SETI applications for NVIDIA GPU improvement - how you can help

Message boards : Number crunching : SETI applications for NVIDIA GPU improvement - how you can help
Message board moderation

To post messages, you must log in.

Previous · 1 . . . 10 · 11 · 12 · 13

AuthorMessage
Profile tullioProject Donor
Volunteer moderator
Volunteer tester

Send message
Joined: 9 Apr 04
Posts: 6296
Credit: 1,680,372
RAC: 1,589
Italy
Message 1835377 - Posted: 11 Dec 2016, 15:44:18 UTC

I have installed the GTX 750 on my Linux box and it works. But it still uses 8.10. CPU usage is very high, near 100% on one of the 2 cores of the Opteron 1210, vintage 2008. The other is shared by Einstein, LHC@home and Atlas@home.
Tullio
ID: 1835377 · Report as offensive     Reply Quote
Profile tullioProject Donor
Volunteer moderator
Volunteer tester

Send message
Joined: 9 Apr 04
Posts: 6296
Credit: 1,680,372
RAC: 1,589
Italy
Message 1835575 - Posted: 12 Dec 2016, 14:58:47 UTC
Last modified: 12 Dec 2016, 15:04:02 UTC

A 8.01 version with CUDA 60 instead of opencl_nvidia_sah lowers the CPU usage to 7%.
Tullio
ID: 1835575 · Report as offensive     Reply Quote
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 5806
Credit: 76,039,961
RAC: 50,982
Russia
Message 1835576 - Posted: 12 Dec 2016, 15:07:15 UTC - in response to Message 1835575.  

Per ReadMe use -use_sleep option if low CPU usage preferable for you. Performance degradation expected.
SETI apps news
We're not gonna fight them. We're gonna transcend them.
ID: 1835576 · Report as offensive     Reply Quote
Profile tullioProject Donor
Volunteer moderator
Volunteer tester

Send message
Joined: 9 Apr 04
Posts: 6296
Credit: 1,680,372
RAC: 1,589
Italy
Message 1835580 - Posted: 12 Dec 2016, 15:54:15 UTC
Last modified: 12 Dec 2016, 15:55:06 UTC

Now I have an Atlas@home task running on two cores using VirtualBox and SETI@home using only 7% of CPU. Performance is not essential, I am running CERN projects using VirtualBox, Einstein@home and SETI@home using CPU. The Windows 10 PC, with GTX 1050 is using 8.19 with opencl_nvidia_SoG. That is my main SETI PC. I was wondering about the difference between OpenCL and CUDA on Linux.
Tullio
ID: 1835580 · Report as offensive     Reply Quote
Profile Greg Eckert
Volunteer tester

Send message
Joined: 30 Jul 99
Posts: 1
Credit: 22,570,616
RAC: 9,833
United States
Message 1836354 - Posted: 16 Dec 2016, 20:32:12 UTC - in response to Message 1795063.  

Hey, I have a large number of computers running the app, plus I worked on the original optimizations back in the Pentium !!! days. I'd like to help, though I'm out of touch with how far this has come in the past ten years so I may need some help configuring them the way you want them. I have the following nVidia cards in the PCs I am running:

    - NVIDIA GeForce GTS 250 (1024MB) driver: 342.00 - This one is getting shockingly good results on the .44 Lunatics defaults, Core 2 Quad with this card getting 2,900 avg credits!
    - NVIDIA GeForce 210 (512MB) driver: 341.95 and Intel(R) Pentium(R) Dual CPU E2200 @ 2.20GHz [Family 6 Model 15 Stepping 13] (6 PCs with this card and processor) - PCs with this card are averaging only 40-80 avg credits
    - NVIDIA GeForce 210 (512MB) driver: 341.95 and Intel(R) Xeon(R) CPU X5550 @ 2.67GHz [Family 6 Model 26 Stepping 5] (2 PCs with this card and processor) avg 1900 credits
    - NVIDIA GeForce GTX 550 Ti (1024MB) driver: 376.19 OpenCL: 1.1 with Intel(R) Core(TM) i7 CPU 960 @ 3.20GHz [Family 6 Model 26 Stepping 5], avg 1900 credits
    - And a PC running dual cards with the NVIDIA GeForce GT 640 (2048MB) driver: 376.19 OpenCL: 1.2 running as the PhysX engine and the nVidia GeForce 470 GTX, OpenCL:1.1 running as the primary card for video.



In addition to these I have a bunch of computers doing the CPU/iGPU thing and several AMD low-end systems running as well. Here's my full list: https://setiathome.berkeley.edu/hosts_user.php?userid=3023
Anyhow, message me with instructions for how you want me to install and configure them and I'd be happy to help!


ID: 1836354 · Report as offensive     Reply Quote
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 5806
Credit: 76,039,961
RAC: 50,982
Russia
Message 1836356 - Posted: 16 Dec 2016, 20:41:24 UTC - in response to Message 1836354.  


Anyhow, message me with instructions for how you want me to install and configure them and I'd be happy to help!

Thanks for proposal. I'll PM you when/if some of those cards will be needed for specific testing.
SETI apps news
We're not gonna fight them. We're gonna transcend them.
ID: 1836356 · Report as offensive     Reply Quote
Ashutosh Agarwal

Send message
Joined: 6 May 15
Posts: 1
Credit: 6,850
RAC: 0
India
Message 1844014 - Posted: 23 Jan 2017, 12:49:37 UTC

Thanks!
ID: 1844014 · Report as offensive     Reply Quote
Profile Tom Miller
Volunteer tester
Avatar

Send message
Joined: 28 Nov 02
Posts: 768
Credit: 18,728,159
RAC: 20,308
United States
Message 1851226 - Posted: 25 Feb 2017, 14:34:06 UTC - in response to Message 1795063.  
Last modified: 25 Feb 2017, 14:34:53 UTC

I have a GT 710 that does Seti GPU tasks on my Developer computer. I will be migrating from a Intel Duo (2 cpu) to an Intel i5 (4 core) shortly. I would prefer to continue to run Win7.

I am perfectly willing to setup the GPU for any kind of testing needed. It is running nicely, doing a standard Seti WU with no parameter files in about an hour and half, and then dropped to maybe a half hour when I put in the stock small card parms for the SOG command line.

I have two GT750 Ti cards I am running on my Xeon W3565. I am running a mixture that includes Seti Beta there. And it is picking up beta GPU runs. I am willing to tinker with these systems as long as I can get fairly specific instructions.

I am an amateur Web Developer with some exposure to software testing.

So to sum up. I would like to have at least one 750 gpu continue doing nothing but production. But I would be willing to devote the other one to "whatever" :)
If possible.

I sometimes don't see PM's immediately. If push comes to shove email me at tlgalenson at yahoo.com and include SETI in the subject line.

Tom
"You are entitled to your own opinion but not to your own facts." Senator and Professor Patrick Moynihan
---
https://GalensonConsulting.WordPress.com
ID: 1851226 · Report as offensive     Reply Quote
Profile Juan Jiménez

Send message
Joined: 15 May 99
Posts: 9
Credit: 15,474,502
RAC: 383
Germany
Message 1871159 - Posted: 4 Jun 2017, 21:09:07 UTC - in response to Message 1795063.  

One of the compute nodes on my cluster has two Xeon E5-2670 10-core CPU's with hyperthreading, 128gb of memory and two Tesla K20Xm cards. I may be able to help with test builds (I'm running Centos 7) when the GPU's are idle.
ID: 1871159 · Report as offensive     Reply Quote
Bill GreeneProject Donor
Volunteer tester

Send message
Joined: 3 Jul 99
Posts: 74
Credit: 86,264,752
RAC: 47,609
United States
Message 1891855 - Posted: 26 Sep 2017, 2:04:21 UTC - in response to Message 1818588.  

Did the same for my 780ti and 980. Had to back out of that update.
ID: 1891855 · Report as offensive     Reply Quote
robertmiles
Volunteer tester

Send message
Joined: 16 Jan 12
Posts: 176
Credit: 2,178,696
RAC: 2,482
United States
Message 1901772 - Posted: 19 Nov 2017, 2:50:32 UTC - in response to Message 1795063.  
Last modified: 19 Nov 2017, 2:52:32 UTC

[snip]

Have no spare hardware or money but have testing skills and familiar with application tuning?
3. Run special test builds to help to adopt existing software to wide range of hardware devices of NVIDIA platform and help others to setup such builds for test.
Builds to be published in this thread or dedicated threads on Lunatics site (currently out of reach).

Have nothing but will to help?
4. Run special test builds on your own host and ask for help to install them on own host. There are enough peoples around who could help with initial setup.
Post links to build's processing results to help in testing.

I could do some test builds, but I'll need extra instructions for how to get started since I've never done builds for a BOINC project before.

Also, these builds can be done ONLY under Windows 10.

I have about 10 graphics board to get started with the testing, though. All Nvidia-based.
ID: 1901772 · Report as offensive     Reply Quote
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 5806
Credit: 76,039,961
RAC: 50,982
Russia
Message 1901804 - Posted: 19 Nov 2017, 6:31:03 UTC - in response to Message 1901772.  

I could do some test builds, but I'll need extra instructions for how to get started since I've never done builds for a BOINC project before.

Also, these builds can be done ONLY under Windows 10.

I have about 10 graphics board to get started with the testing, though. All Nvidia-based.

What toolchain do you plan to use? VisualStudio? GCC? smth else?
SETI apps news
We're not gonna fight them. We're gonna transcend them.
ID: 1901804 · Report as offensive     Reply Quote
robertmiles
Volunteer tester

Send message
Joined: 16 Jan 12
Posts: 176
Credit: 2,178,696
RAC: 2,482
United States
Message 1901848 - Posted: 19 Nov 2017, 13:21:16 UTC - in response to Message 1901804.  

I could do some test builds, but I'll need extra instructions for how to get started since I've never done builds for a BOINC project before.

Also, these builds can be done ONLY under Windows 10.

I have about 10 graphics board to get started with the testing, though. All Nvidia-based.

What toolchain do you plan to use? VisualStudio? GCC? smth else?

I have PARTS of the Visual Studio, GCC, and Cygwin tool chains installed. I'm familiar enough with the Cygwin tool chain to tell that it is probably not suitable for GPU work. I'm not familiar enough with the others to tell what I'm missing. However, I'll probably install any toolchain that is free and appears to be from a reliable source. I'm likely to need help getting started with any toolchain other than Cygwin, though.

I've finished an online CUDA class that used a toolchain on the class's server, but did not say enough about how to install a usable toolchain on my own computer and start using it. I no longer have access to the class's server.
ID: 1901848 · Report as offensive     Reply Quote
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 5806
Credit: 76,039,961
RAC: 50,982
Russia
Message 1901921 - Posted: 19 Nov 2017, 20:16:09 UTC - in response to Message 1901848.  

Then you have to learn some environment to work with first perhaps.
All Windows GPU builds were done with VS AFAIK.
All OpenCL ones - definitely.
So easiest way is to install VisualStudio environment and use VC++ toolchain.
For GPU build you'll need SDK from corresponding GPU vendor also.
SETI apps news
We're not gonna fight them. We're gonna transcend them.
ID: 1901921 · Report as offensive     Reply Quote
Previous · 1 . . . 10 · 11 · 12 · 13

Message boards : Number crunching : SETI applications for NVIDIA GPU improvement - how you can help


 
©2017 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.