HCC GPU Launched.

[bluestang]

Hmmm, thanks fallwind. Well, maybe I'll keep the "free" 7870 and return the other 7870 and call it a day. Time will tell as I have a week to figure it out, but this is XS, so I'm not sure.

[werdwerdus]

depends on how many cpu cores you have available. remember each GPU task needs a dedicated core!

[Movieman]

depends on how many cpu cores you have available. remember each GPU task needs a dedicated core!

So on the Beast I could run 32 GPU WU?

[Snow Crash]

So on the Beast I could run 32 GPU WU?

You mean you're NOT???

[my87csx481]

You had me worried for a minute. My 7970 is completing 18WUs in that time frame... still wondering what a GTX 680 loaded with multiple tasks would do. Nvidia was always the dominant card with F@H GPU WUs. I about shiet my drawers when I read the ATI appeared to be the best for HCC WCG!!!

I could test this. You guys just tell me what to do. I read in this thread to not run multiple gpu threads if your usage is ~95%. On my GTX 680, it is at 95% with one thread, so I never bothered screwing around with multiple threads. It currently runs 1 gpu thread HCC, another gpu thread HCC on a GTX 460, and 6 threads of GFAM on the cpu (3770k at 4.7).

I'm gonna' pull the 460 and put it into something else, probably this evening, depending on how I feel (feel like today, shouldn't be at work).....so, let me know what to do/try.

Thanks.

Edit: I mis-typed....mine is a 670, NOT a 680.

[Bun-Bun]

Well my GTX670 produces roughly half what a 7850 does. I don't imagine a GTX680 would do much better. My GTX670 does gain a bit from 2 threads vs 1. About a 4:3 ratio.

[Mydog]

My GTX480 is about 30% faster than my GTX680L from what I can see here running 5 WU's on each in the same rig.

[nanoprobe]

Fixed
6 WU's made 110,000 Boinc PPD


And you couldn't do 40 GPU WU's because you'd run out of CPU cores/threads, I would say max 32 but I'd reccomend 30. Testing 8 WU's now on the 7970 and the can't seem to get out of sync so that the load evens out. The WU's act a bit like the traffic in the rush hour, as soon as one or two leaves the GPU and start on the CPU the others play catch up as the GPU load is getting easier. I also get 100% CPU usage on my 3770K when the WU's finish on the GPU.

I'm pretty sure I read that WCG has a daily limit of 2500 WUs per machine so having a machine that can do more than that would be left idling at times. If I can find the link again I'll post it.

[Mydog]

I'm pretty sure I read that WCG has a limit of 2500 WUs per machine so having a machine that can do more than that would be left idling at times. If I can find the link again I'll post it.

I can't see my numbers going down on that rig other that the poweroutage for 1.5 hours yesterday. Looks like it still chews thru 3500 WU'S every day however the pending validation list is increasing fast so maybe we only get 2500 validated.

[my87csx481]

My GTX480 is about 30% faster than my GTX680L from what I can see here running 5 WU's on each in the same rig.

How do you tell which one is which in the results (or in boinc manager)?

This is a GPU only rig with that configuration, right? Care to share your app_info file (so I can test mine out)?

Thanks.

[Mydog]

How do you tell which one is which in the results (or in boinc manager)?

This is a GPU only rig with that configuration, right? Care to share your app_info file (so I can test mine out)?

Thanks.

I just stopped networking on that rig and checked the times on the WU's, approx 10 mins(GTX480) vs 14 mins(GTX680L) You can also see it if you go in to your WCG page and look at the result details.

app_info.xml for Nvidia that I use, GPU WU's only on that rig.

<app_info>
<app>
<name>hcc1</name>
<user_friendly_name>Help Conquer Cancer</user_friendly_name>
</app>
<file_info>
<name>wcg_hcc1_img_6.56_windows_intelx86__nvidia_h cc1</name>
<executable/>
</file_info>
<file_info>
<name>hcckernel.cl.6.56</name>
<executable/>
</file_info>
<app_version>
<app_name>hcc1</app_name>
<version_num>656</version_num>
<platform>windows_intelx86</platform>
<plan_class>nvidia_hcc1</plan_class>
<avg_ncpus>1.0</avg_ncpus>
<max_ncpus>0.783972</max_ncpus>
<coproc>
<type>CUDA</type>
<count>.5</count>
</coproc>
<file_ref>
<file_name>wcg_hcc1_img_6.56_windows_intelx86__nvi dia_hcc1</file_name>
<main_program/>
</file_ref>
<file_ref>
<file_name>hcckernel.cl.6.56</file_name>
<open_name>hcckernel.cl</open_name>
</file_ref>
</app_version>
</app_info>

[0ne.shot]

[s]Is the space intentional?

line 26

<file_name>wcg_hcc1_img_6.56_windows_intelx86__n vi dia_hcc1</file_name>[/s]

Nevermind. It added another space when I copy and pasted. Is the space in the actual code?

[bluestang]

No spaces.

[my87csx481]

Do you have this in your cc_config to run on both gpus?:

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

Also, I have to modify the count right? To get 5 per gpu, I need to have .2 as the count, right?

[Mydog]

Do you have this in your cc_config to run on both gpus?:

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

Also, I have to modify the count right? To get 5 per gpu, I need to have .2 as the count, right?

Yes to all

[my87csx481]

Thanks for the clarification.

I'll play with it when I get home.

[my87csx481]

I don't know....event log says "not reporting or requesting new tasks"

I'm trying to run 4 on one gpu, and 4 on another gpu.

I read something in Snow Crash's thread about telling BOINC where to get the files or something....but didn't see any further explanation about that.

Here's my app_info:

<app_info>
   <app>
      <name>hcc1</name>
      <user_friendly_name>XtremeSystems Crunches Cancer</user_friendly_name>
   </app>

   <app_version>
      <app_name>hcc1</app_name>
      <version_num>656</version_num>
      <platform>windows_intelx86</platform>
      <plan_class>nvidia_hcc1</plan_class>
      <avg_ncpus>1.0</avg_ncpus>
      <max_ncpus>1.0</max_ncpus>
      <coproc>
         <type>CUDA</type>
         <count>.25</count>
      </coproc>
      <file_ref>
          <file_name>wcg_hcc1_img_6.56_windows_intelx86__nvidia_hcc1</file_name>
          <main_program/>
      </file_ref>
      <file_ref>
         <file_name>hcckernel.cl.6.56</file_name>
         <open_name>hcckernel.cl</open_name>
      </file_ref>
   </app_version>

   <file_info>
      <name>wcg_hcc1_img_6.56_windows_intelx86__nvidia_hcc1</name>
      <executable/> 
   </file_info>
   <file_info>
      <name>hcckernel.cl.6.56</name>
      <executable/>
   </file_info>

</app_info>

[Bun-Bun]

BOINC defaults to not doing GPU work while computer is in use. Manually set activity to use GPU always and it will start downloading WU's.

[my87csx481]

Did that already....I don't think that got wiped out when I added the app_info.xml....but I double checked, and it's ticked now.

Edit: Maybe this is the problem?

0001.jpg

[pirogue]

In order to get the application downloaded, I think you need to download at least one WU without the app_info.

[Mydog]

Did that already....I don't think that got wiped out when I added the app_info.xml....but I double checked, and it's ticked now.

Edit: Maybe this is the problem?

0001.jpg

I got the same thin when I copied from here but when I copied it from the WCG forum it worked.

try this one

app_info.txt

Just rename it

[Sparky]

OK now I'm confused, I haven't changed a thing but now my log file says "tasks are available for ATI GPU but your preferences are set to not accept them." WTH!
*edit* found a check box on the WCG page to allow GPU units, and it was unchecked. I didn't do that. Set it back to checked. Wacky. Now to figure out how to load down my card better.
*edit2* My card temp barely wavers a couple degrees, but even at 84% load it reports the clocks are still at 2D 500MHz?
*edit3* dang, I'm still on 12.4 drivers That might be part of it. Downloading the latest drivers now.

[nanoprobe]

Did that already....I don't think that got wiped out when I added the app_info.xml....but I double checked, and it's ticked now.

Edit: Maybe this is the problem?

0001.jpg

Try detaching and reattaching to WCG. When you reattach let it download some GPU tasks. Then shut down Boinc and drop your app_info file back into the WCG project folder. Restart boinc. Hopefully that will fix your problem.

[my87csx481]

In order to get the application downloaded, I think you need to do at least one WU without the app_info.

This did it....sort of. I stopped everything, removed the app_info, restarted everything, BOINC took a minute but d/l'd some new work units (GPU only now), cards only worked on 1 each, I suspended and stopped everything again, replaced the app_info, restarted everything again....all is working now.

Thanks all for your help.

[pirogue]

This did it....sort of. I stopped everything, removed the app_info, restarted everything, BOINC took a minute but d/l'd some new work units (GPU only now), cards only worked on 1 each, I suspended and stopped everything again, replaced the app_info, restarted everything again....all is working now.

Thanks all for your help.

Good. Glad to hear that.