PSHED.dll+10c0 or ntoskrnl.exe+50187b per BlueScreenView - the same BSOD code has happened five or six times over the past month with no overclocking on the GPU/CPU except XMP enabled. Memtest86 passes a full test. Whea uncorrectable BSOD while playing Star Citizen, and the BSOD seems fairly random with the common point being playing Star Citizen (I don't play many games outside of Star Citizen right now) - this most recent one happened a week after the last BSOD. SFC /scannow comes back clean.
 
Windows 11 pro (21H2)
3090 Ti FTW3 Ultra Hybrid (516.94, no OC, temps in the 60C range)
X299 FTW3 (1.28)
10940X (stock, AiO water cooled temps in the 60C range)
Mushkin MRC4U360GKKP32GX2 (4x32) 128GB 3600MHz C16 (stock XMP settings)
EVGA 1600T2 PSU
Intel 900P C: drive
Micron 9300 ssd game drive
Adaptec 8805 w/ eight 8TB seagate archive drives
 
Anyone have an idea where I should be looking for a PSHED.dll / ntoskrnl.exe BSOD? I thought that it might have been the eight sticks of 3200MHz C16 RAM that I had, since eight sticks put additional strain on the IMC over four sticks - yet I'm still getting (less frequent) BSODs with four sticks.

0x00000124

XMP disabled reduces the BSODs as well, however it makes the one game that that I play play like crap due to it being severely single thread bound.
 
EDIT: I'd have to restart and look which I can't do atm (copying files to a backup), yet there were two voltages that seemed to have helped a little when I bumped them up from auto, VSA and VCCIO or something like that? 0.950V and 1.1V.

Well, when this really big backup finishes I'll take a look at upping the VCORE. 1.2V wasn't completely stable, and the temps quickly shoot up with more.

You say you have no CPU OC but what exactly do you mean by that, that your BIOS CPU is on 'auto' or that it isn't on auto but multipliers are on defaults?
 
With Star Citizen being common crash point and codes pointing to PSHED.dll / ntoskrnl.exe it is possible instruction set triggers it so I would test stability with OCCT of not just regular instruction sets but also AVX2 and AVX3 to assure offsets are not too small. I was able to achieve best stability on my 10920X X299 Dark when I used per core OC ratio with x43 multiplier, AVX2 offset 3, AVX3 offset 5, mesh ratio x24, auto voltages, and Windows ultimate power plan.

arestavo

ZoranC
You say you have no CPU OC but what exactly do you mean by that, that your BIOS CPU is on 'auto' or that it isn't on auto but multipliers are on defaults?
 
With Star Citizen being common crash point and codes pointing to PSHED.dll / ntoskrnl.exe it is possible instruction set triggers it so I would test stability with OCCT of not just regular instruction sets but also AVX2 and AVX3 to assure offsets are not too small. I was able to achieve best stability on my 10920X X299 Dark when I used per core OC ratio with x43 multiplier, AVX2 offset 3, AVX3 offset 5, mesh ratio x24, auto voltages, and Windows ultimate power plan.

Yeah, Star Citizen uses AVX now (I think just the original AVX instruction set) - I'll likely need to stability test with AVX to verify true stability. Auto settings for cores/vmesh. My power plan is also set to ultimate performance.

Would you mind sharing a screenshot of your BIOS CPU settings and what HWInfo reports are your CPU multipliers? If memory serves me well EVGA's auto settings result in 0 offset for AVX2/AVX3 which was too aggressive for my 10920X. Easiest way to find out which instruction set Star Citizen might be using is by using two different offsets for AVX2/AVX3 and then observing did frequency drop while running and by how much.
 
Word of warning: When using auto for CPU mesh multiplier ends up 24 (Intel's native). But if you switch CPU multipliers to non-auto EVGA's 'auto' multiplier for mesh ends up 30 (I think) and sometimes I would see it spike and I would have occasional crashes. Setting mesh multiplier manually to 30 eliminated spikes and reduced frequency of crashes. Manually setting mesh multiplier to 24 gave me best stability. You might want to use HWInfo64 to check your mesh ratio.
 
P.S. Intel's processor diagnostic tool has brief health checks of CPU. Not as intensive as others but more tools you check with better are the chances of being confident / catching something.

arestavo

ZoranC
 
 
Would you mind sharing a screenshot of your BIOS CPU settings and what HWInfo reports are your CPU multipliers? If memory serves me well EVGA's auto settings result in 0 offset for AVX2/AVX3 which was too aggressive for my 10920X. Easiest way to find out which instruction set Star Citizen might be using is by using two different offsets for AVX2/AVX3 and then observing did frequency drop while running and by how much.
 
Word of warning: When using auto for CPU mesh multiplier ends up 24 (Intel's native). But if you switch CPU multipliers to non-auto EVGA's 'auto' multiplier for mesh ends up 30 (I think) and sometimes I would see it spike and I would have occasional crashes. Setting mesh multiplier manually to 30 eliminated spikes and reduced frequency of crashes. Manually setting mesh multiplier to 24 gave me best stability. You might want to use HWInfo64 to check your mesh ratio.
 
P.S. Intel's processor diagnostic tool has brief health checks of CPU. Not as intensive as others but more tools you check with better are the chances of being confident / catching something.

I'm still running my backup so I can't get a screencap of the BIOS settings at the moment. Intel's processor diagnostic came back good using their tool. That mesh multiplier changing sounds like it may have contributed to my crashes back when I was overclocking. I'll have to keep that in mind and manually set it. The mesh is currently 24 with auto as per HWinfo64.

It was worth trying Intel's Processor Diagnostic even though it giving CPU clean bill of health isn't guarantee everything is OK. Post BIOS screenshot once you can and we will take it from there.

So I hit the stability testing pretty hard with OCCT and aida64 - I think that I've got a handle on it now. It was likely my core 5 (if I remember correctly) as it's a pig on voltage, plus a combo of my vmesh spiking. 
 
Since locking down the vmesh to 30 and matching it's voltage to same as the VCCIO (1.15V iirc), everything has been good. I even later tried the OC Robot for per core OCing and it's been solid in Star Citizen with AVX (3) and AVX 2 (5) offsets in place. That poor core 5 only does 3900MHz while my core 12 does 5000. I may look at a remount since I did install a new Arctic Liquid Freezer 2 420 before doing the overclocking.
 
Thanks to everyone that helped, and thanks to this great guide that was really good for understanding X299 overclocking - https://www.hardwareluxx....2066-oc-guide.1172969/

ZoranC

arestavoSo I hit the stability testing pretty hard with OCCT and aida64 - I think that I've got a handle on it now.

Yay! I’m very glad to hear things are working now and we were of help :) Would you mind, please, sharing screenshots of your BIOS settings under CPU tab so I can compare mine to them?
 
While on the subject of VMesh, mine seemed OK at 30x but eventually I lowered it to 24x because I couldn't measure any performance difference between the two in real world _BUT_ CPU temperatures were measurably higher when using 30x, ESPECIALLY with AVX2 and AVX3 instruction sets. Do you have any real world benchmark data that justifies overclocking mesh, please?
 
P.S. Thank you VERY much for sharing that link on X299 tuning! I wish I could understand German :( :( :(

I just went off the guide (I use the Brave browser which is chromium based and have the google translate plugin installed) as it stated that 30 on the mesh ratio should be doable for just about all processors - mine happens to work at fine at 1.1V with a matching VCCIO of 1.1V. I didn't look into what the benefits were other than remembering (from overclocking reviews years ago) that it helped in CPU bound games, and Star Citizen is still very CPU bound.

ZoranC
Thank you for sharing screenshots of your settings! If you don't mind once I have a chance to go over them I might ping you with questions about them. In the meantime two quick things, please:
 
1. Is your Intel Turbo Boost 3.0 Driver Support BIOS setting enabled or disabled?
 
2. It would be interesting to see what difference you see in performance and for temperatures x30 vs. x24 Vmesh, would Star Citizen benchmark show same thing I observed.

I do have turbo boost 3.0 enabled. I'll have to see if I notice a difference in Star Citizen, yet performance is still heavily influenced by the particular server's health (it's still very much an alpha) - so it may difficult to see if the difference isn't big.

@arestavo: I took a quick look at your settings and one thing about them is confusing me, which might be due to my lack of knowledge. If we take a look at your core multipliers left side is what I think Intel says they tested each core can do and they provide voltage for while right one is what you set them at. So you have overclocked some cores but underclocked others. In turn best performing cores will hit 5GHz which is only 4% improvement (50/48) but total processing power of all cores together (sum of left side ratios vs. sum of your ratios) went DOWN 2% with your settings (650/638).
 
If I am correct wouldn’t you be better off using Intel’s settings? You would be getting better total computing power at overall lower temperature and in turn getting better stability?

arestavo

ZoranC
@arestavo: I took a quick look at your settings and one thing about them is confusing me, which might be due to my lack of knowledge. If we take a look at your core multipliers left side is what I think Intel says they tested each core can do and they provide voltage for while right one is what you set them at. So you have overclocked some cores but underclocked others. In turn best performing cores will hit 5GHz which is only 4% improvement (50/48) but total processing power of all cores together (sum of left side ratios vs. sum of your ratios) went DOWN 2% with your settings (650/638).
 
If I am correct wouldn’t you be better off using Intel’s settings? You would be getting better total computing power at overall lower temperature and in turn getting better stability?

No clue.
 
I got the another WHEA uncorrectable BSOD and stopped using per core again. The OC robot settings were not fully stable, and so I am now trying a manually selected 4.4GHz all core at 1.215V on the VCORE.
 
With the 4.4GHz all core overclock while playing my CPU bound game (Star Citizen), I'm getting less 0.1% lows that kids these days call stuttering (vs. the OC robots per core OC). And while that game is still very much CPU bound, the 4.4GHz all core overclock is raising the average FPS when in places where I'm CPU bound vs stock CPU clocks. I did bump up the VCCIO and VSA up from 1.1V to 1.15V and locked the mesh frequency multiplier to 24 (instead of 30) to see if that would be stable. Temperatures aren't that far from what I had with the per core OC - averaging roughly around 65C vs 72C (eyeballing it, not using a program to parse the data).

In my experiments I am most stable when I have CPU multiplier control as 'Per core', multiplier for all cores set to Intel's official all core boost frequency (that is 43 for my 920x, it would be 41 for your 940x), and mesh at Intel's official x24. I also have CPU states and turbo boost 3.0 disabled. Don't forget to manually set TJMax to whatever it is official for your 940x, auto support on X299 Dark is broken.
 
My theory behind this is that Windows and/or BIOS do not seem to handle variations in frequencies between cores correctly, especially when it comes to AVX2/AVX3 offsets. If memory serves me well (and don't quote me on this, it has been a while) when I tried to use something with AVX2/AVX3 instruction sets final frequency wouldn't be offset for each core but all cores would end up running at whatever best core can do minus the offset. In the end I gave up chasing few extra percent of performance in exchange for stability, it wasn't worth my time.
 
So you might end up doing same.

arestavo
So I kept getting WHEA uncorrectable (code 124) again and again - even after upping VCORE to 1.3V. I gave up and went back to stock - and am still getting those WHEA uncorrectable BSODs again. Well, my RAM is at 3600MHz so I'm going to have to try 3200 and see if it stops. If not, I'm plum out of ideas.

If you are getting WHEAs at stock settings then something is seriously off and memory would be one of primary suspects but I was under impression you said you already tested it. How many passes of MemTest86 you tested with and what are your test parameters? FWIW, I am using G.Skill.
 
Also, have you eliminated possibility it isn't your PSU / video card that is triggering it?

I gave up - I'm going to get an AMD 5800x3d and an X570 with 4 PCIE slots when black friday deals come out.

Too much oc’ing messed up your cpu. 124 bsod points to cpu issue. If you can’t stabilize with more voltage, or by going back to stock settings then you’ll need to rma.

If the cpu has warranty then I don’t see why they wouldn’t.