 Rookie
•
25 Posts
0
March 14th, 2025 23:22
Welp, I installed a PCI-E x4 NVME adapter card (with NVME SSD) into the PEG slot. BTW officially Dell does not support any other device in the PEG slot, except graphics/display adapter. Says so in the documentations. But I did it anyway and it seemed to be more stable. I could boot to desktop and it ran, for about 15 minutes = Rebooted.
I mucked around with the PEG Bifurcation options/configuration. Some settings seemed to provoke reboot while loading the OS, other seemed to work long enough to get to the desktop and even run for a while. But within 15 minutes = reboot. So I remove it.
Tried a different PSU (Antec) and that seemed to improve things, I could boot to the desktop and it ran great! I even ran 3DMark one pass and thought OMG it was the PSU all along? But 15 or 20 minutes of usage = reboot.
The ONLY configuration that is stable is when a graphics card is inserted into the PEG slot. I noted that when I was able to run that pass of 3DMark Night Raid with no graphics card inserted, the result was ~9600. When I insert a graphics card BUT run the benchmark on the iGPU (Intel UHD), the result is lower ~8600. This I have verified twice now, in each configuration (when I was able to get that far without a graphics card inserted). Another interesting note is that when I changed the 'rendering device' from NVIDIA graphics card to the Intel UHD 750 graphics, the application warned "rendering device is not connected directly to the selected display" but in fact the display IS plugged to the onboard graphics port, which is the Intel UHD graphics. It further suggests to me there is some kind of PCIe routing, lane reversal or switching bug here.
So changing these things seems to be altering something, getting a lot further than mid-boot but in the end, whether it is 5, 10, 15, or 20 minutes, it will reboot spontaneously. With graphics card inserted = UPTIME FOR HOURS AND HOURS.
mazzinia_
4 Operator
•
1.4K Posts
0
March 15th, 2025 00:20
If the benchmark value was higher without the discrete gpu, maybe some setting somehow causes a kind of overclocking
mazzinia_
4 Operator
•
1.4K Posts
0
March 18th, 2025 11:01
@tcsenter_d1db75 it's a pity. installing 2 old 5usd gpu would allow to let them be usable for a while more