FAHBench (OpenMM 5.1)

foldy · Post by **foldy** » Sun Apr 12, 2020 4:21 pm

FAHbench 2.3.2 for Windows build with OpenMM 7.4.1 Core_22 with mixed precision
https://www.file-upload.net/download-13 ... n.zip.html

So that should match the current FAH GPU core for Corona Virus work units.

Assimilator1 · Post by **Assimilator1** » Tue Apr 14, 2020 12:14 pm

PantherX
Thanks for the update to the benchmarks

, totally agree about no point in gathering more for now.

So what is the long list of stress tests you use then?

I was thinking of using OCCTs GPU stress test, but I wonder if it's artificially throttled as (IIRC) it's counted as a 'power virus'. As is furmark.
So I would like to know what is a realistic GPU stress test (as in not throttled, or drawing an unrealistic power load) with error reporting?

Neil-B · Post by **Neil-B** » Tue Apr 14, 2020 1:13 pm

You could tape over half the cooling fan slots to simulate animal hair/dust/other detritus that might realistically build up over time? … or put a cardboard box around the back of the case to simulate the tight spaces the kit might be squeezed into in an overcrowded home-office? … or randomly pull out the power lead to simulate spouse/kids wanting to plug in their stuff? … Realistic might vary dependant upon your circumstances …

Post by **toTOW** » Tue Apr 14, 2020 8:57 pm

foldy wrote:FAHbench 2.3.2 for Windows build with OpenMM 7.4.1 Core_22 with mixed precision
https://www.file-upload.net/download-13 ... n.zip.html

So that should match the current FAH GPU core for Corona Virus work units.

I tested your version, but I'm seeing weird things on DHFR : it's not able to use 100% of my 1070, only 65% ... I can't explain why. GPU load is better on NAV with 95%.

There's also a CUDA option in your build, but it doesn't work.

Assimilator1 · Post by **Assimilator1** » Wed Apr 15, 2020 2:25 pm

Neil-B wrote:You could tape over half the cooling fan slots to simulate animal hair/dust/other detritus that might realistically build up over time? … or put a cardboard box around the back of the case to simulate the tight spaces the kit might be squeezed into in an overcrowded home-office? … or randomly pull out the power lead to simulate spouse/kids wanting to plug in their stuff? … Realistic might vary dependant upon your circumstances …

Riiight, something like a stress test which doesn't trigger 'power virus' restrictions along with error reporting is more what I was thinking of

.

foldy
Only just noticed your FaHbench link! I'll try it now

.
[update]Ok I've tried it now, at 1st I was getting very on/off GPU load, I then paused BOINC (but it had left a CPU core free for normal folding anyway, which works fine there) & I got a consistent GPU load, but it was only at 75-80%.
Here's a screenshot for ya. Main rig spec in sig. Btw, what mix of SP/DP did you go for? Or is that not chosen by the benchmark program?

Post by **PantherX** » Sat Apr 18, 2020 1:14 am

Assimilator1 wrote:...So what is the long list of stress tests you use then?
I was thinking of using OCCTs GPU stress test, but I wonder if it's artificially throttled as (IIRC) it's counted as a 'power virus'. As is furmark.
So I would like to know what is a realistic GPU stress test (as in not throttled, or drawing an unrealistic power load) with error reporting?

This list of mine was last updated in 2016 since that was the last time I assembled by system. I haven't done any research into current stress test software... that might change next year if I decide to get myself a AMD Zen 3 + Nvidia 3000 Ti GPUs

Catzilla Computer Benchmark: https://www.catzilla.com/download
Cinebench: https://www.maxon.net/en/support/downloads/
FurMark: https://geeks3d.com/furmark/downloads/
Heaven Benchmark: https://benchmark.unigine.com/heaven
IntelBurnTest: https://www.techpowerup.com/download/intelburntest/
Linpack Xtreme: https://www.techpowerup.com/download/linpack-xtreme/
MemTest86: https://www.memtest86.com/download.htm
Superposition Benchmark: https://benchmark.unigine.com/superposition
Tropics Benchmarks: https://benchmark.unigine.com/tropics
Valley Benchmark: https://benchmark.unigine.com/valley
stressapptest: https://github.com/stressapptest/stressapptest
Futuremark 3DMark 11: https://www.techpowerup.com/download/fu ... 3dmark-11/
MSI Kombustor: https://geeks3d.com/furmark/kombustor/downloads/
AIDA64 Extreme: https://www.aida64.com/downloads
Prime95: https://www.mersenne.org/download/
OCCT: https://www.ocbase.com/

I like stressing out CPU/GPU/RAM simultaneously as opposed to individual components. I also ensure that the ambient temperature is as high as possible... including using a heat as I like simulating the worst conditions my system would perform in. However, with what I have been reading about "automatic overclocking" I think in my next build, I will focus on maximum cooling and let the hardware do it's job as opposed to manually overclocking anything

Assimilator1 · Post by **Assimilator1** » Sat Apr 18, 2020 9:10 am

Thanks

, some I'm familiar with, I use OCCT for CPU testing & it's better than P95, so I don't use that anymore.
Heaven doesn't have error reporting for the free version, so I don't know how you would know when an error occurs? (out right crashes aside).
Furmark is detected by the drivers as a 'power virus' (as is one other I can't remember the name of), and so power is throttled for that test, so I'm not sure how good a test that makes it.
Are their any on your list which are GPU tests, do error testing+reporting & don't apply an 'unrealistic' load? Tall order I know!

I remember when ATITool would automatically lower GPU clock speeds on finding an error, magic! (didn't that become Furmark?).

Agreed about CPU overclocking, my Ryzen 3600 is the 1st CPU I haven't manually overclocked in over 20 yrs!

Post by **PantherX** » Sat Apr 18, 2020 9:26 am

Assimilator1 wrote:...Are their any on your list which are GPU tests, do error testing+reporting & don't apply an 'unrealistic' load? Tall order I know! ...

I never bothered to do reporting... I tend to sit in front my screen while all this runs and make notes on paper with a pencil... talk about old school

Assimilator1 · Post by **Assimilator1** » Sat Apr 18, 2020 9:34 am

Lol, you sit & watch the screen for hours??

Neil-B · Post by **Neil-B** » Sat Apr 18, 2020 10:08 am

Hours isn't really old school .. PantherX probably means days

… although he is using a pencil not a slate

davidcoton · Post by **davidcoton** » Sat Apr 18, 2020 10:36 am

I thought it was a lump of chalk on the cave wall

Post by **PantherX** » Sat Apr 18, 2020 11:21 am

Nah... I have upgraded to markings in the sand

Of course, if the wind blows, I have to start all over again

calkapokole · Post by **calkapokole** » Sat Apr 18, 2020 11:25 pm

foldy wrote:FAHbench 2.3.2 for Windows build with OpenMM 7.4.1 Core_22 with mixed precision
https://www.file-upload.net/download-13 ... n.zip.html

So that should match the current FAH GPU core for Corona Virus work units.

Just for fun I've benchmarked my GeForce GTX 660M (retired from active FAH duty a few years ago)

Compute: OpenCL
Precision: mixed
WU: real
Atoms: 64614
Accuracy Check: Enabled
NaN Check: Disabled
Run lenght: 5 mins
Score: 4.52456
Scaled Score: 13.6536

Folding Forum

FAHBench (OpenMM 5.1)

Re: FAHBench (OpenMM 5.1)

Re: FAHBench (OpenMM 5.1)

Re: FAHBench (OpenMM 5.1)

Re: FAHBench (OpenMM 5.1)

Re: FAHBench (OpenMM 5.1)

Re: FAHBench (OpenMM 5.1)

Re: FAHBench (OpenMM 5.1)

Re: FAHBench (OpenMM 5.1)

Re: FAHBench (OpenMM 5.1)

Re: FAHBench (OpenMM 5.1)

Re: FAHBench (OpenMM 5.1)

Re: FAHBench (OpenMM 5.1)

Re: FAHBench (OpenMM 5.1)