Anandtech vs Tom's Hardware Folding@Home Coronavirus Race thread

[Markfw]

Well something happened. All of a sudden a bunch of points showed up ! 16.7 million last 24 hours !

 

[Markfw]

I have 5 unsent, waiting that have failed, but I am still at 18.3 million ppd.

 

[Endgame124]

You know, I never stress tested my primary folding box... I just aborted My KVM project, setup f@h, and have let it run ever since. I should really probably find some temperature software for Linux to make sure I’m not throttling and running at 95c... anyone have any suggestions?

 

[Howdy]

PSensor is a start for temperature monitoring, accuracy unknown.

Then there is Coolbits for the GPUs depending on your linux version allowing fan control. There is a Folding Install for Linux thread started by @TennesseeTony around here somewhere, I just can't find it.

 

[StefanR5R]

There is a Folding Install for Linux thread started by @TennesseeTony around here somewhere,

It's this one: GUIDE: Linux Mint 18.3 Cinnamon with Nvidia GPU(s), November 2017

To monitor NVIDIA GPU sensors:
nvidia-smi dmon -d 60
(The last parameter is the logging interval in seconds.)

 

[Endgame124]

My cpu slot size defaulted to 15 since I also have a gpu in the box. Is there a better size I should change it to?

 

[TennesseeTony]

In the past, odd numbers of threads were harder to get tasks for. I don't know if that is still the case today.

 

[Endgame124]

In the past, odd numbers of threads were harder to get tasks for. I don't know if that is still the case today.

I’ve been reading that. My cpu slot has been hit or miss for work until recently, but I don’t know if that is due to the work shortage or the odd slot configuration

 

[TennesseeTony]

Regarding points, 'only' 15 threads will seem minuscule compared to a GPU, so maybe try 2 CPU slots of 6 threads each, and add a third with 3 threads?

 

[Endgame124]

Regarding points, 'only' 15 threads will seem minuscule compared to a GPU, so maybe try 2 CPU slots of 6 threads each, and add a third with 3 threads?

I was considering switching to 10 cpu and putting 5 cpu into boinc

 

[Endgame124]

I was considering switching to 10 cpu and putting 5 cpu into boinc

Moved 5 CPUs to boinc. At first I was going to let all 5 run with climate prediction, but it only seemed to use 1 task and 1 cpu, so I also added Rosetta and 4 CPUs

 

[StefanR5R]

Regarding points, 'only' 15 threads will seem minuscule compared to a GPU, so maybe try 2 CPU slots of 6 threads each, and add a third with 3 threads?

Dividing a computer into two or more CPU slots will give fewer PPD than one CPU slot with a maximum of threads. (Exception: Crippled variants of the Windows OS which don't support slots larger than 32 threads. Arguably such operating systems should not be used on a computer with more than 32 hardware threads in the first place.)

15 threads in a slot are a multiple of 3 and 5, which is not ideal for domain decomposition but still OK.

I doubt that the number of threads of a slot influences its chances to receive work. (Exception: There may be no work available for slots with more than 64 threads occasionally for some weeks or months. But during the F@h race, work for 64-, 88-, and 128-thread slots was available. And when I tested today, work was still available for 128 thread slots.)

In general, folding on CPUs is a waste of energy. My 7 nm CPUs (with 14 nm I/O die) with 128 threads in one slot gave PPD/Watt comparable with 1080Ti, which is a 16 nm GPU. But smaller CPU slots give dramatically less PPD/Watt.

And since the F@h folks say that PPD reflect scientific value of the work, mainstream CPUs give very, very poor scientific value per power consumed. I say, use the CPUs for a different project than F@h if you find one which you like, or just leave the CPUs idle except for the GPU driver thread. (Exception: When you are in a F@h contest and want to wring out the last bit of performance for F@h alone.)

(Whenever I refer to PPD above, I consider quick return bonus included.)

PS, slots whose thread count is a multiple of larger factors, such as 11, or slots with very large thread count, may receive work which then fails to run due to inability of Fahcore_a7 to decompose the computational domain in a manner which would match the thread count.

 

[Endgame124]

Dividing a computer into two or more CPU slots will give fewer PPD than one CPU slot with a maximum of threads. (Exception: Crippled variants of the Windows OS which don't support slots larger than 32 threads. Arguably such operating systems should not be used on a computer with more than 32 hardware threads in the first place.)

15 threads in a slot are a multiple of 3 and 5, which is not ideal for domain decomposition but still OK.

I doubt that the number of threads of a slot influences its chances to receive work. (Exception: There may be no work available for slots with more than 64 threads occasionally for some weeks or months. But during the F@h race, work for 64-, 88-, and 128-thread slots was available. And today work was available for 128 thread slots too.)

In general, folding on CPUs is a waste of energy. My 7 nm CPUs (with 14 nm I/O die) with 128 threads in one slot gave PPD/Watt comparable with 1080Ti, which is a 16 nm GPU. But smaller CPU slots give dramatically less PPD/Watt.

And since the F@h folks say that PPD reflect scientific value of the work, mainstream CPUs give very, very poor scientific value per power consumed. I say, use the CPUs for a different project than F@h if you find one which you like, or just leave the CPUs idle except for the GPU driver thread. (Exception: When you are in a F@h contest and want to wring out the last bit of performance for F@h alone.)

(Whenever I refer to PPD above, I consider quick return bonus included.)

PS, slots whose thread count is a multiple of larger factors, such as 11, or slots with very large thread count, may receive work which then fails to run due to inability of Fahcore_a7 to decompose the computational domain in a manner which would match the thread count.

In general I would agree to save the CPUs for other projects, but F@H has been putting out quite a few CPU only projects recently.

Also, I figure those 10 CPUs give me a slight edge vs team members who are running a gpu setup similar to mine

 

[StefanR5R]

In general I would agree to save the CPUs for other projects, but F@H has been putting out quite a few CPU only projects recently.

People continue to donate many CPU-hours to F@h. Therefore they continue to create CPU work accordingly, while giving but very poor points accordingly.

 

[mopardude87]

Perhaps give GPUGRID a go on BOINC. It's medical research too. I mean, not forever, just until interest wanes in F@H (although I hope it doesn't wane too much).

Yeah i got a buddy who does BOINC but i can't get him to do F@H. Maybe he ran into issues with F@H and just decided BOINC is more reliable?

 

[Assimilator1]

Well, I guess I spoke too soon about giving all CPU threads to BOINC, running F@H WU https://stats.foldingathome.org/project?p=14545 and I noticed it's av GPU load was only ~65-70%, freeing up a core for F@H GPU raised that to ~85%. You can easily see it in the GPU-Z screenshot here too, mid way through that graph I set BOINC from 100% to 85% CPU. Back to 85% it is! Although I wonder how many WUs are this CPU dependant?

 

[mopardude87]

Yeah i keep finding that using the cpu keeps impacting gpu based folding . I don't run it anymore . Maybe that can change with the 3700x/3900x upgrade? Doubt it but we going from a stock 7700k here.

 

[StefanR5R]

I wonder how many WUs are this CPU dependant?

Maybe it's different between FahCore_21 and FahCore_22? Most GPU work is with core 22 now, but some still with 21. FAHControl as well as taskmanager should show which one is running at the moment.

 

[Assimilator1]

Too late, it's gone, but as per the link it's a Covid WU, I thought all Covid WUs were done with core22?

mopardude87
Is that with AMD or Nvidia GPUs? Or both?

 

[Endgame124]

If it would be interesting, I can snag a screen shot of my A10-7870K with 4 Boinc processes running (currently Seti), and folding on both the r250x and Integrated r7. The GPUs are at 100% while the CPUs are also at 100% from Boinc. Of course, combined, both GPUs only manage a little over 100k PPD, so I'm not sure it means much

On a different topic, I've got 3 stuck "sending" uploads on one host, and 1 on another. Guessing this is going to be the new problem, as getting work has been pretty consistent recently.

 

[mopardude87]

mopardude87
Is that with AMD or Nvidia GPUs? Or both?

I only run nvidia and i got a 1080ti. I run my Intel HD630 as primary driver then got the client running the 1080ti. HD630 lets me play SOME games and its not looking to be effecting ppd barely at all on the 1080ti as the cpu barely is tasked at like 19% currently with folding.

So many must be having wu issues cause somehow i landed in the top 10 on the team when the week before i was closer to 25. I feel bad cause i know others clearly have more then me running but somehow i got more points for the week. I feel bad

When i get the 3700x/3900x upgrade , i will prob consider grabbing a discount 1050ti as primary/secondary folder late nights. I could play older games on that baby while the 1080ti just chugs about. I could prob power a 3700x/1050ti/1080ti combo on my Seasonic Focus 650watt without issue especially if i stay stock.Kinda partial to a x570/3900x/nvme/dual gpu setup i think but the 3900x/1080ti/1050ti power consumption may push the 650w? Why i may go 3700x over the 3900x merely for power consumption.

 

[Markfw]

I now have 10 units that can't seem to be sent. The ppd is all but gone. Any way to clear these ? From 316k to 28k. So all ten units are now worth less than one.

 

[Endgame124]

I now have 10 units that can't seem to be sent. The ppd is all but gone. Any way to clear these ? From 316k to 28k. So all ten units are now worth less than one.

I had 3 stuck for 3 days, but they eventually cleared themselves. I have since collected 2 more that won’t upload.

 

[Assimilator1]

So many must be having wu issues cause somehow i landed in the top 10 on the team when the week before i was closer to 25. I feel bad cause i know others clearly have more then me running but somehow i got more points for the week. I feel bad

No don't worry about it mate, it's not down to you who gets the WUs, just luck of the draw, but if you're having a good run of luck, feel free to spread it around .

Ref GPUs, ah yea, as per earlier posts apparently Nvidia GPUs need more CPU time, for F@H at least. I thought I could get away with BOINC using all my CPU & run GPU F@H without losing ppd, I did for a while but it seems that doesn't work with all F@H WUs . So I've left a core free again now.

 

[TennesseeTony]

I now have 10 units that can't seem to be sent. .... Any way to clear these ?...

Find the folder in which the work unit is stored, and delete it.

For Windows it is

C:\Users\your user name\AppData\Roaming\FAHClient\work

.

edit: For Mint 19.3 Linux it is

File System / var / lib / fahclient / work

You will see a few files, and some folders such as 00, 01, 02 etc. These folders are the work units. Look at your Folding client/interface to see what ID numbers are active, and delete the inactive folder(s). In the example below, I would delete the folder named 00