Tesla K80 on Google Cloud platform - not recognised

Message boards : Number crunching : Tesla K80 on Google Cloud platform - not recognised
Message board moderation

To post messages, you must log in.

AuthorMessage
Jason Xiros

Send message
Joined: 16 Mar 07
Posts: 4
Credit: 5,368,683
RAC: 151
Australia
Message 1882016 - Posted: 4 Aug 2017, 10:21:29 UTC

I have a Google Cloud Compute server with an attached Tesla K8 GPU. The GPU appears in Windows Device Manager, and the latest NVIDIA drivers are installed.

SETI@home does not use the GPU. It shows against the list of computers on my account, but no GPU tasks are assigned to it by BOINC manager.

What am I missing??
ID: 1882016 · Report as offensive
rob smith Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer moderator
Volunteer tester

Send message
Joined: 7 Mar 03
Posts: 22158
Credit: 416,307,556
RAC: 380
United Kingdom
Message 1882019 - Posted: 4 Aug 2017, 10:27:05 UTC

Starter for ten:
Have you selected all the "use nVidia GPU" options, both locally, and on your account web page?
Also, have you got a sensible value for the "use GPU when less than x%" busy in your local options?
Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?
ID: 1882019 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14649
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1882022 - Posted: 4 Aug 2017, 10:39:21 UTC

Are you 'running as a daemon' - i.e. headless, without a logged-in Windows user?
ID: 1882022 · Report as offensive
Jason Xiros

Send message
Joined: 16 Mar 07
Posts: 4
Credit: 5,368,683
RAC: 151
Australia
Message 1882023 - Posted: 4 Aug 2017, 10:51:17 UTC - in response to Message 1882019.  

Hi Bob,

I appreciate your help. Both web-based and local preferences are identical. GPU is never suspended.

I know the client can see the GPU... this one has me scratching my head :-(
ID: 1882023 · Report as offensive
Jason Xiros

Send message
Joined: 16 Mar 07
Posts: 4
Credit: 5,368,683
RAC: 151
Australia
Message 1882024 - Posted: 4 Aug 2017, 10:53:19 UTC - in response to Message 1882022.  

Are you 'running as a daemon' - i.e. headless, without a logged-in Windows user?


No... I didn't choose the "run as a service option"

I connect to the server via RDP, but the GPU isn't used after I log off either.
ID: 1882024 · Report as offensive
Jason Xiros

Send message
Joined: 16 Mar 07
Posts: 4
Credit: 5,368,683
RAC: 151
Australia
Message 1882025 - Posted: 4 Aug 2017, 11:09:58 UTC

For reference, these log entries are written when starting BOINC

04-Aug-2017 08:37:19 [---] Starting BOINC client version 7.6.33 for windows_x86_64
04-Aug-2017 08:37:19 [---] log flags: file_xfer, sched_ops, task
04-Aug-2017 08:37:19 [---] Libraries: libcurl/7.47.1 OpenSSL/1.0.2g zlib/1.2.8
04-Aug-2017 08:37:19 [---] Data directory: C:\ProgramData\BOINC
04-Aug-2017 08:37:19 [---] Running under account Administrator
04-Aug-2017 08:37:20 [---] CUDA: NVIDIA GPU 0: Tesla K80 (driver version 369.30, CUDA version 8.0, compute capability 3.7, 4096MB, 4010MB available, 4111 GFLOPS peak)
04-Aug-2017 08:37:20 [---] OpenCL: NVIDIA GPU 0: Tesla K80 (driver version 369.30, device version OpenCL 1.2 CUDA, 11424MB, 4010MB available, 4111 GFLOPS peak)
04-Aug-2017 08:37:20 [---] Host name: thebrain
04-Aug-2017 08:37:20 [---] Processor: 8 GenuineIntel Intel(R) Xeon(R) CPU @ 2.50GHz [Family 6 Model 62 Stepping 4]
04-Aug-2017 08:37:20 [---] Processor features: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ss htt pni ssse3 cx16 sse4_1 sse4_2 popcnt aes f16c rdrandsyscall nx lm avx fsgsbase smep
04-Aug-2017 08:37:20 [---] OS: Microsoft Windows Server 2012 R2: Datacenter x64 Edition, (06.03.9600.00)
04-Aug-2017 08:37:20 [---] Memory: 32.00 GB physical, 33.00 GB virtual
04-Aug-2017 08:37:20 [---] Disk: 320.00 GB total, 299.49 GB free
04-Aug-2017 08:37:20 [---] Local time is UTC +0 hours

so it can definately see the GPU :-(
ID: 1882025 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14649
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1882035 - Posted: 4 Aug 2017, 12:43:04 UTC - in response to Message 1882024.  
Last modified: 4 Aug 2017, 12:43:53 UTC

I connect to the server via RDP, but the GPU isn't used after I log off either.
Both running RDP, and logging off from Windows, will kill GPU computation: the necessary tools (CUDA or OpenCL) can only run in the context of a local user - they are blocked from the local system account for security reasons, and have been ever since Windows Vista.

Can you install a different remote access service, like VNC? That doesn't have the same driver restrictions as Microsoft's RDP. And then, can you disconnect from the machine without logging out of Windows?
ID: 1882035 · Report as offensive
Profile -= Vyper =-
Volunteer tester
Avatar

Send message
Joined: 5 Sep 99
Posts: 1652
Credit: 1,065,191,981
RAC: 2,537
Sweden
Message 1882113 - Posted: 4 Aug 2017, 20:18:46 UTC

Try teamviewer.. It worx flawless..

_________________________________________________________________________
Addicted to SETI crunching!
Founder of GPU Users Group
ID: 1882113 · Report as offensive

Message boards : Number crunching : Tesla K80 on Google Cloud platform - not recognised


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.