PDA

View Full Version : GPU Crunching question



lazyshopper
10-29-14, 05:15 PM
I just bought 3 video cards & want to use 2 for crunching SETI. They are 1: GEForce GTX 650 2GB GDDR5 2: GEForce GTX 750 TI 2GB GDDR% Overclocked 3: GTX 280 1 GB GDDR3

What order should I implement these cards into my machines? I also have a Radeon HD6870 1 GB GDDR to throw in the mix. Any ideas? I guess I will add these to 3 of my machines & the bottom feeder is promised to someone else. I want to get the most bang for my buck. I will put these into 1 intel quad core machine & 2 AMD quad core machines. They may not be the best machine to crunch with but better than nothing.

Maxwell
10-29-14, 05:50 PM
I would keep the AMD and the NVidia cards in separate machines. Though it's possible to run both together, it's much easier to keep them in separate machines.

The GTX 280 is an older card, right? If you're going to avoid crunching on one of them, that would be the one. Otherwise, I'd stick one of each into the boxes. Doesn't really matter which card goes into which box for the most part.

For SETI, you'll want to run multiple WUs on each card, and probably reserve a full thread for GPU crunching (we can give more details on this later). Latest NVidia drivers. There is also an optimized package for SETI that would be great to install (can someone dig up the link?).

Keep asking questions. We're happy to help. :D

lazyshopper
10-29-14, 06:05 PM
Great. Between the GTX 280 & the HD 6870 which would you choose? I am totally novice to the world of GPU crunching. I also have one machine with a Zotac GT 630 & a Radeon HD 6870 installed but seti is only doing 1 WU at a time on the 4 cores & the HD 6870 card isn't being used at all from what I can see but I wouldn't know how to tell unless it bit me.

Maxwell
10-29-14, 06:13 PM
So, to get both cards crunching, you'll want to add a "use_all_gpus" tag to your cc_config.xml file. Check out this to get an idea of what I'm talking about: http://www.setiusa.us/faq.php?faq=manage#faq_manage4

I'd dig up more, but I'm still at work. I'll try and get back to this tonight, unless others want to jump in.

lazyshopper
10-29-14, 06:25 PM
I added this -

<cc_config>
<log_flags>
[ ... ]
</log_flags>
<options>
[ ... ]
</options>
</cc_config>

<cc_config>
<options>
<use_all_gpus>
</options>
</cc_config>

I also got this message after adding the config

10/29/2014 6:27:50 PM | | Unexpected text in cc_config.xml: [ ... ]
10/29/2014 6:27:50 PM | | Not using a proxy
10/29/2014 6:27:50 PM | | log flags: file_xfer, sched_ops, task
10/29/2014 6:27:56 PM | | Re-reading cc_config.xml

Al
10-29-14, 07:44 PM
<cc_config>
<options>
<report_results_immediately>1 </report_results_immediately>
<use_all_gpus>
</options>
</cc_config>

Just use the above and you should be golden.

John P. Myers
10-29-14, 08:21 PM
Doesn't it need to be <use_all__gpus>1</use_all_gpus> or has something changed?

Sent from my Galaxy S4 using Tapatalk Pro

zombie67
10-29-14, 08:46 PM
D'oh! Yeah, I was so busy looking at the syntax, I missed the content.

http://boinc.berkeley.edu/wiki/Client_configuration

<cc_config>
<options>
<use_all_gpus>1</use_all_gpus>
</options>
</cc_config>

Al
10-29-14, 09:12 PM
D'oh! Yeah, I was so busy looking at the syntax, I missed the content.

http://boinc.berkeley.edu/wiki/Client_configuration

<cc_config>
<options>
<use_all_gpus>1</use_all_gpus>
</options>
</cc_config>
Whoops, me too.

lazyshopper
10-29-14, 09:16 PM
Thanks for all the help. I got everything updated & didn't get any errors. Now, I will sit back & see what it does.

Fire$torm
10-29-14, 11:00 PM
I would keep the AMD and the NVidia cards in separate machines. Though it's possible to run both together, it's much easier to keep them in separate machines.....

Sorry Max but you got that part wrong. It is very easy, and my preferred setup, to run one ATI/AMD & one nVidia on the same machine. I never have to worry about matching GPU family types/series, never have to add any special flags to cc_config, etc...

Also I never have to worry about any machine being completely idle if one type of GPU project runs out of work, has a new app that fails, the server crashes or just goes dark for an unspecified period of time.

John P. Myers
10-30-14, 04:17 AM
I used to mix AMD and Nvidia but i keep them separate now too. It definitely is less of headache imo :)

FourOh
10-30-14, 09:43 AM
Great. Between the GTX 280 & the HD 6870 which would you choose? I am totally novice to the world of GPU crunching. I also have one machine with a Zotac GT 630 & a Radeon HD 6870 installed but seti is only doing 1 WU at a time on the 4 cores & the HD 6870 card isn't being used at all from what I can see but I wouldn't know how to tell unless it bit me.

Definitely the 6870. It uses way less electricity and should get comparable results on SETI or Primegrid. In fact, you might consider selling the 280 and putting the $$ towards something more efficient. I really like the 750 Ti, glad to see you have one of those. The 750 Ti is a standout on the GPUGrid project... are you interested in SETI@Home only for now?

lazyshopper
10-30-14, 12:47 PM
Right now I am running seti & WCG. Later after I get my machines up to speed I may expand into some of the other projects. Thanks for the input on the 6870.

lazyshopper
10-30-14, 10:56 PM
Another question - How do I get more than 1 seti WU to run at a time using GPUs? Also, is there a way to optimize the processes?

zombie67
10-31-14, 02:23 AM
Another question - How do I get more than 1 seti WU to run at a time using GPUs?

http://boinc.berkeley.edu/wiki/Client_configuration

Add a file called app_config.xml to the project folder. The format is:

<app_config>
[<app>
<name>uppercase</name>
<max_concurrent>1</max_concurrent>
[<fraction_done_exact/>]
<gpu_versions>
<gpu_usage>.5</gpu_usage>
<cpu_usage>.4</cpu_usage>
</gpu_versions>
</app>]
...
[<app_version>
<app_name>uppercase</app_name>
[<plan_class>mt</plan_class>]
[<avg_ncpus>x</avg_ncpus>]
[<ngpus>x</ngpus>]
[<cmdline>--nthreads 7</cmdline>]
</app_version>]
...
[<project_max_concurrent>N</project_max_concurrent>]
</app_config>

In this example:
- app is astropulse app
- gpu usage .5 = two tasks, .333 = three tasks, .25 = four tasks, etc.
- CPU usage .333 reserves a third of a thread, .5 reservs half a thread, 1 a full thread, etc.

<app_config>
<app>
<name>astropulse_v7</name>
<gpu_versions>
<gpu_usage>.5</gpu_usage>
<cpu_usage>.333</cpu_usage>
</gpu_versions>
</app>
</app_config>


Also, is there a way to optimize the processes?

Yes, install the lunatics optimized apps. If you are windows, use the all-in-one installer from here:

http://www.arkayn.us/forum/index.php?PHPSESSID=9b10b7271446ac228a13a864238031 d9&

It will do all the app_info.xml and other file stuff. You can combine that with an app_config.xml to more easily manage the number of GPU tasks per GPU and the threads reserved for GPU tasks.