Folding Forum

Without knowing whether its part of the problem - I would recommend having the CPU (which model btw) not above 75°C. CPUs tend to be a bit more sensitve than GPUs. By any chance: Do you have OC as well? Your GPU comes with factory-side OC and boost as well. You might want to try stabilty with a tool such as Furmark3D with regular settings. If it produces the same kind of errors - disable all OC, clock the card back to recommend base clock speed. Could be one source of error, althought your symptomps really dont sound good.

Not sure I quite agree on the CPU temps … Better to check what you CPUs TJmax (or equivalent) are as many current CPUs will readily function 24/7 in the 90C to 100C range before they start to throttle speeds … The fans may well be "noisy" but they should cope well on a properly configured system … Fairly sure that even latest GPUs really aren't happy when that hot.

Reducing clock speeds and under-volting the GPU has been discussed in other threads where then has been issues with GPU stability … A slightly weak (but not weak enough to fail manufacturing spec tests can sometimes be stressed beyond where it likes to be by the intensity of the work of a FAH core … Hopefully on of the GPU gurus who has done this to stabilise their own systems will spot this and join the conversation.

If the CPU is continuing to fold after the freeze has happened and the GPU slot has failed it might hint at GPU rather than CPU?

If you don't have it, download and install MSI Afterburner.

Once running it will show various temperatures including your cpu and gpu. Try to re-start your folding and have Afterburner open.

Afterburner is typically used for overclocking but it can also be used to manually increase fan speed on your 1070ti and also if it becomes necessary to downclock it by reducing the power limit to say 80%.

You say the card runs at 50C normally? That is simply because the NVidia BIOS won't bother to run the fan unless the card is getting hot. Once fully loaded, the fan profile will try to keep the core temperature to around 80-85C.

I had a 1070ti and ran the fan at 80% because that was still very quiet but kept the temps to low 70s.

Neil-B wrote:If the CPU is continuing to fold after the freeze has happened and the GPU slot has failed it might hint at GPU rather than CPU?

Also my thoughts, hence the OC hint. I just dont know if OC is enough of an explanation for this utter failure, it killed the driver. Then again, I'm not a hardware professional. Wanna agree with HaloJones on the next steps.

Neil-B wrote:Not sure I quite agree on the CPU temps … Better to check what you CPUs TJmax (or equivalent) are as many current CPUs will readily function 24/7 in the 90C to 100C range before they start to throttle speeds … The fans may well be "noisy" but they should cope well on a properly configured system … Fairly sure that even latest GPUs really aren't happy when that hot.

I would not let my machine run continuously at a setting where CPU auto-throttling takes place on a recurring basis. Personally, I give at least 20° buffer between observed peak temp and TjMax. On my i7-4770 rig, I found that disabling Turbo mode reduced CPU temp significantly. Since it is a backup machine where only F@H runs actively, raw CPU speed is not important to me.

For the OP, you still need to monitor voltage and temp on that GPU. If those are both within reason, you might try uninstalling the GPU driver and trying a new (or older) version. Sometimes a particular driver/GPU combination just doesn't play nice. Make sure you completely uninstall the old driver before installing a new one (utilities such as DisplayDriverUninstaller are available to help get rid of undesirable Registry entries and leftover files), and disable any automated updater. Once you find a driver that works reliably, keep it!

Good that you have a bigger power supply, it will make your system much more stable for sure.

Warning 2020-04-17 12:49:41 PM Display 4101 None
Display driver nvlddmkm stopped responding and has successfully recovered.

I almost suggested to increase the timeout of the gpu / driver in registry, but you already did that. (I had the same issue when first fooling around gpu computing)

16:57:00:WU02:FS01:0x22:Folding@home Core Shutdown: BAD_WORK_UNIT
16:57:01:WARNING:WU02:FS01:FahCore returned: BAD_WORK_UNIT (114 = 0x72)

the log says bad work unit. Do you have the error with new work unit too? probably the error wasn't on your side, sometimes bad workunits happen.

Platform: win32 10
16:27:02: Bits: 32

Do you use 32 bit version of the software? why? you seem to have 16GB of ram, and a 64 bit processor.
Basically everything that is younger than circa 2006, can run 64 bit code.
I don't know why the 32 bits stuff is still considered as some "default" option for many people. it was already obsolete a decade ago!

Temperature of the CPU was around 87 Celsius,

I guess this is your processor:
https://ark.intel.com/content/www/us/en ... 0-ghz.html
and it says TCASE 72.72°C , Case Temperature is the maximum temperature allowed at the processor Integrated Heat Spreader (IHS).
so 87 celsius is really too high for that processor, and for a 84 watts TDP it means that your cooling fan is dusty or that you use some stock cooler...

Airflow should be okay because you said

I have a front intake and back exhaust fans, each ~120 mm.

I suggest to stress-test the CPU with CPUz to see if the system freezes due to this or not.

I have a GIGABYTE AORUS GTX 1080 Ti. I used a monitoring app and the fans are spinning and GPU temp goes to a high of about 79 Celsius. I've actually seen the system freeze when it drops a bit to 75 Celsius.

Your GPU temperatures may be not bad, although if you already dust off the cpu you can clean the whole case and all coolers at once.

Darth_Peter_dualxeon wrote:

Platform: win32 10
16:27:02: Bits: 32

Do you use 32 bit version of the software? why? you seem to have 16GB of ram, and a 64 bit processor.
Basically everything that is younger than circa 2006, can run 64 bit code.
I don't know why the 32 bits stuff is still considered as some "default" option for many people. it was already obsolete a decade ago!

The client program is 32 bit, no need for 64. The folding cores are 64 bit. His system shows as 64 bit.

16:27:02:            CWD: C:\Users\#######\AppData\Roaming\FAHClient
16:27:02:             OS: Windows 10 Enterprise
16:27:02:        OS Arch: AMD64
16:27:02:           GPUs: 1

Jan wrote:Without knowing whether its part of the problem - I would recommend having the CPU (which model btw) not above 75°C. CPUs tend to be a bit more sensitve than GPUs. By any chance: Do you have OC as well? Your GPU comes with factory-side OC and boost as well. You might want to try stabilty with a tool such as Furmark3D with regular settings. If it produces the same kind of errors - disable all OC, clock the card back to recommend base clock speed. Could be one source of error, althought your symptomps really dont sound good.

Thanks for the suggestion. I ran FurMark3D (what a funny name) for 20 minutes with the default settings. Essentially 1024x768 resolution, no AA. Temperature stabilized at 79 Celsius. Power was ~100% almost the whole time. No issues seen during the 20 minutes, which was almost twice as long as it has run Folding@Home on a good day.

I also ran the OpenGL benchmark at the 1080 Preset. The results are not the highest among recent reports for the same level 1080 TI, with a score of 9811 points.

https://gpuscore.top/furmark/show.php?id=49946

Temperature of the CPU was around 87 Celsius,

I guess this is your processor:
https://ark.intel.com/content/www/us/en ... 0-ghz.html
and it says TCASE 72.72°C , Case Temperature is the maximum temperature allowed at the processor Integrated Heat Spreader (IHS).
so 87 celsius is really too high for that processor, and for a 84 watts TDP it means that your cooling fan is dusty or that you use some stock cooler...

Close. Mine is the 3.5 GHz version, not the 3.9 GHz version.

I suggest to stress-test the CPU with CPUz to see if the system freezes due to this or not.

I did this, and didn't see CPU temperature go over 73 Celsius. I use HWiNFO64 to probe the temps at the Cores. There could be something to it going over 73 for Folding@Home though. The HWiNFO64 registered a maximum of 94 Celsius at some point while Folding only to come down to 87 Celsius, but it is also doing CPU folding at the same time. I've not found that disabling the CPU fold changed the results. I should try again with the probes on.

I do have a stock Intel cooler on this CPU. The heat sink looks pretty clear though.

I ran Folding@Home again, with only GPU folding (removed the CPU slot) and had MSI Afterburner running at the same time. It froze as before.

HWiNFO64 reported CPU core temps at 62 Celsius max. Most were 60 Celsius or below. I think we can rule out this being a CPU heating issue.

At the time of the freeze, MSI Afterburner was reporting 70 Celsius for the GPU. The GPU clock was 1936 MHz and the Mem clock was 5005 MHz. I'm not quite sure how to read this but in the GPU clock circle, the white "Base" triangle is set around 1550 MHz and the red "Boost" triangle is set about 1700 MHz. 1936 MHz goes above that a fair ways. FurMark3D stayed pretty stable at the base frequency, whereas a different benchmark "Unigine Valley Benchmark 1.0" went as high as Folding at Home but didn't freeze after 10+ minutes with temperatures of in the mid-70s Celsius. No issues seen with those runs.

Could Folding at Home be even more stressing than these Benchmarks? Is there anyway to dial-back the folding speed?

Its possible that FAH may stress the GPU a bit more on a different level, producing this error. Its not the temps. I would go with decreasing the clock speed for maybe 100-200MHz and disabling any turbo features for further testing.

Have you considered Superposition instead of Valley Benchmark (https://benchmark.unigine.com/superposition)? Reason is that Superposition is more recent. Also, choose the highest screen resolution and graphic options to push the GPU harder.

However, please note that F@H uses the "compute" sections of the GPU heavily while most gaming benchmarks (including Superposition) use the "rendering" sections of the GPU heavily. Thus, stable while running isn't an indication of stability while folding... just an indicator to allow to you tweak.

LancerDL wrote:I ran Folding@Home again, with only GPU folding (removed the CPU slot) and had MSI Afterburner running at the same time. It froze as before.

HWiNFO64 reported CPU core temps at 62 Celsius max. Most were 60 Celsius or below. I think we can rule out this being a CPU heating issue.

What about voltages? Are you monitoring the 12V rail?

jrweiss wrote:What about voltages? Are you monitoring the 12V rail?

I'll focus on that on my next attempt. I'll also give the Superposition benchmark a try, as recommended.

Does anyone know what the white and red triangles are in the MSI Afterburner display? They're on the GPU clock circle, it's all I know. Saw a few games stay right at the white triangle, so am wondering what would make it go higher than even the red triangle? If it's supposed to be a limiter, it doesn't seem to be doing that for Folding@Home and at least one benchmark (though the benchmark didn't crash).

FYI, I haven't done any Overclocking, unless something came that way.

So the white traingle is the base speed, the red triangle the boost speed. Now the whole thing about NVidia Boost is that the BIOS base and boost settings are kinda guaranteed numbers for the chip. When the system is being used, the cooling capabilities will potentially allow the card to go over those numbers and if you've moved any of the sliders in Afterburner to the right you've told the card it's allowed to go higher.

So, for example, you can't really affect the voltages any more so the top slider doesn't do much for the 2xxx series. The Power Limit slider tells the card it's allowed to go over 100% which will give it one option to go faster. The Temp Limit says to the card that it's OK to potentially go over the 80C that NVidia prefer. The Core Clock is for you to actively tell the card to go faster than Boost and then Memory Clock simply clocks the memory faster (or slower, up to you).

So, for example, my main desktop rig has a water-cooled 1070. Its base/boost is 1700/1900 but because I have the Core Voltage at +100 (probably being ignored although the right hand dial shows 1093mV), the Power Limit at 108%, the Temp Limit at 92, the Core Clock at +132 and the Memory Clock at +201, the card is currently hitting 2062MHz at 46C.

Without that cooling, it wouldn't probably get to such a high clock, but NVidia's Boost technology really likes low temps.