Using Nvidia Tesla Dual GPU card

Message boards : GPUs : Using Nvidia Tesla Dual GPU card
Message board moderation

To post messages, you must log in.

AuthorMessage
kimsejin5

Send message
Joined: 17 Jun 18
Posts: 2
United States
Message 86588 - Posted: 17 Jun 2018, 1:18:50 UTC

I am using BOINC 7.8.3 (x64 version), and am trying to use a Nvidia Tesla card for compute purposes in my computer to its fullest potential. Right now, I have three GPUs on two cards: one Quadro (purely for graphics) and one Tesla (purely for compute) with two GPUs on board. The computer recognizes GPU1 as Tesla's first GPU, GPU2 as Tesla's second GPU, and GPU3 is the Quadro GPU (unless I have that all screwed up, lmk). Right now, BOINC is only using one of the Tesla cores, GPU1, and is ignoring GPU2, even though both can be used. I've already tried adjusting the cc_config.xml file, and right now, it reads
<cc_config>
   <use_all_gpus>1<use_all_gpus>
   <ignore_nvidia_dev>3<ignore_nvidia_dev>
</cc_config>

I'm trying to use it for GPUGRID or Einstein@Home, either one would be splendid to get working.
The BOINC log is below:
6/16/2018 9:03:15 PM |  | Unrecognized tag in cc_config.xml: <use_all_gpus>
6/16/2018 9:03:15 PM |  | Missing end tag in cc_config.xml
6/16/2018 9:03:15 PM |  | Starting BOINC client version 7.8.3 for windows_x86_64
6/16/2018 9:03:15 PM |  | log flags: file_xfer, sched_ops, task
6/16/2018 9:03:15 PM |  | Libraries: libcurl/7.47.1 OpenSSL/1.0.2g zlib/1.2.8
6/16/2018 9:03:15 PM |  | Data directory: C:\ProgramData\BOINC
6/16/2018 9:03:15 PM |  | Running under account Sejin Kim
6/16/2018 9:03:15 PM |  | CUDA: NVIDIA GPU 0: Quadro K4000 (driver version 391.24, CUDA version 9.1, compute capability 3.0, 3072MB, 2555MB available, 1245 GFLOPS peak)
6/16/2018 9:03:15 PM |  | OpenCL: NVIDIA GPU 0: Quadro K4000 (driver version 391.24, device version OpenCL 1.2 CUDA, 3072MB, 2555MB available, 1245 GFLOPS peak)
6/16/2018 9:03:15 PM |  | Host name: DESKTOP-UP4T9S7
6/16/2018 9:03:15 PM |  | Processor: 12 GenuineIntel Intel(R) Core(TM) i7-6850K CPU @ 3.60GHz [Family 6 Model 79 Stepping 1]
6/16/2018 9:03:15 PM |  | Processor features: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss htt tm pni ssse3 fma cx16 sse4_1 sse4_2 movebe popcnt aes f16c rdrandsyscall nx lm avx avx2 vmx tm2 dca pbe fsgsbase bmi1 hle smep bmi2
6/16/2018 9:03:15 PM |  | OS: Microsoft Windows 10: Professional x64 Edition, (10.00.16299.00)
6/16/2018 9:03:15 PM |  | Memory: 31.84 GB physical, 36.59 GB virtual
6/16/2018 9:03:15 PM |  | Disk: 446.53 GB total, 318.23 GB free
6/16/2018 9:03:15 PM |  | Local time is UTC -4 hours
6/16/2018 9:03:15 PM |  | A new version of BOINC is available (7.10.2). <a href=https://boinc.berkeley.edu/download.php>Download</a>
6/16/2018 9:03:15 PM | Einstein@Home | URL http://einstein.phys.uwm.edu/; Computer ID 12635138; resource share 100
6/16/2018 9:03:15 PM | GPUGRID | URL http://www.gpugrid.net/; Computer ID 471497; resource share 100
6/16/2018 9:03:15 PM | NFS@Home | URL http://escatter11.fullerton.edu/nfs/; Computer ID 6287398; resource share 100
6/16/2018 9:03:15 PM | Einstein@Home | General prefs: from Einstein@Home (last modified 01-Dec-2017 17:08:08)
6/16/2018 9:03:15 PM | Einstein@Home | Host location: none
6/16/2018 9:03:15 PM | Einstein@Home | General prefs: using your defaults
6/16/2018 9:03:15 PM |  | Reading preferences override file
6/16/2018 9:03:15 PM |  | Preferences:
6/16/2018 9:03:15 PM |  | max memory usage when active: 16302.89 MB
6/16/2018 9:03:15 PM |  | max memory usage when idle: 29345.20 MB
6/16/2018 9:03:15 PM |  | max disk usage: 10.00 GB
6/16/2018 9:03:15 PM |  | suspend work if non-BOINC CPU load exceeds 12%
6/16/2018 9:03:15 PM |  | (to change preferences, visit a project web site or select Preferences in the Manager)
6/16/2018 9:08:21 PM | NFS@Home | Computation for task S10p338_935148_0 finished
6/16/2018 9:08:22 PM | NFS@Home | Starting task S10p338_935164_0
6/16/2018 9:08:24 PM | NFS@Home | Started upload of S10p338_935148_0_r988184853_0
6/16/2018 9:08:27 PM | NFS@Home | Finished upload of S10p338_935148_0_r988184853_0

I am aware that something's not right about the XML file, I just don't know what. If someone could give me a valid cc_config.xml code snippit to use, that would be great.
The log's not showing it, but I have an Intel Core i7-6850K, Nvidia Quadro K4000, and Nvidia Tesla K10. BOINC is using one of the Tesla GPUs properly, but is not utilizing the other one at all. I want BOINC to use both of the GPUs on the Tesla, and not do any computation on the Quadro card. I am using Quadro driver version 391.24, and have Nvidia Maximus enabled. I am using Windows 10 x64, version 1709, build 16299.492.
I know that all of this is kind of pandering, let me know if you need more info. Thanks in advance!
ID: 86588 · Report as offensive
kimsejin5

Send message
Joined: 17 Jun 18
Posts: 2
United States
Message 86589 - Posted: 17 Jun 2018, 1:20:25 UTC - in response to Message 86588.  

It's also worth noting that if the use_all_gpus XML flag were working right, it should ignore the Quadro, since I've set that card to be dedicated to graphics tasks, and the Tesla GPUs to be dedicated to compute needs.
ID: 86589 · Report as offensive
MarkJ
Volunteer tester
Help desk expert

Send message
Joined: 5 Mar 08
Posts: 272
Australia
Message 86592 - Posted: 17 Jun 2018, 10:11:20 UTC
Last modified: 17 Jun 2018, 10:13:47 UTC

Did you install the Tesla Driver for Windows?

According to the log it can only see the Quadro and isn't aware of the Tesla. If it found it you'd have 3 lines (one for each device) and then you can tell it which one to ignore.
MarkJ
ID: 86592 · Report as offensive
Profile Jord
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 29 Aug 05
Posts: 15492
Netherlands
Message 86594 - Posted: 17 Jun 2018, 11:12:15 UTC - in response to Message 86588.  

<cc_config>
   <use_all_gpus>1<use_all_gpus>
   <ignore_nvidia_dev>3<ignore_nvidia_dev>
</cc_config>

It's missing the end tags, as the log tells you. Tags start with
<something>
and end with
</something>
, with a slash at the start of the tag.

So in your case, that's
<cc_config>
   <use_all_gpus>1</use_all_gpus>
   <ignore_nvidia_dev>3</ignore_nvidia_dev>
</cc_config>

However, <code>use_all_gpus</code> is only used when you have two differing same brand GPUs you want to use, for instance a GTX1080 and your Quadro. When a dual GPU is detected, both GPUs on there ought to be exactly the same already, as Nvidia doesn't release dual GPU cards with two wildly differing GPUs.

I checked your system on GPUGrid and the one task you ran there does say it ran on the Tesla, which is possible. But that you may have had a forced driver update from Windows in the mean time. Windows drivers take away at least OpenCL capability. If you haven't done so yet, you may want to disable automatic driver updates in Windows 10.
Another possibility is that the GPUGrid application can detect the GPUs in your system and use them.
ID: 86594 · Report as offensive

Message boards : GPUs : Using Nvidia Tesla Dual GPU card

Copyright © 2024 University of California.
Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.