Cuda capable GTX 1070 wont' crunch CUDA apps, only opencl

Pierre Poulin
Pierre Poulin
Joined: 12 Dec 05
Posts: 28
Credit: 2875094701
RAC: 351263
Topic 224665

Hi,

Help with the following would be greatly appreciated.

I just set up an Nvidia 1070 running perfectly with opencl on MacOS 10.3.6 and Boinc 7.16.14.

Cuda drivers are installed, recognized in the MacOS Control Panel, no update required.

Einstein@home appears to detect the Cuda capability of the GPU but for some unknown reason, chooses to only crunch OpenCL apps .

Here’s  the account_einstein.phys.uwm.edu.xml file content followed by the Boinc Log:

Also, I’m running the GPU from « home » configuration with 0.25 work unit factor.

 

<account>

    <master_url>http://einstein.phys.uwm.edu/</master_url>

    <authenticator>0cf7d12d1164ed1648c74afeeadbcXXX</authenticator>

    <project_name>Einstein@Home</project_name>

<project_preferences>

 

  <venue name="home">

    <project_specific>

      <graphics fps="20" width="800" height="600" quality="low"/>

      <app_id>19</app_id>

      <app_id>32</app_id>

      <app_id>46</app_id>

      <app_id>49</app_id>

      <app_id>52</app_id>

      <app_id>53</app_id>

      <libc215>0</libc215>

      <also_run_cpu>0</also_run_cpu>

      <allow_non_preferred_apps>0</allow_non_preferred_apps>

      <gpu_util_brp>1</gpu_util_brp>

      <gpu_util_fgrp>1</gpu_util_fgrp>

      <gpu_util_gw>1</gpu_util_gw>

    </project_specific>

    <resource_share>100</resource_share>

    <no_cpu>0</no_cpu>

    <no_cuda>1</no_cuda>

    <no_ati>1</no_ati>

    <no_intel_gpu>1</no_intel_gpu>

    <allow_beta_work>0</allow_beta_work>

  </venue>

 

And the Boinc Log : (sam is samedi, Saturday in French)

 

Sam 30 jan 15:56:43 2021 |  | cc_config.xml not found - using defaults

Sam 30 jan 15:56:43 2021 |  | Starting BOINC client version 7.16.14 for x86_64-apple-darwin

Sam 30 jan 15:56:43 2021 |  | log flags: file_xfer, sched_ops, task

Sam 30 jan 15:56:43 2021 |  | Libraries: libcurl/7.58.0 OpenSSL/1.1.0g zlib/1.2.11 c-ares/1.13.0

Sam 30 jan 15:56:43 2021 |  | Data directory: /Library/Application Support/BOINC Data

Sam 30 jan 15:56:43 2021 |  | CUDA: NVIDIA GPU 0: GeForce GTX 1070 (driver version 10.1.163, CUDA version 10.1, compute capability 6.1, 4096MB, 3972MB available, 6852 GFLOPS peak)

Sam 30 jan 15:56:43 2021 |  | CUDA: NVIDIA GPU 1 (not used): GeForce GT 640M (driver version 10.1.163, CUDA version 10.1, compute capability 3.0, 512MB, 84MB available, 572 GFLOPS peak)

Sam 30 jan 15:56:43 2021 |  | OpenCL: NVIDIA GPU 0: GeForce GTX 1070 (driver version 10.33.0 387.10.10.10.40.140, device version OpenCL 1.2, 8192MB, 3972MB available, 6852 GFLOPS peak)

Sam 30 jan 15:56:43 2021 |  | OpenCL CPU: Intel(R) Core(TM) i5-3330S CPU @ 2.70GHz (OpenCL driver vendor: Apple, driver version 1.1, device version OpenCL 1.2)

Sam 30 jan 15:56:43 2021 |  | Host name: iMac-2012-3.local

Sam 30 jan 15:56:43 2021 |  | Processor: 4 GenuineIntel Intel(R) Core(TM) i5-3330S CPU @ 2.70GHz [x86 Family 6 Model 58 Stepping 9]

Sam 30 jan 15:56:43 2021 |  | Processor features: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clfsh ds acpi mmx fxsr sse sse2 ss htt tm pbe pni pclmulqdq dtes64 mon dscpl vmx est tm2 ssse3 cx16 tpr pdcm sse4_1 sse4_2 x2apic popcnt aes pcid xsave osxsave tsctmr avx rdrand f16c

Sam 30 jan 15:56:43 2021 |  | OS: Mac OS X 10.13.6 (Darwin 17.7.0)

Sam 30 jan 15:56:43 2021 |  | Memory: 8.00 GB physical, 911.16 GB virtual

Sam 30 jan 15:56:43 2021 |  | Disk: 930.71 GB total, 910.92 GB free

Sam 30 jan 15:56:43 2021 |  | Local time is UTC -5 hours

Sam 30 jan 15:56:43 2021 | Einstein@Home | General prefs: from Einstein@Home (last modified 30-Jan-2021 01:29:39)

Sam 30 jan 15:56:43 2021 | Einstein@Home | Host location: none

Sam 30 jan 15:56:43 2021 | Einstein@Home | General prefs: using your defaults

Sam 30 jan 15:56:43 2021 |  | Preferences:

Sam 30 jan 15:56:43 2021 |  | max memory usage when active: 4096.00 MB

Sam 30 jan 15:56:43 2021 |  | max memory usage when idle: 5734.40 MB

Sam 30 jan 15:56:43 2021 |  | max disk usage: 465.36 GB

Sam 30 jan 15:56:43 2021 |  | suspend work if non-BOINC CPU load exceeds 75%

Sam 30 jan 15:56:43 2021 |  | (to change preferences, visit a project web site or select Preferences in the Manager)

Sam 30 jan 15:56:43 2021 |  | Setting up project and slot directories

Sam 30 jan 15:56:43 2021 |  | Checking active tasks

Sam 30 jan 15:56:43 2021 | Einstein@Home | URL http://einstein.phys.uwm.edu/; Computer ID 12869257; resource share 100

Sam 30 jan 15:56:43 2021 | Einstein@Home | Not using CPU: project preferences

Sam 30 jan 15:56:43 2021 |  | Setting up GUI RPC socket

Sam 30 jan 15:56:43 2021 |  | Checking presence of 331 project files

Sam 30 jan 15:56:43 2021 | Einstein@Home | Started upload of h1_0543.20_O2C02Cl4In0__O2MDFS3_Spotlight_543.80Hz_475_0_1

 

I’m not a coder at all but I read from this that CUDA should work.

Anyway I can change a « ) » or « 1 » somewhere  to make it crunch ONLY CUDA ?

Thanks for any help

Pierre Poulin

Harri Liljeroos
Harri Liljeroos
Joined: 10 Dec 05
Posts: 4341
Credit: 3204648514
RAC: 1972486

Einstein @ Home does not have

Einstein @ Home does not have Cuda applications available for GPUs, only opencl.

[edit] See here: https://einsteinathome.org/apps.php

Pierre Poulin
Pierre Poulin
Joined: 12 Dec 05
Posts: 28
Credit: 2875094701
RAC: 351263

I guess I was too used to

I guess I was too used to crunching CUDA Seti@home apps and took for granted it was widely used...https://boinc.berkeley.edu/wiki/GPU_computing

Thank you !

Pierre Poulin
Pierre Poulin
Joined: 12 Dec 05
Posts: 28
Credit: 2875094701
RAC: 351263

P.S. Wiki says Aricebo radio

P.S. Wiki says Aricebo radio was CUDA compatible...Too bad Arecibo is gone forever, what a tremendous loss...

See "CUDA" on    https://en.wikipedia.org/wiki/Einstein@Home#Scientific_objectives

James Bradshaw
James Bradshaw
Joined: 1 Mar 14
Posts: 23
Credit: 1278981605
RAC: 2343409

I am trying to put together

I am trying to put together various posts to find out if I am understanding the GPU issues correctly.  I am running two systems with GTX1070.  When running primarily SETI the 1070s were sucking up all the power on the grid.  When SETI started to go away I returned from partial Einstein to ONLY Einstein, and have noticed that my crunching numbers in Einstein have dropped dramatically.  Example, on one machine, in May 2020,I averaged 175,726 Einstein credits per day.  In December, on the same machine I averaged 74,780 credits per day.  I did not knowingly change any setting.  Ran diagnostics on the 1070s.  WNL, temp OK, all other parameters OK, except that the 1070 were running at about 3 to 5% utilization.  NOTE:  The built in Intel HUD Graphics 630 is running near max output.

If I understand all the comments, I can forget about running either 1070 for Einstein, and I need to find a new home for the GPUs.  I am not a gamer and the GPUs are only installed for distributed computing.

Disclaimer:  I am not a computer jock and do not write code.

Your comments, suggestions, etc., most appreciate.

 

Thanks.  JB

Keith Myers
Keith Myers
Joined: 11 Feb 11
Posts: 4964
Credit: 18713718606
RAC: 6378293

There is probably nothing

There is probably nothing wrong with your  1070.  It should be running around 90% utilization.

There have been reports of issues with the latest Windows Nvidia drivers and distributed computing.

Reported solutions is to backlevel several versions to get back to a working driver.

 

mikey
mikey
Joined: 22 Jan 05
Posts: 12680
Credit: 1839084099
RAC: 3910

Keith Myers wrote:There is

Keith Myers wrote:

There is probably nothing wrong with your  1070.  It should be running around 90% utilization.

There have been reports of issues with the latest Windows Nvidia drivers and distributed computing.

Reported solutions is to backlevel several versions to get back to a working driver.

 

He's using version 451.67 and that's the same one I'm using and it works just fine, I'm guessing it's more the bad tasks and problems with the uploads sometimes instead. Part of it could also be he stopped crunching the Gamma Ray Pulsar Binary Search #1 for gpu tasks and moved on to the GW tasks and they had some major problems recently. They also have a much longer wait time to get validated by a wingman than the GRPB tasks do and you get fewer credits on top of it.

James Bradshaw
James Bradshaw
Joined: 1 Mar 14
Posts: 23
Credit: 1278981605
RAC: 2343409

Thanks Keith and Mikey for

Thanks Keith and Mikey for your comments.  Regarding stopping crunching Gamma Ray Pulsar Binary Search #1 and going to GW tasks.....If so, it was not a change I knowingly made...I have been crunching whatever "they" send me.  It sounds as though some people have the ability to select only the tasks that provide the most credits the quickest.  This morning the 1070 is blazing its way at 13%, Intel 650 GPU at 30-40%, and CPUs at 100%, all with diagnostics showing WNL.  I would rather have the crunching done with GPUs than CPUs, but....    In any case, I appreciate the time you guys took to respond.  

Keith Myers
Keith Myers
Joined: 11 Feb 11
Posts: 4964
Credit: 18713718606
RAC: 6378293

You too can select which

You too can select which tasks you would like to crunch through your project preferences page.

Project Preferences

By selecting which apps and which hardware you would like to run, you take control over how you crunch for each project.

I also suspect you are getting confused with what Task Manager shows for your 1070 utilization.  I haven't run Windows in years but I have seen many posts about this confusing issue with Task Manager.

Supposedly you have to choose the correct page or screen in Task Manager to show Compute utilization other than the default Graphics utilization.  Many people are confused with this apparently and complaining their gpus are not being used properly.

I think you will find that your 1070 is actually running close to 90% utilization IF you choose the correct Task Manager screen.

[Edit] Another way to prove a sanity check for utilization is use the Nvidia provided utility called nvidia-smi in the Terminal.   That shows the temps, fan speed and utilization of all detected Nvidia cards in your system.

That utility is located at C:\Program Files\NVIDIA Corporation\NVSMI

 

mikey
mikey
Joined: 22 Jan 05
Posts: 12680
Credit: 1839084099
RAC: 3910

Keith Myers wrote:I think

Keith Myers wrote:

I think you will find that your 1070 is actually running close to 90% utilization IF you choose the correct Task Manager screen.

[Edit] Another way to prove a sanity check for utilization is use the Nvidia provided utility called nvidia-smi in the Terminal.   That shows the temps, fan speed and utilization of all detected Nvidia cards in your system.

That utility is located at C:\Program Files\NVIDIA Corporation\NVSMI 

I could not find the file in my Win10 folders and I don't see the option to see "Compute utilization" either

My 1660 TI is using 89 to 94% running a single task at a time on the Gamma Ray Pulsar #1 tasks. I DO keep a cpu core free to feed the gpu so that could be a problem as well.

BUT I track my gpu usage, both AMD and Nvidia by using MSIAfterburner, it's free and you can even adjust the fans, power usage etc depending on the gpu you have.

https://www.guru3d.com/files-details/msi-afterburner-beta-download.html

I don't do the Beta versions but if you go to the bottom of the page you will see the download link for the both the latest Beta and the Stable/Final previous version that I use

Keith Myers
Keith Myers
Joined: 11 Feb 11
Posts: 4964
Credit: 18713718606
RAC: 6378293

You only get this exe if you

You only get this exe if you install the Nvidia drivers from Nvidia.  It does not come with the Nvidia drivers provided by Microsoft.

Try again and search for it with File Manager. The exe is named nvidia-smi.exe.

Just try navigating to the C:\Program Files\NVIDIA Corporation\NVSMI directory and open a Command Line Terminal or otherwise known as a DOS box.

Type nvidia-smi -l 10 and the program will open and poll your gpus every ten seconds.

Again, I have not used Task Manager in a long while and I am not aware of how it is designed anymore.

Juha tells how to find it in this message. Richie tells where to find it in the pull-down menus.

Task Manager Compute_0

Task Manager pull-down menus

 

Comment viewing options

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