HCC GPU Launched.

[Movieman]

I'd love to see someone try. Also it'd be interesting to see if performance dips at all by using a narrower PCI-e lane.

It's a X8 so shouldn't be an issue..So, send the card and we'll find out!

[Conquistador SW]

Did I miss it? Did we get some clarity on what is "primary cpu time" and what is "secondary cpu time"?

Not really haha. I think he meant CPU time before the GPU kicks in and after it stops. But in my case, the CPU is at full load all the time, so except if I am going to measure those times with a stopwatch there is no way to find these.

[bluestang]

Ok, so I pulled my HD 4670 card out and replaced it with the 7870. Can I put the 4670 back in and run them both...the 7870 for crunching only and the 4670 just to handle my display? That would help avoid any lagging (very, very minor), correct? I would love to do this and then run 2 WU at once on the 7870

[OldChap]

Did I miss it? Did we get some clarity on what is "primary cpu time" and what is "secondary cpu time"?

After the more clever of our members pointed out that those numbers combined are logged as cpu time under the properties button on the task page I am thinking that comparing that cpu time between various types of rig/processor may give useful info..... especially if individuals were to give those numbers they get at different clocks or ht states.

What a waste of a day here.... I forgot that unless you change settings windows shuts a rig down if you don't use it ....can't they tell it is working

[pirogue]

While doing a little GPU overclocking, I discovered the hardware monitor in Afterburner and did an "EKG" (ECG?). Anyway, both the peaks and troughs on the GPU usage last around a minute. The GPU utilization peaks around 84% and the wattage (measured by meter on UPS) goes from 83 to 135 during the peaks. The temp ranges from 42C to 49C. The fan ranges from 20% to 23%.



Link to full view: http://www.wcgdaws.com/HCC-GPU-EKG.PNG

Maybe it's interesting. Maybe it's not.

P.S. This is for a 2600K running a 7870 with GPU work only.

10/15/2012 0:000:17:36:03 104,456 458
10/14/2012 0:001:12:48:41 208,592 921
10/13/2012 0:001:10:54:03 181,188 806

[Movieman]

^^ What I found during my tests which took one short evening of reading is that I need to buy a 7870 for each machine and then not ever have to think about anything..

[Rob_B]

That is one way to do it.

[0ne.shot]

A 7870 has twice the SPs as a 7770, but does the performance scale linearly? Or does the longer CPU processing time take away some of the time saved using a faster card?

[shoota]

haha. Just when I sell off all my ati cards and powerful psus to run them.. this happens. duh.

[pirogue]

^^ What I found during my tests which took one short evening of reading is that I need to buy a 7870 for each machine and then not ever have to think about anything..

Needless to say, I'm pretty happy with my decision. At +60 watts over the 5450 I was using, I'll hopefully be able to continue to run it over the winter without going broke.

[artemm]

Has anyone tested the effects of single CPU core performance on WU completion times? If you're using the tweak to run only GPU work units, wouldn't it be highly beneficial to turn HT off, fr instance, in order to maximize performance? If you want to get really extreme about it, overclocking cores individually and setting affinity or even disabling all cores but the fastest would work too, right?

[OldChap]

Got a few answers on that in about half an hour.....

[pirogue]

A 7870 has twice the SPs as a 7770, but does the performance scale linearly? Or does the longer CPU processing time take away some of the time saved using a faster card?

I don't have a 7770 to compare, but if you do, here are the times for a random WU on a 7870:

Estimated kernel execution time = 0.22682 [sec]
Starting analysis of ../../projects/www.worldcommunitygrid.org/X0960069740712200605191425_X0960069740712200605191425.jp2...
Extracting GLCM features...
Total kernel time: 29.500572 (1026 kernel executions)
Total memory transfer time: 1.330714
Average kernel time: 0.028753
Min kernel time: 0.027096 (dx=25 dy=5 sample_dist=24 )
Max kernel time: 0.030491 dx=1 dy=1 sample_dist=0
INFO: GPU calculations complete.
Total time for ../../projects/www.worldcommunitygrid.org/X0960069740712200605191425_X0960069740712200605191425.jp2: 84 seconds
Finished Image #0, pctComplete = 1.000000
CPU time used = 53.430343

[Otis11]

While doing a little GPU overclocking, I discovered the hardware monitor in Afterburner and did an "EKG" (ECG?). Anyway, both the peaks and troughs on the GPU usage last around a minute. The GPU utilization peaks around 84% and the wattage (measured by meter on UPS) goes from 83 to 135 during the peaks. The temp ranges from 42C to 49C. The fan ranges from 20% to 23%.



Link to full view: http://www.wcgdaws.com/HCC-GPU-EKG.PNG

Oh... that raises some points - first, what does this usage graph look like if you run 2WUs on a card? Can we smooth out that idle time with another WU?

Second - those fluctuations are worse for the card than simply running hot. The largest cause of Solid State Electronics failure (atleast at two of the companies I've worked for) was mechanical stresses caused by continuous fluctuations in temperature. It's actually much better to just run them hot than to let them fluctuate wildly. Now these 2-6C fluctuations don't matter, just throwing that out there for anyone who sees more rapid temp changes (say if you're cooling with water maybe? idk...)

Third - fans are designed to run at relatively constant speeds for long, long periods of time without failure, but accelerating and decelerating causes them problems - so it would be better to fix the fan at say, 25% in your case and just let it run. Just be careful that that doesn't cause larger temperature swings!

I'm PROBABLY nitpicking about nothing - but I want to raise the issue before things fail because of it. Crunching already gets a bad rap for hardware failure a lot when it's not the crunching that killed it. Just thinking if I say something before hand it might hold more water IF something goes wrong.

[Conquistador SW]

Oh... that raises some points - first, what does this usage graph look like if you run 2WUs on a card? Can we smooth out that idle time with another WU?

Second - those fluctuations are worse for the card than simply running hot. The largest cause of Solid State Electronics failure (atleast at two of the companies I've worked for) was mechanical stresses caused by continuous fluctuations in temperature. It's actually much better to just run them hot than to let them fluctuate wildly. Now these 2-6C fluctuations don't matter, just throwing that out there for anyone who sees more rapid temp changes (say if you're cooling with water maybe? idk...)

Third - fans are designed to run at relatively constant speeds for long, long periods of time without failure, but accelerating and decelerating causes them problems - so it would be better to fix the fan at say, 25% in your case and just let it run. Just be careful that that doesn't cause larger temperature swings!

I'm PROBABLY nitpicking about nothing - but I want to raise the issue before things fail because of it. Crunching already gets a bad rap for hardware failure a lot when it's not the crunching that killed it. Just thinking if I say something before hand it might hold more water IF something goes wrong.

Yeah, it would be better if the GPU was at full load all the time.

With watercooling I noticed that my GTX 260 idles at 36C and full load for an hour at 40C (according to MSI afterburner), so a very small difference.

[pirogue]

Oh... that raises some points - first, what does this usage graph look like if you run 2WUs on a card? Can we smooth out that idle time with another WU?

I don't think it's possible without constantly pausing/resuming. I tried.

Based on a short sampling:

• GPU utilization goes up 6%
• the temps vary between 48C and 57C. (It's also a warm day today).
• the power usage goes up by 7 or 8 watts
• the fan varies between 23-30%
• total runtime per WU went up by around 25 seconds.

I'll have to see if I notice a change in my stats tomorrow.

Second - those fluctuations are worse for the card than simply running hot. The largest cause of Solid State Electronics failure (atleast at two of the companies I've worked for) was mechanical stresses caused by continuous fluctuations in temperature. It's actually much better to just run them hot than to let them fluctuate wildly. Now these 2-6C fluctuations don't matter, just throwing that out there for anyone who sees more rapid temp changes (say if you're cooling with water maybe? idk...)

I'm hoping it can handle the fluctuations I'm seeing. If it can't, I'm hoping it craps out during the warranty period.

Third - fans are designed to run at relatively constant speeds for long, long periods of time without failure, but accelerating and decelerating causes them problems - so it would be better to fix the fan at say, 25% in your case and just let it run. Just be careful that that doesn't cause larger temperature swings!

Based on the handy log files afterburner will create, I can see the the fan varies from around 995-1035RPM. (See warranty statement above.)

I'm PROBABLY nitpicking about nothing - but I want to raise the issue before things fail because of it. Crunching already gets a bad rap for hardware failure a lot when it's not the crunching that killed it. Just thinking if I say something before hand it might hold more water IF something goes wrong.

See warranty statement above.

[Manicdan]

did anyone test a 680 yet? and how hard would it be for me to do it if i never used bionic before?

[Mydog]

did anyone test a 680 yet? and how hard would it be for me to do it if i never used bionic before?

Running a 680L here without any problems, only one WU at the time on the GPU now.

[Otis11]

did anyone test a 680 yet? and how hard would it be for me to do it if i never used bionic before?

Harder than CPU Crunching... but then again, that's not really saying much.

Given a few minutes of patience and talking to the guys on this thread I'd say no trouble at all. Trying to get the max performance out of it on the other hand, will take some tweaking. But just get it running for now and then let these other smart guys fill us in on the results of their performance tests.

[OldChap]

Got a few answers on that in about half an hour.....

Sorry need to re- run that test sequence ... I'll be back later

EDIT: It is hard to get good averages over a short period so the figures don't seem to add up particularly well however the trend can be clearly seen so pick the bones out of this:

My thought is that a single cpu thread is getting between 800 and 1000ppd and ht off is therefore costing 3200 and 4000ppd. Ask yourself.... how many cards do I have to run to gain that much?

Mine are low end ati cards.... you will have to see for yourselves what the difference is with better cards.

[Snow Crash]

<use with caution :->
this app_info file runs 2 WU per GPU AND runs CPU (both for hcc only)

<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_hcc1</name>
      <executable/> 
   </file_info>
   <file_info>
      <name>hcckernel.cl.6.56</name>
      <executable/>
   </file_info>

   <file_info>
      <name>wcg_hcc1_img_6.56_windows_intelx86</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__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>

<app_version>
    <app_name>hcc1</app_name>
    <version_num>656</version_num>
    <platform>windows_intelx86</platform>
    <avg_ncpus>1.000000</avg_ncpus>
    <max_ncpus>1.000000</max_ncpus>
    <flops>3231932486.733310</flops>
    <api_version>6.13.0</api_version>
    <file_ref>
        <file_name>wcg_hcc1_img_6.56_windows_intelx86</file_name>
        <main_program/>
    </file_ref>
</app_version>

</app_info>

[kjeldoran]

I just transcribed the HCC CPU part to my app_info.xml and I get computation errors on the CPU WUs instantly. Message is 10/15/2012 9:34:06 PM | World Community Grid | Output file X0930069780341200605191747_1_0 for task X0930069780341200605191747_1 absent. Not really sure what to do, but I am just removing the code for now.

My code looks like this:

<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__ati_hcc1</name>
<executable/>
</file_info>
<file_info>
<name>hcckernel.cl.6.56</name>
<executable/>
</file_info>
<file_info>
<name>wcg_hcc1_img_6.56_windows_intelx86</name>
<executable/>
</file_info>

<app_version>
<app_name>hcc1</app_name>
<version_num>656</version_num>
<platform>windows_intelx86</platform>
<plan_class>ati_hcc1</plan_class>
<avg_ncpus>1.0</avg_ncpus>
<max_ncpus>1.0</max_ncpus>
<coproc>
<type>ATI</type>
<count>.5</count>
</coproc>
<file_ref>
<file_name>wcg_hcc1_img_6.56_windows_intelx86__ati_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_version>
    <app_name>hcc1</app_name>
    <version_num>656</version_num>
    <platform>windows_intelx86</platform>
    <avg_ncpus>1.000000</avg_ncpus>
    <max_ncpus>1.000000</max_ncpus>
    <api_version>6.13.0</api_version>
    <file_ref>
        <file_name>wcg_hcc1_img_6.56_windows_intelx86</file_name>
    </file_ref>
    <main_program/>
</app_version>

</app_info>

[Red Maw]

For random information etc.

i7 930 @3.8 Ghz, HD 5850 - 1 WU takes ~4:15, however it reaches 99.415% before the 3:15 mark lol

It also makes my rig really slow and annoying to use.

[=[PULSAR]=]

Apparently I've returned just in time from over a year long hiatus. Enabled GPU's? Now this gives me a reason to start stock piling cards again. Running on a GTX480 right now lets see how this goes

EDIT: For nvidia cards what drives the WU gpu clock or shader clocks?

[fallwind]

And here's my version to run 2 HCC on my GPU and SN2S on the rest of the CPU:

<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_hcc1</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>2.0</max_ncpus>
      <coproc>
         <type>CUDA</type>
         <count>.5</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>

<app>
    <name>sn2s</name>
    <user_friendly_name>Say No to Schistosoma</user_friendly_name>
</app>

   <file_info>
      <name>wcgrid_sn2s_vina_6.20_windows_x86_64</name>
      <executable/>
   </file_info> 
   <file_info>
      <name>wcgrid_sn2s_vina_prod_x86_64.exe.6.20</name>
      <executable/>
   </file_info>
   <file_info>
      <name>wcgrid_sn2s_gfx_prod_x86_64.exe.6.20</name>
       <executable/>
   </file_info>

<app_version>
    <app_name>sn2s</app_name>
    <version_num>620</version_num>
    <platform>windows_x86_64</platform>
    <avg_ncpus>1.0</avg_ncpus>
    <max_ncpus>1.0</max_ncpus>
    <flops>3347548492.458962</flops>
    <api_version>7.1.0</api_version>
    <file_ref>
        <file_name>wcgrid_sn2s_vina_6.20_windows_x86_64</file_name>
        <main_program/>
    </file_ref>
    <file_ref>
         <file_name>wcgrid_sn2s_vina_prod_x86_64.exe.6.20</file_name>
          <open_name>AutoDockVina64.exe</open_name>
     </file_ref>
     <file_ref>
         <file_name>wcgrid_sn2s_gfx_prod_x86_64.exe.6.20</file_name>
          <open_name>graphics_app</open_name>
     </file_ref>
</app_version>

<app_version>
    <app_name>sn2s</app_name>
    <version_num>620</version_num>
    <platform>windows_intelx86</platform>
    <avg_ncpus>1.000000</avg_ncpus>
    <max_ncpus>1.000000</max_ncpus>
    <flops>3347548492.458962</flops>
    <api_version>7.1.0</api_version>
    <file_ref>
        <file_name>wcgrid_sn2s_vina_6.20_windows_x86_64</file_name>
        <main_program/>
    </file_ref>
    <file_ref>
         <file_name>wcgrid_sn2s_vina_prod_x86_64.exe.6.20</file_name>
          <open_name>AutoDockVina64.exe</open_name>
     </file_ref>
     <file_ref>
         <file_name>wcgrid_sn2s_gfx_prod_x86_64.exe.6.20</file_name>
          <open_name>graphics_app</open_name>
     </file_ref>
</app_version>
</app_info>

Man did I ever blow out alot of WU's getting this to work properly!