Einstein causes PC to restart, Milky Way works fine
16 Dec 2022 0:42:14 UTC
Topic 228732
(moderation:
I'm currently using BOINC 7.20.2 and running Einstein@home and Milkyway@home. My computer consists of:
Ryzen 5800X3D
MSI B550-A Pro
32GB DDR4 3600
Radeon 6800XT
Seasonic 750w 80+ titanium
Most recent Windows 10 update
Most recent BIOS update
Computer is custom water cooled
When running einstein@home tasks, the computer abruptly restarts within a few minutes. No blue screen, no warning, nothing. I attempt looking in Windows event viewer but searching the specific error doesn't get me anywhere and I don't know enough about Windows errors to narrow my search. It's always a "bug check" that causes a sudden reboot followed by lots of digits in parenthesis.
I can run Milky Way all day long with CPU and GPU tasks. If I disable GPU tasks and try to run einstein, it still restarts after a few minutes. If I change the CPU utilization setting to 25%, I can get more time but it always eventually restarts. Sometimes I'll get a Radeon driver pop up message after Windows loads saying it detected a black screen or crash and asks if I want to submit a crash report.
This computer is my VR gaming computer. I recently formatted the hard drive and reloaded Windows due to frequent crashes with Flight Simulator 2020, Half Life Alyx, and BOINC. After reloading the OS, drivers, games and everything else, I did lots of trial and error with Steam VR settings and all my VR games work perfectly now. FS2020 needed extra in-game tweaking but it also hasn't crashed yet. Einstein is the last hold out.
Some settings that could potentially impact BOINC stability: in BIOS I have "MSI Kombo Strike" activated for the 5800X3D. It is an overclocking feature specific to the 5800X3D alone. I also have smart access memory activated. The other name for it is resizable BAR. I have my Windows power plan set to "high performance". I don't think the computer is running out of juice or getting too hot. I think this is a software issue. Here are some screen shots of MSI Afterburner after running Milky Way for over an hour.
If you think any info from event viewer would be helpful please let me know.
Just a quick guess, but the
)
Just a quick guess, but the power supply would cause all of the symptoms you describe including crashing while gaming.
Einstein may be drawing more resources than your other project causing the supply to trip. I personally have had this happen with a Seasonic 850. I had to move to the Seasonic 1300 watt to eliminate all random crashing. While your 750 watt SHOULD be enough, with your system I wouldn't go any less than an 850 and quite frankly Id buy a 1000w, just cause.....
Adam wrote: When running
)
I firmly agree with Greg (Gandolph1). Get a 1000W PSU as a minimum, Platinum or Titanium rated. 1300W or even higher if you can afford it. My preference would be either EVGA or Seasonic, with first choice being an EVGA 1300W Platinum. It should fit in your computer case.
For your GPU a 750W PSU is rated as a MINIMUM, and that is from AMD. That assumes that you don't do any overclocking, and are doing 'normal' things with your computer. It does not include running BOINC projects 24/7.
Proud member of the Old Farts Association
Adam wrote:Some settings that
)
Of course you should first try to get your system at default speed to run stable and than see how (and if) you can overclock it. Not every system can be overclocked, specially not for running 24/7 at BOINC load.
Don't think/try to guess, check.
But I agree with the two posts above, all you write sounds like not stable power delivery.
.
Thank you everyone. Since my
)
Thank you everyone. Since my post I've tried turning off Kombo Strike and that didn't make a difference. The pictures I posted show that temps for the CPU and GPU are well below their thermal throttling points so that's why I've ruled out thermals.
Does anyone know the name of the tool I need to measure power draw from the wall? I'm able to see with software that the CPU is pulling around 100w and the GPU can pull as much as 265w. I've seen tech tubers like Linus and Steve use the tool but I'm not sure what it's name is.
I should mention I have a second very similar computer that I use for pancake gaming and einstein works fine on that one. I didn't mention it yet to try and minimize the already huge wall of text in my initial post. The second computer consists of:
Ryzen 5800X
MSI X570 Gaming Edge Wifi
32GB DDR4 3600
Same Radeon 6800XT
Same Seasonic 750w 80+ titanium
Same Most recent Windows 10 update
Older BIOS version
Older AMD graphics drivers from 1 year ago
Same custom water cooled setup
It's also using PCIe gen 3 instead of 4 due to the riser cable, no smart access memory, no kombo strike. I agree when computers abruptly restart that sounds like slam dunk power supply issue. I just want to be absolutely sure it's not software before dropping a good chunk of change on a kilowatt PSU especially when my other very similar computer with the same PSU is running einstein with no problems.
Adam wrote:Does anyone know
)
Kill-a-Watt meter is the most common, but there are many others depending on your budget and preference.
https://www.amazon.com/s?k=power+draw+meter&crid=2UP42WC9BKXS&sprefix=power+draw+meter%2Caps%2C106&ref=nb_sb_ss_ts-doa-p_1_16
Keep in mind that the useful limit of drawing power from the wall is about 75% of the rating of your power supply for drawing power 24/7 with overclocking. The point that we're trying to make is the voltage spikes that come from your graphics card (GPU). Some are higher than others, and some power supply units (PSU) will suppress power spikes better than others.
You'd be better served to not have a power supply with a rating below 1000W, regardless. The higher the better, and also with a 80+ rating of Platinum or Titanium.
.....[EDIT].....
BTW, your GPU can draw as much as 300W TDP, not 265W.
https://www.techpowerup.com/gpu-specs/radeon-rx-6800-xt.c3694
Proud member of the Old Farts Association
From the looks of it you are
)
From the looks of it you are dumping on the new GW tasks which put a much bigger load on both the cpu and gpu. No issues with the much easier GR and Meerkat tasks. And of course the Separation tasks at MW are a trifle for any gpu.
You need a larger power supply to handle the increased loading of the gpu from 0-90% and the big load on the cpu for the final 10% toplist calculation on the cpu for each task.
Kill-a-watt meters and the
)
Kill-a-watt meters and the like are great for figuring out your actual average power consumption from a cost of computing point of view. They are useless for the problem at hand, as they don't have fine enough time resolution.
A modern GPU can and does pull transients of far higher than average power consumption.
Plenty of capacitors exist both on the power supply and inside the system so that the shorter-duration spikes may have little effect--but something sneaks through. If ANYTHING in the system does not like the resulting transients, trouble may come.
More likely, if you have outright restarts, the power supply's internal protection circuitry detected a transient for which the combination of width (time duration) and height (amperage above a limit) exceed the actual triggering of protection.
If you want to instrument that, you might look into just using a low-value (extremely) resistor in series with the wall supply line and an oscilloscope. Good luck finding out what value you need to avoid, but at least you'd have the capability of comparing different scenarios. If you peak inside a Kill-a-watt, you'll find it is measure voltage across a low-value resistor.
The one thing I'd exclude is that the Einstein software has some magic flaw, which mysteriously figures out how to activate on your system, but not on thousands of other user systems running the same code.
You are not the first, and won't be the last, to post here something on the lines of "everything else works, so it must be that Einstein is at fault". They all forget the other side of the coin "everybody else's system does not have this problem, and yours does--so there has to be something different (and quite likely faulty) about your system".
Good luck.
Keith Myers wrote: From the
)
How were you able to tell that? I'd love to know for future troubleshooting.
Thanks for the advice everyone. I was just hoping to exhaust my free options first (driver problem, Windows settings etc) since a Seasonic or EVGA 1000w+ titanium PSU are easily $300+.
I turned kombo boost back on and found it actually reduced my CPU power usage from ~100w to 88w while boosting clock speed so kudos to MSI for figuring out how to pull that off. I also rearranged the cables on my power supply since it's fully modular thinking maybe there was a faulty plug. Neither made a difference, computer restarted after just a few minutes of running Einstein. Sounds like it's not worth pursuing measuring the computer's power draw. I'll start thinking about my next move.
Adam wrote:I was just hoping
)
One free option that comes to my mind is swapping those two "identical" power supplies. Maybe the other can handle this system better.
.
True! I was originally
)
True! I was originally thinking about swapping the identical GPU's but that would require the time consuming process of draining and refilling the coolant. Swapping PSU's would take no time since they're modular, all the wires can stay put. Can't believe I didn't think of that ugh! Also I realized they aren't identical, the other one is a titanium 750 and this one is a platinum 750. Maybe that's all this one needs? I'll report back what happens, thank you.