Focusing on World Community Grid

[ZipSpeed]

I've been crunching away the last few months on WCG/HCC, on my LLano @ 3.0Ghz, 8GB RAM. Set up to do GPU WUs on the IGP (of which it does fairly well), as well as my three other CPU cores doing regular WCG CPU WUs. (One CPU core dedicated to HCC GPU WUs.)

Crunching while sipping power! Have you measure how much power it draws?

 

[blckgrffn]

Crunching while sipping power! Have you measure how much power it draws?

The A8 Trinity I was running on WCG for a while was pulling ~140W, GPU and CPU locked an loaded. It beat out a 2600k in average work per day by a couple k, I believe, while pulling the same power.

The cool thing was that it ran at full boost frequencies the entire time on all cores. Heatsink was an Artic Cooler Silencer and that didn't even get loud.

If you did a project where GPU power was better rewarded (poem, etc.) then the differential in PPD CPU vs APU would be huge.

 

[StitchExperimen]

We moved up one more Mar 13 to 65th place in world... AND 30th in Country... Mar 13, 2013.
We passed KAUST.

In 3 more days we get 64th position world and 29th country from "Crunching@EVGA"

And I missed it but we passed "Ubuntu Linux" at 67th place this March 2013 also.

 

[GLeeM]

And I missed it but we passed "Ubuntu Linux" at 67th place this March 2013 also.

You did not miss it ... you posted it in the "One-Month WCG Team Competition" thread (post #52) - probably by mistake

We moved up one more to 66th place in world... AND 30th in Country... March 7, 2013.
We passed Ubuntu Linux.

 

[sangyup81]

We passed a billion points. I see the team at 1,000,958,491. It couldn't have been that long ago that we passed a billion!

 

[GLeeM]

I see the team at 1,000,958,491.

Those are WCG points.

BOINC Points = 143,535,469

 

[Icecold]

Why does my team status on WCG say 'Joined-Retired' ? Here is a screenshot:

 

[GLeeM]

I think "Joined=03/23/2013" and "Retired=Current".

Meaning you are Current and not yet Retired

 

[Icecold]

I think "Joined=03/23/2013" and "Retired=Current".

Meaning you are Current and not yet Retired

Hehe, you're right, I was definitely reading that wrong. Thanks! I guess that's what I get for posting before my morning coffee on a Saturday lol.

 

[Icecold]

Are Nvidia cards not really worth it for crunching HCC on WCG at this point? I've seen posts on other forums where people are using fairly old or low end AMD cards and are saying they can run multiple WU's on their card at the same time, and it's completing each within a couple of minutes.(but bear in mind most information I've found is from when it was still in BETA; I'm having difficulty finding current information)

On my GTX470 it seems to take around 8-9 minutes per WU, and that's only running 1 at a time, but maybe something has changed since it came out of BETA or there's something else that I don't understand. Any thoughts? Or is it just time to purchase a used 5850 on the cheap if I want to crunch WCG on a GPU?

Thanks for any insight!

 

[GLeeM]

On my GTX560 Ti 448 core (OCed to 850) it seems to take around 9-11 minutes per WU when running three WUs at a time.

I get a little desktop lag though. And each instance uses one thread (CPU). I can give you easy directions if you want to try two or more.

 

[Icecold]

I had set my app_config.xml earlier and put it in

C:\Users\All Users\BOINC\projects\www.worldcommunitygrid.org

and

C:\ProgramData\BOINC\projects\www.worldcommunitygrid.org

but it didn't seem to take.

I'm pretty sure I had my app_config.xml setup correctly, so I figured either I was putting it in the wrong directories or that it was because I have a ton of other WU pending for Correlizer. I was going to let everything finish for all projects and then try it again.

I was mostly curious if users of AMD cards really are getting such good performance out of them, or if that was primarily just when the project was in beta. I'm not necessarily against buying an AMD card(have 2 4890's already that I'm unsure if I can use to crunch any projects, maybe Poem@home but i know they're not new enough for WCG). I just don't want to 'waste' my GTX 470 on WCG if an AMD card would stomp all over it, since I could always put the GTX 470 towards folding(which gets 17-22k ppd) and buy a 5850, 6870, or 7770 for ~$100 and throw that at WCG. BUT, if you think it's worthwhile to leave it running on WCG, I will do that.

Edit: c : \ was making a :\ icon. Is there any way to make it not show smile faces when typing in c : \ other than putting it in [ code ] boxes which makes it not show up inline with other text?

 

[GLeeM]

CProgramData\BOINC\projects\www.worldcommunitygrid.org but it didn't seem to take.

BOINC must be restarted so it can read the file.

BUT, if you think it's worthwhile to leave it running on WCG, I will do that.

I am same as you in NOT knowing how well AMD does, so I cannot answer this.

I know what you mean trying to put resources where they do best ... last summer I had a AMD 4850 quit ... it did MANY times better than Nvidia in Milkyway. It doesn't seem worth it to use Nvidia to crunch Milkyway.

 

[Rattledagger]

BOINC must be restarted so it can read the file.

If you've got one of the latest test-clients it should be enough to re-read config-file in BOINC Manager, atleast v7.0.58 says it finds the file:
01.04.2013 03:04:05 | World Community Grid | Found app_config.xml

I am same as you in NOT knowing how well AMD does, so I cannot answer this.

I know what you mean trying to put resources where they do best ... last summer I had a AMD 4850 quit ... it did MANY times better than Nvidia in Milkyway. It doesn't seem worth it to use Nvidia to crunch Milkyway.

Milkyway is double-precision, and especially the Nvidia-6xx-series of cards has really dreadful double-precision-performance, so performance in Milkyway can't directly be compared to any other project.

WCG uses OpenCL, and this puts Nvidia-cards at a disadvantage since Nvidia wants everyone to use Cuda instead. I've never managed to get my Ati-5850 to run two tasks at once without one task becoming basically stuck after some hours crunching, so I've left it at one task at once, something that seems to take 9-11 minutes.

On an Amd-7950 on the other hand, it seems to average around 8 minutes/8 tasks.

 

[GLeeM]

On an Amd-7950 on the other hand, it seems to average around 8 minutes/8 tasks.

w00t!

Wow

How much longer are there going to be GPU WUs in WCG?

 

[StitchExperimen]

hcc1 is going to be done in a couple of months with so many people participating so buying an AMD card might not be the best thing to do because I would imagine that a lot of people are going to move over to Poem and flood that program and they might not have enough work is speculation. I bought two Nvidia cards and put them on GPUGRID. They are 660 Ti's and do long work units and get extra credit if done under a certain time period.

 

[blckgrffn]

hcc1 is going to be done in a couple of months with so many people participating so buying an AMD card might not be the best thing to do because I would imagine that a lot of people are going to move over to Poem and flood that program and they might not have enough work is speculation. I bought two Nvidia cards and put them on GPUGRID. They are 660 Ti's and do long work units and get extra credit if done under a certain time period.

I can't even get any work for POEM right now, darn it. My old 4770 was doing 90k ppd running only 1WU, back when it was running four it was turning ~140k easily

 

[Rattledagger]

w00t!

Wow

Unfortunately HCC has given me problems last couple times tried running it, maybe it's the drivers it's not liking...

hcc1 is going to be done in a couple of months with so many people participating so buying an AMD card might not be the best thing to do because I would imagine that a lot of people are going to move over to Poem and flood that program and they might not have enough work is speculation. I bought two Nvidia cards and put them on GPUGRID. They are 660 Ti's and do long work units and get extra credit if done under a certain time period.

Hopefully WCG will have another GPU-project ready by the time HCC is finished.

In the mean-time, AFAIK all GPU-projects except GPUGRID works with Amd-cards and while Amd isn't running circles around Nvidia-cards like in Milkyway, they'll still have good performance in most of the other projects.

By the time HCC is finished it's even possible Folding@home finally manages to release a decent Amd-GPU-core...

 

[Rattledagger]

Apparently HCC has now less than 1 month to go:

Based on current estimates, there is now approximately 25 days of new work remaining for the Help Conquer Cancer project. Once all new work has been sent out for the project, only resend work units will be available (work units not previously computed correctly or on time).

The launch of GPU processing has significantly accelerated the completion of this research and as a result, Help Conquer Cancer will complete over 2 1/2 years sooner than would be possible otherwise. We are also looking at our future projects to see if they could benefit from GPU processing as well.

Seippel

 

[ZipSpeed]

Hmm, time to put more GPUs on HCC before the work runs out! I've actually had to shut down a couple rigs the last couple days due to unusually warm weather. The weather is cooling down later this week, so I will put those two rigs back to work then

 

[ljtatej]

Hi all! I have decided to donate some computer time to the team. I downloaded the latest client, but I can't seem to use the second gpu on my 690 for HCC. Anyone know what settings to change.

 

[GLeeM]

but I can't seem to use the second gpu on my 690 for HCC. Anyone know what settings to change.

You need a cc_config.xml file with "<use_all_gpus>1</use_all_gpus>" inside.

Google search "cc_config.xml use all gpu".

Then look into app_config.xml to run more than one WU at a time.

Make sure to use text editor and not word processor
Make sure of path where file belongs
Make sure to restart BOINC (exception ... see post #364 above)
If you need help just ask

 

[ljtatej]

Thanks. Got it going.

 

[GLeeM]

Thanks. Got it going.

Curious if you are running more than one WU at a time and how long per WU?

 

[ljtatej]

I'm only running one per core and it averages about 6min give or take for each WU