Why do I get no GPU work?

alintope
alintope
Joined: 27 Jan 12
Posts: 52
Credit: 232,774,818
RAC: 43,239
Topic 209896

Today I reactivated one of my computers and inserted a Geforce 1050 TI graphics card to speed it up for GPU work. I wonder why Einstein@home ignores this rather new card and doesn't send any GPU work. Now this computer is doing only "O1Spot1Hi_GalCent" work units, which do not need the GPU. What can I do to make this machine download GPU work units? What is it that holds it from fetching those work units? In the project's preferencies I switched on all available kinds of applications including "Gamma-ray pulsar binary search #1 (GPU)" and "Binary Radio Pulsar Search (Arecibo, GPU)".

Thank you for any helpful idea.

Heinrich

 

 

Zalster
Zalster
Joined: 26 Nov 13
Posts: 2,976
Credit: 3,226,958,391
RAC: 746,652

What are your preferences on

What are your preferences on your website? Do you have the location where that computer is set to run your GPU?  It sounds like it's set to default and has it only running CPU work units

archae86
archae86
Joined: 6 Dec 05
Posts: 2,673
Credit: 2,344,851,773
RAC: 2,994,401

There has been a history in

There has been a history in the last year of some Windows 10 updates setting the Nvidia driver or perhaps some condition flag to a state in which Einstein did not see software support it needed.  People had success with one combination or another of installing a manually downloaded driver direct from Nvidia, using DDU to clean up first, and some thinking they needed to select the clean install option.  However that problem generally comes with readily observed signs of pain, if you just look around the event log for your machine.  If you have already looked there, and don't see obvious trouble, most likely your problem is something else.

The first thing to do when you don't like work fetch behavior is to review the web page at which Einstein posts the most recent work request cycle summary for the machine in question.  The URL is simply formed using your host ID.

For your 1050 machine the Einstein ID is 12136875 so this site is found at:

https://einsteinathome.org/host/12136875/log

While reading these logs is a bit of an art, it is clear that the most recent interaction when I looked just now, your machine did not request either CPU nor GPU work at all.

That sounds as though BOINC running on your machine thinks it has enough work to do.  Do you run another project?  

All five tasks that host has returned recently have errored out extremely quickly, so there is something more wrong at your host than just work fetch problems.

 

 

 

 

alintope
alintope
Joined: 27 Jan 12
Posts: 52
Credit: 232,774,818
RAC: 43,239

Meanwhile this machine is

Meanwhile this machine is running fine and doesn't return false results any more. May be the reason was, I fiddled so much around with some old and overdue tasks after I had installed the BOINC client. This might have caused the extremely quickly error returns.

The last server protocoll gives me the following information. Especially line #5 is interesting: "0.00 Cuda instances". What does that mean? Does it point at the problem? I asked GPU-Z which told me the graphics driver installed does support CUDA. Its mode is 6.1. (PhysX is not installed - as I didn't need it for E@H so far.)

2017-09-26 04:56:09.6828 [PID=5118 ]    [send] effective_ncpus 4 max_jobs_on_host_cpu 999999 max_jobs_on_host 999999
2017-09-26 04:56:09.6828 [PID=5118 ]    [send] effective_ngpus 1 max_jobs_on_host_gpu 999999
2017-09-26 04:56:09.6828 [PID=5118 ]    [send] Not using matchmaker scheduling; Not using EDF sim
2017-09-26 04:56:09.6828 [PID=5118 ]    [send] CPU: req 172800.00 sec, 1.00 instances; est delay 0.00
2017-09-26 04:56:09.6829 [PID=5118 ]    [send] CUDA: req 0.00 sec, 0.00 instances; est delay 0.00
2017-09-26 04:56:09.6829 [PID=5118 ]    [send] work_req_seconds: 172800.00 secs
2017-09-26 04:56:09.6829 [PID=5118 ]    [send] available disk 9.47 GB, work_buf_min 86400
2017-09-26 04:56:09.6829 [PID=5118 ]    [send] active_frac 0.120578 on_frac 0.962648 DCF 3.453397
2017-09-26 04:56:09.6837 [PID=5118 ]    [mixed] sending locality work first (0.0807)
2017-09-26 04:56:09.6854 [PID=5118 ]    [send] send_old_work() no feasible result older than 336.0 hours
2017-09-26 04:56:09.7385 [PID=5118 ]    [version] Checking plan class 'AVX'
2017-09-26 04:56:09.7422 [PID=5118 ]    [version] reading plan classes from file '/BOINC/projects/EinsteinAtHome/plan_class_spec.xml'
2017-09-26 04:56:09.7423 [PID=5118 ]    [version] plan class ok
2017-09-26 04:56:09.7423 [PID=5118 ]    [version] Best version of app einstein_O1Spot1Hi is 1.00 ID 985 AVX (7.45 GFLOPS)

Does this help finding out why the GPU is ignored?

Heinrich

 

alintope
alintope
Joined: 27 Jan 12
Posts: 52
Credit: 232,774,818
RAC: 43,239

I overlooked some trivial

I overlooked some trivial aspect: I always thought my machine would be at location "home" where I consciously switched on GPU work. But in fact it was at location "generic" where GPU work is switched off. Sorry for asking for your help and wasting your time!

Heinrich

 

 

 

archae86
archae86
Joined: 6 Dec 05
Posts: 2,673
Credit: 2,344,851,773
RAC: 2,994,401

So Zalster asked you the

So Zalster asked you the right question back there.  Glad you found it.

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.