Folding Forum

After experimenting with overclocking over the last couple weeks, there is a clear correlation between clock speed and getting BAD_WORK_UNIT (114 = 0x72) errors. Even an increase of only 130 MHz on GPU causes starts causing errors sometimes.

vtankovich wrote:I've noticed that I have a small chance to get an error when I try to use the GPU while it's folding - games or even highres video playing or a lot of "heavy" tabs open at the same time.

Running games and folding would cause issues for both as both applications are competing with each other over the GPU. Would not recommend that unless you have a high-end GPU, the game doesn't use too much resources and you don't mind getting varying frame rates. In that case, you may try that but not suggested.

I have no issues playing back 4K and 8K videos on my system while folding on the GTX 1080 Ti. I also have no issues with several tabs open that are content rich. However, my PPD did drop and I am fine with that as this isn't a dedicated system. IMO, it boils down to the type of GPU you have. A high-end one will allow a mostly seamless experience while folding on the GPU and allowing you to perform day-to-day tasks. Mid-range might struggle under some conditions while low-end may cause screen lag which can be partly resolved by disabling hardware acceleration.

bikeaddict wrote:After experimenting with overclocking over the last couple weeks, there is a clear correlation between clock speed and getting BAD_WORK_UNIT (114 = 0x72) errors...

Welcome to the F@H Forum bikeaddict,

Please note that testing the overclock with live WUs isn't ideal at all as it means you're delaying science and causing WUs to be unnecessary duplicated F@H doesn't assigned the same WU to multiple donors for verification, instead, all WUs are unique to that individual donor under normal conditions.

bikeaddict wrote:After experimenting with overclocking over the last couple weeks, there is a clear correlation between clock speed and getting BAD_WORK_UNIT (114 = 0x72) errors. Even an increase of only 130 MHz on GPU causes starts causing errors sometimes.

'Only' 130 MHz - WTF! Days of that sort of overclocking are gone.

If the card could run at that speed with any degree of stability, don't you think it might have shipped something like that?

In reality, with a lot of understanding and trial and error in 5 MHz steps, you might be able to improve on the GPU manufacturers millions of dollars research slightly, probably at the expense of some power efficiency.

Cards are already 'factory overclocked' to the nth degree - GPU Boost 2.0 or whatever they call it now. For example, the 2080Ti founders edition has a 'base clock' or 1350 MHz. That is what it is guaranteed to do all day everyday no matter what you throw at it.

However, there is then the nominal boost clock of 1635 (which is in turn an overclock of a 'stock' 1545 boost). Provided various heat, load and power conditions are met, the card dynamic overclocks making many measurements and adjustments to real time optimise.

In fact, if conditions allow, the card can choose to exceed that. I've seen 1760 MHz quite often.

Now, throw another 130Mhz on that and we've now got a total overclock of 540Mhz - quite extreme and although games might run and the odd pixel in a million is slightly the wrong shade, who cares.

Try and do molecular dynamic simulations and 1 atom ends up colliding with another and the project goes boom.

In summary, manual overclocking is naive, counter productive and should be discouraged.

FAH was processing multiple WUs with no errors with GPU clock boosted 100-120 MHz. The machine is completely stable with even 150 MHz increase (GTX 1650 Super) while loading pages or playing full screen 1440p YouTube videos with Firefox using hardware acceleration.

But I've given up on overclocking to avoid wasting any more processing time. I want to see how high I can get my daily avg. points. It's been in the 400K range and I'd like to see if I can top 500K.

bikeaddict wrote:FAH was processing multiple WUs with no errors with GPU clock boosted 100-120 MHz. The machine is completely stable with even 150 MHz increase (GTX 1650 Super) while loading pages or playing full screen 1440p YouTube videos with Firefox using hardware acceleration.

Without wishing to sound rude, but that sort of 'testing' is equivalent to testing a hot-rodded nitrous injected engine by driving at 50 down a highway. Only on a racetrack will you find if it's going to blow.
None of that is going to tax any of the hardware or frequencies that folding does. I doubt it even touches the shaders nor invokes any degree of boost - the card has dedicated pathways for that sort of media rendering.

bikeaddict wrote:But I've given up on overclocking to avoid wasting any more processing time. I want to see how high I can get my daily avg. points. It's been in the 400K range and I'd like to see if I can top 500K.

Cool, I only wish more would too and stop wasting WUs

I don't know the settings for a GTX 1080Ti.
But I would suggest doing without an overclock for a while.
Also, keep temps down, by increasing fan speed, and possibly cutting down on power a bit (in case you're running your PSU at 85% or higher of it's max rated load.

I tried this with my old GPU GT9500 and it broke forced me to buy another gpu xD I replaced it with https://www.adorama.com/pn16606sdo.html

Modern GPUs are good at "overclocking" themselves up to a stable rate without any input required from user. Mine is advertised to have a boost clock of 1750ish, and it actually runs between 1890-1935, depending on temp. So all I had to do was to increase power limit to 110% to reduce throttling and set fan to 65%, as cooler temp allows for slightly higher clock.

bikeaddict wrote:...But I've given up on overclocking to avoid wasting any more processing time.

As I said on the previous page, FAH does not support overclocking. FAH designes their software to take advantage of every bit of the hardware that is designed into it. Call it a benchmark if you want to, but it's highly unlikely that you can generate more points by overclocking that you'll waste with the crashes you generate. If you want to play the overclocking game elsewhere, fine, but stop trying it here.

Just want to throw out a bit of what i have experienced...

5500XT Asus OC Evo - CONSTANT bad work units - stock (ASUS stock OC). Downloaded OC Tweaker - lowered clock all the way to AMD stock (~1620) met with more bad work units. Lowered wattage to about 60% and increased fan speed... Still hit with bad work units. Those steps are just the more significant ones that I took, I spent almost 2 days on this; irritated because my 5700XT has had 0 issues.. all while folding and playing COD Warzone on High settings...

My (potential) fix: I uninstalled GPU Tweak and installed the AMD Adrenaline software and UNDERVOLTED the GPU... left it on overnight with no issues. (going on about 10 hours now)






I really hope this helps anyone with similar issues.

Edit: Typo

AMD seems to run their GPU voltages quite high compared to Nvidia.
On Nvidia we're running 0.9-1.05V on average.

rwh202 wrote:

bikeaddict wrote:After experimenting with overclocking over the last couple weeks, there is a clear correlation between clock speed and getting BAD_WORK_UNIT (114 = 0x72) errors. Even an increase of only 130 MHz on GPU causes starts causing errors sometimes.

'Only' 130 MHz - WTF! Days of that sort of overclocking are gone.

If the card could run at that speed with any degree of stability, don't you think it might have shipped something like that?

In reality, with a lot of understanding and trial and error in 5 MHz steps, you might be able to improve on the GPU manufacturers millions of dollars research slightly, probably at the expense of some power efficiency.

Cards are already 'factory overclocked' to the nth degree - GPU Boost 2.0 or whatever they call it now. For example, the 2080Ti founders edition has a 'base clock' or 1350 MHz. That is what it is guaranteed to do all day everyday no matter what you throw at it.

However, there is then the nominal boost clock of 1635 (which is in turn an overclock of a 'stock' 1545 boost). Provided various heat, load and power conditions are met, the card dynamic overclocks making many measurements and adjustments to real time optimise.

In fact, if conditions allow, the card can choose to exceed that. I've seen 1760 MHz quite often.

Now, throw another 130Mhz on that and we've now got a total overclock of 540Mhz - quite extreme and although games might run and the odd pixel in a million is slightly the wrong shade, who cares.

Try and do molecular dynamic simulations and 1 atom ends up colliding with another and the project goes boom.

In summary, manual overclocking is naive, counter productive and should be discouraged.

Eh... I don't think you're really correct here. I have a 1080ti running at +186MHz. Combine that with factory boost, and it's running at 2050MHz core compared to an official default clock of 1480, for an overall 570 over the advertised clock. Despite that, it's been folding 24/7 for months with the only BAD_WU errors happening on 13404 and 13405 WUs, which are experimental WUs known to be prone to some error and instability. All other WUs complete 100% without error.

Many modern chips still have headroom for overclocking.

In addition, many people here have stated that folding is a harder workload than furmark. This is plainly and obviously false. Most folding WUs run between 65 and 100% TDP on my 1080ti, with some occasionally hitting 115% or so (I have the power limit set to 120%). If I remember right, the high load ones are 14415 and similar (the myosin projects) - I've seen those consistently hitting my GPU harder than most other projects. Even so, my GPU is able to maintain 1950MHz or above on basically any folding workload. Furmark on the other hand is able to push my GPU all the way to its power cap at 120% TDP with the card clocked all the way down at 1750MHz. No folding workload is this demanding. Not even close.

However, where Folding is more demanding is in tolerance of small errors - Furmark will still run with an occasional bad bit or small error. Folding will not tolerate any errors at all, so what seems to be stable in furmark may end up not being stable after all. In no way is folding as demanding of a workload for the GPU though.

rwh202 wrote:

bikeaddict wrote:After experimenting with overclocking over the last couple weeks, there is a clear correlation between clock speed and getting BAD_WORK_UNIT (114 = 0x72) errors. Even an increase of only 130 MHz on GPU causes starts causing errors sometimes.

'Only' 130 MHz - WTF! Days of that sort of overclocking are gone.

If the card could run at that speed with any degree of stability, don't you think it might have shipped something like that?

In reality, with a lot of understanding and trial and error in 5 MHz steps, you might be able to improve on the GPU manufacturers millions of dollars research slightly, probably at the expense of some power efficiency.

Cards are already 'factory overclocked' to the nth degree - GPU Boost 2.0 or whatever they call it now. For example, the 2080Ti founders edition has a 'base clock' or 1350 MHz. That is what it is guaranteed to do all day everyday no matter what you throw at it.

However, there is then the nominal boost clock of 1635 (which is in turn an overclock of a 'stock' 1545 boost). Provided various heat, load and power conditions are met, the card dynamic overclocks making many measurements and adjustments to real time optimise.

In fact, if conditions allow, the card can choose to exceed that. I've seen 1760 MHz quite often.

Now, throw another 130Mhz on that and we've now got a total overclock of 540Mhz - quite extreme and although games might run and the odd pixel in a million is slightly the wrong shade, who cares.

Try and do molecular dynamic simulations and 1 atom ends up colliding with another and the project goes boom.

In summary, manual overclocking is naive, counter productive and should be discouraged.

On a low load, some of my RTX 2080Tis can run well over 2100Mhz (2150Mhz sometimes).
Nividia really upped the standard with each production. The earlier models could hit 1935Mhz, occasionally.
However, under load, most of my GPUs run at 1750-1875Mhz.
Big projects on Core 22 put a heavier toll on the GPU, in as such as when I'd run an RTX 2080Ti under 170W, it'll drop so drastically in speed (1500Mhz) that it's not even worth running it there. 180-190W minimum. Though from 190-220W it is running in the 1875Mhz range, which imho is a good balance between energy efficiency and frequency.
Once I go over 225W (~1905-1935Mhz) the gain in frequency becomes so little it also doesn't make sense.
You could run it at 300W, and get 2050Mhz. But at 200W you'd get 1875Mhz. That's 175Mhz slower, but 1/3rd less power.

bruce wrote:

bikeaddict wrote:...But I've given up on overclocking to avoid wasting any more processing time.

As I said on the previous page, FAH does not support overclocking. FAH designes their software to take advantage of every bit of the hardware that is designed into it. Call it a benchmark if you want to, but it's highly unlikely that you can generate more points by overclocking that you'll waste with the crashes you generate. If you want to play the overclocking game elsewhere, fine, but stop trying it here.

I've been folding on overclocked systems for years, and you absolutely will generate more points and science (this is plainly true, since the system is faster).

Folding is very demanding that you have a stable overclock of course, and a lot of marginal overclocks that work just fine on games don't work for FaH, but it's false to state that overclocking doesn't work on FaH. My current system has a 3.3GHz CPU overclocked to 3.9 and a GPU with a 186MHz boost, and it folds for months with no bad WUs at all. The key is just in making sure your overclock is stable and in making sure you have adequate cooling.