World Community Grid: Launch of Graphics Processing Unit (GPU) capability for the Hel [Archive]

View Full Version : World Community Grid: Launch of Graphics Processing Unit (GPU) capability for the Hel

RSS

10-10-12, 04:40 PM

We are excited to announce the release of GPU processing capability for the Help Conquer Cancer project!

More... (http://www.worldcommunitygrid.org/about_us/viewNewsArticle.do?articleId=210)

Mr. Hankey

10-10-12, 05:39 PM

We are excited to announce the release of GPU processing capability for the Help Conquer Cancer project!

More... (http://www.worldcommunitygrid.org/about_us/viewNewsArticle.do?articleId=210)

anyone run this application... does it pay shit like the rest of WCG?

Fire$torm

10-10-12, 06:30 PM

anyone run this application... does it pay shit like the rest of WCG?

I haven't but I'm sure it does.

Mad Matt

10-10-12, 07:45 PM

anyone run this application... does it pay shit like the rest of WCG?

I haven't but I'm sure it does.

=)) =)) =))

That was mind-reading... :D

nanoprobe

10-10-12, 08:53 PM

anyone run this application... does it pay shit like the rest of WCG?

Actually it runs very well. They've been beta testing since January. What it pays can't be measured in points IMHO

Now it's with a little sadness that I let y'all know I will be setting my GPUs to run this and will be leaving POEM for a while. I lost my mom to cancer and I'm a cancer survivor myself so this project has always been close to my heart. Once everything gets settled I will use some of my CPUs to help with challenges where I can. I will be around the forums and I will be watching when SUSA takes over the #1 position in POEM.

Crunch on.

Mad Matt

10-10-12, 09:25 PM

Actually it runs very well. They've been beta testing since January. What it pays can't be measured in points IMHO

Now it's with a little sadness that I let y'all know I will be setting my GPUs to run this and will be leaving POEM for a while. I lost my mom to cancer and I'm a cancer survivor myself so this project has always been close to my heart. Once everything gets settled I will use some of my CPUs to help with challenges where I can. I will be around the forums and I will be watching when SUSA takes over the #1 position in POEM.

Crunch on.

I hear you, we all lost or almost lost friends and relatives to cancer and crunched against this. If it just helps making you feel better and one day may help somebody, it's the best way to go.

We will miss you on the POEM team, but I am hoping we will see you back there soon. %%-

10-10-12, 11:00 PM

Actually it runs very well. They've been beta testing since January. What it pays can't be measured in points IMHO

Now it's with a little sadness that I let y'all know I will be setting my GPUs to run this and will be leaving POEM for a while. I lost my mom to cancer and I'm a cancer survivor myself so this project has always been close to my heart. Once everything gets settled I will use some of my CPUs to help with challenges where I can. I will be around the forums and I will be watching when SUSA takes over the #1 position in POEM.

Crunch on.

I'm so sorry nanoprobe. I lost my father to cancer on the 9th of September so I can feel for you. I wish you and your family the best.

zombie67

10-10-12, 11:22 PM

Okay, gonna need an app_info.xml so that my machines get only GPU tasks for HCC, while my CPUs get only other specific projects. The app_info.xml language baffles me, even after all these years. PLEASE?

Rattledagger

10-11-12, 04:11 AM

anyone run this application... does it pay shit like the rest of WCG?
Well, let's see, on my i7-920 with an Ati-HD5850, a HCC-task on CPU apparently takes 1.91 hours = 115 minutes. On the GPU, a HCC-task takes roughly 5 minutes.

My guess is WCG will give the same per task regardless of how it's crunched, meaning GPU should give roughly 23x more cobblestones/day per core.

If takes into account the cpu has 8 cores, this means cpu can do 100.2 tasks/day if runs only HCC on cpu. With the GPU, cpu-production drops to 87.7 tasks/day but with an additional 144 from GPU the total is 231.7 tasks/day and this again should equal 2.32x more cobblestones/day.

If assumes the same speedup across the board, this means HCC instead of using 736 days to finish will only use 318 days to finish.

Mike029

10-11-12, 09:26 AM

Actually it runs very well. They've been beta testing since January. What it pays can't be measured in points IMHO

Now it's with a little sadness that I let y'all know I will be setting my GPUs to run this and will be leaving POEM for a while. I lost my mom to cancer and I'm a cancer survivor myself so this project has always been close to my heart. Once everything gets settled I will use some of my CPUs to help with challenges where I can. I will be around the forums and I will be watching when SUSA takes over the #1 position in POEM.

Crunch on.

No problem. We are glad your here and please continue to use the forums. Also if you are up to it could you post your results on your different gpus?

John P. Myers

10-11-12, 10:28 AM

How will statistics be updated for graphics card participation?
World Community Grid will award run time based on the elapsed time from when a work unit is started and when it finishes running. Credit and result counts will be awarded in the same manner as CPU participation.

Will there be a badge for graphics card participation?
No. Your contributions utilizing your graphics card will contribute to a specific research project and count towards earning a badge for that project.

Sounds like it's paying the same as CPU WUs?

Mr. Hankey

10-11-12, 01:25 PM

Sounds like it's paying the same as CPU WUs?

wow that sounds even worse.... If you have the same WU running on CPU and it take 24 hrs, but the GPU does it in 12 you are only going to get credit based on elapsed run time, or half the credit for the same WU? So while the project gets twice as much work done the crunchers get the same payout. So sad, this project could get so much more done if they were more equitable with the credit. I guess they have all the data they need, or more data than they can use at any given time?

This project sure makes some strange decisions. The project site deviation from the standard BOINC layout makes info way harder to find as well. Many people run this project off an on, but if the credits were good (or even comparable) people would park on this project.

Mike029

10-11-12, 01:53 PM

This project sure makes some strange decisions. The project site deviation from the standard BOINC layout makes info way harder to find as well. Many people run this project off an on, but if the credits were good (or even comparable) people would park on this project.

Agreed.

nanoprobe

10-11-12, 02:39 PM

No problem. We are glad your here and please continue to use the forums. Also if you are up to it could you post your results on your different gpus?
You'd like to see my WCG GPU results?

nanoprobe

10-11-12, 02:45 PM

wow that sounds even worse.... If you have the same WU running on CPU and it take 24 hrs, but the GPU does it in 12 you are only going to get credit based on elapsed run time, or half the credit for the same WU? So while the project gets twice as much work done the crunchers get the same payout. So sad, this project could get so much more done if they were more equitable with the credit. I guess they have all the data they need, or more data than they can use at any given time?

This project sure makes some strange decisions. The project site deviation from the standard BOINC layout makes info way harder to find as well. Many people run this project off an on, but if the credits were good (or even comparable) people would park on this project.

You'll get the same amount of credit whether it's a CPU result or a GPU result and since the GPU is much faster you're daily results will be much higher. My first 12 hours produced 73,000+ boinc credits which is equal to about 421,000 WCG points. Why the BOINC credits are only 1:7 I have no idea but since points is not my motivation it doesn't really matter to me.

Rattledagger

10-11-12, 02:49 PM

wow that sounds even worse.... If you have the same WU running on CPU and it take 24 hrs, but the GPU does it in 12 you are only going to get credit based on elapsed run time, or half the credit for the same WU?
No, it's the same credit per wu, regardless of CPU or GPU, meaning going by your example the GPU will get 2x credit/day compared to the CPU.

In reality the speedup is much higher, for my Ati-HD5850 running on i7-920 the speed-up is around 22x, and this means GPU gets roughly 22x more credit/day compared to a CPU-core.

Mr. Hankey

10-11-12, 04:14 PM

No, it's the same credit per wu, regardless of CPU or GPU, meaning going by your example the GPU will get 2x credit/day compared to the CPU.

In reality the speedup is much higher, for my Ati-HD5850 running on i7-920 the speed-up is around 22x, and this means GPU gets roughly 22x more credit/day compared to a CPU-core.

Well that is much better. Thank you for clarifying.

10-11-12, 05:06 PM

Rattledagger

10-11-12, 05:23 PM

Is this normal? Wus on my 5870s get to 99.415% in 3m20s. Getting the rest of the way to 100% takes an additional 1m10s. Seems a little odd, but I've never run these wus before. Gpu usage is at 0% for the last minute ten.
Yes, this is normal. The start and the end of each wu uses only the cpu with the end being longest and it taking 1+ minute is normal.

As for app_info.xml, JanusBK using nvidia posted a file over at the WCG-forums, see http://www.worldcommunitygrid.org/forums/wcg/viewpostinthread?post=395470
Obviously the app_info.xml won't be usable for Amd-cards, but it atleast shows it can be done, and can give a boost in production.

10-11-12, 05:42 PM

Thanks! I'll look at the app_info tonight.

John P. Myers

10-11-12, 06:05 PM

No, it's the same credit per wu, regardless of CPU or GPU, meaning going by your example the GPU will get 2x credit/day compared to the CPU.

In reality the speedup is much higher, for my Ati-HD5850 running on i7-920 the speed-up is around 22x, and this means GPU gets roughly 22x more credit/day compared to a CPU-core.

Sounds great for getting credits, but for those chasing badges (based on time, not credits) there's no improvement right? More work and more electricity to gain badges at the same rate?

zombie67

10-11-12, 06:13 PM

Badges are based on CPU time, which is small for GPU tasks. So you will get some small help on badges, but nothing compared to CPU tasks.

This will be an interesting chart to watch:

http://boincstats.com/en/charts/15/project/perDay/credit/new/chart.png

grathomp

10-11-12, 07:57 PM

For those of you looking for an AMD app_info.xml file I was able to modify the CUDA file that Rattledagger found on the WCG forums to work with my 7970.

Here it is:

<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>
<nbytes>3499008.000000</nbytes>
<max_nbytes>0.000000</max_nbytes>
<status>1</status>
<executable/>
<signature_required/>
<file_signature>
</file_signature>
</file_info>
<file_info>
<name>hcckernel.cl.6.56</name>
<nbytes>63144.000000</nbytes>
<max_nbytes>0.000000</max_nbytes>
<status>1</status>
<signature_required/>
<file_signature>
792397d3ec34c91119da45507eb1f8e732452b9d40c1d4adc3 13cb4d0b6db671
0382de7f034a2dea10b75e2ab2e012b1ce9c3effa738a48f50 daa615fa3e6b82
3f4e28d9986968d347552f0298571f3755a601f8834ab9f554 d457ea2f0a71c4
843ef8c8b5e4b04633591321627f9a403f12f380bcb5702593 f2e42ca0ea5514
.
</file_signature>
</file_info>
<file_info>
<name>hcc1_image01_6.56.tga</name>
<nbytes>262188.000000</nbytes>
<max_nbytes>0.000000</max_nbytes>
<status>1</status>
<signature_required/>
<file_signature>
7117813faaff1bceb99d172432ff67417397b809172fc09c47 a0db414106c68d
de9bf88ed64f518e78b5e7ab79637335c9b166003bd3f32a29 915101c1758746
d1db7a90bc201c35b1c65fcfb2a9b091253b1979bc6f9db748 feb5a6e9d5c804
cd875399b2ee4016fc210f5b76f15aa0902477ce1c0a945fb5 6c178a0d65de4c
.
</file_signature>
</file_info>
<file_info>
<name>hcc1_image02_6.56.tga</name>
<nbytes>32812.000000</nbytes>
<max_nbytes>0.000000</max_nbytes>
<status>1</status>
<signature_required/>
<file_signature>
0d3314db1591dcf067aa6551e3582f5274ff10287fb384c119 34a98245fd1631
be5249db0ae6d0138b191618ae9502e853a681f879e7fa3c94 fec8effdd6685b
a325ec7538ec8abb1fd719629c691bf721da031a87cf69d20b 0f501a38e16a92
dbdd7442d90b64f30deff58bc46c5096fb017f4d6410eb8e9b facb09b2c6a8ae
.
</file_signature>
</file_info>
<file_info>
<name>hcc1_image03_6.56.tga</name>
<nbytes>131116.000000</nbytes>
<max_nbytes>0.000000</max_nbytes>
<status>1</status>
<signature_required/>
<file_signature>
58915b00c9fae53d18ef72f179c6ff1222c134b0c0e696c3bb fddb91c650b26f
c5856311eea2fe2128d9dbdd558b47963332cf9f9ae910689a 133b03e7113715
39babe2ce7303b73e9bfc69b1f01cc4199c7dfab13322d5b40 adcb8ee167832d
81df23835ba4e12430a50aeb534b8bfed9a82ce54fb0a40181 04f1adb36778d9
.
</file_signature>
</file_info>
<file_info>
<name>hcc1_image04_6.56.tga</name>
<nbytes>262188.000000</nbytes>
<max_nbytes>0.000000</max_nbytes>
<status>1</status>
<signature_required/>
<file_signature>
59882941f8d7b844b3ac20301fba70c98ab9cedd5a41855d24 b0721b6b6275e6
f1de708d8b43ae7f0198f21924f528c1c710b6d36776d881fe e6ebb539cf2f4e
4c4f63997be3eb6157b9a3a5922bdbdebb26fc1cfeae4d1398 0682800c697ffa
54a7a64cc14694ec457132320f0c2bcc9c3fed675f71cbe388 d9ec55332c1bae
.
</file_signature>
</file_info>
<app_version>
<app_name>hcc1</app_name>
<version_num>656</version_num>
<platform>windows_intelx86</platform>
<avg_ncpus>0.500000</avg_ncpus>
<max_ncpus>1.000000</max_ncpus>
<flops>56530729222.498611</flops>
<plan_class>ati_hcc1</plan_class>
<api_version>7.1.0</api_version>
<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>
<file_ref>
<file_name>hcc1_image01_6.56.tga</file_name>
<open_name>boinc_wcg_skin_w-logo.tga</open_name>
</file_ref>
<file_ref>
<file_name>hcc1_image02_6.56.tga</file_name>
<open_name>pbibmablk.tga</open_name>
</file_ref>
<file_ref>
<file_name>hcc1_image03_6.56.tga</file_name>
<open_name>UHN_stacked2.tga</open_name>
</file_ref>
<file_ref>
<file_name>hcc1_image04_6.56.tga</file_name>
<open_name>HCC_LOGO.tga</open_name>
</file_ref>
<coproc>
<type>ATI</type>
<count>0.125</count>
</coproc>
<gpu_ram>266338304.000000</gpu_ram>
</app_version>
</app_info>

nanoprobe

10-11-12, 08:08 PM

You can run 4 tasks at a time? What's the run time and GPU load?

grathomp

10-11-12, 08:19 PM

I'm still trying to figure out how many instances of the application is optimal for my system, but I thought I would post a working app_info for people to adjust their <avg_ncpus> <max_ncpus> and GPU <count> according to their specifications.

When I run 8 tasks my GPU load maxes out around 95% but will dip if enough tasks are nearly complete and switch to CPU only. Tasks are completing after about 5 minutes.

nanoprobe

10-11-12, 09:02 PM

grathomp

10-12-12, 03:30 AM

I attached my app_info.xml as app_info.txt

This time it's for 4 workunits per AMD GPU each using one CPU core. Hope this helps.

nanoprobe

10-12-12, 09:19 AM

I attached my app_info.xml as app_info.txt

This time it's for 4 workunits per AMD GPU each using one CPU core. Hope this helps.
Still not working for me. Get this from message log.

10/12/2012 9:14:39 AM | World Community Grid | Found app_info.xml; using anonymous platform
10/12/2012 9:17:19 AM | World Community Grid | update requested by user
10/12/2012 9:17:23 AM | World Community Grid | Sending scheduler request: Requested by user.
10/12/2012 9:17:23 AM | World Community Grid | Not requesting tasks: don't need
10/12/2012 9:17:25 AM | World Community Grid | Scheduler request completed

Duke of Buckingham

10-12-12, 09:56 AM

stop all tasks that you have on Boinc or just canceled them without any tasks Boinc will start downloadind.

nanoprobe

10-12-12, 10:05 AM

stop all tasks that you have on Boinc or just canceled them without any tasks Boinc will start downloadind.

No tasks running. I finished them all before using the app_info file.

grathomp

10-12-12, 11:26 AM

Are you getting workunits for your AMD card before you restart BOINC with the app_info.xml file in your project folder?

nanoprobe

10-12-12, 02:23 PM

Are you getting workunits for your AMD card before you restart BOINC with the app_info.xml file in your project folder?

Yes

Duke of Buckingham

10-12-12, 02:35 PM

Out of my league nanoprobe.

nanoprobe

10-12-12, 03:03 PM

Out of my league nanoprobe.

I've run app_info files with 3 other projects without a hitch but this one has me stumped. Had a friend of mine who wrote a working one for his Nvidia cards write one he said should work on ATI but it doesn't work on mine. :confused:

Duke of Buckingham

10-12-12, 03:16 PM

I would try to run WCG without the app_info and if I could do it I would presume that the problem was in the app else I would restart the Project or I would remove and reattach the project maybe that could solve it. Attention to the Boinc version you are using maybe an update could help you.

My knowledge about this is most generic. There are a lot of members that for sure know much more than me.

STE\/E

10-12-12, 05:26 PM

The app worked on 1 of my 7970 Box's, had to get work first so I had the right files then shut down BOINC & add the app file. Running just 6 Wu's, 3 on each 7970, they run from 80 sec's up to 120 sec's each ... good, bad, indifferent, I don't know, only going to run a few then put the Box back on POEM until after the Challenge then may run more WCG on them ...

grathomp

10-12-12, 06:22 PM

Hey, nanoprobe. Try this one. I also included the Nvidea version of this app_info. This one works for me as well and was posted by glennaxl here (http://www.dslreports.com/forum/r27614814-).

zombie67

10-12-12, 07:55 PM

Just FYI everyone:

Upside of app_info.xml: You can run multiple tasks per GPU

Upside of no app_info.xml: You can run HCC on your GPU, and a different sub-project on your CPU.

If you aren't planning on running any other sub-project other than HCC, go with the app_info. But if you want to crunch other sub-projects, then no app_info.xml.

Rattledagger

10-12-12, 10:03 PM

If you aren't planning on running any other sub-project other than HCC, go with the app_info. But if you want to crunch other sub-projects, then no app_info.xml.
Haven't tested it, but it should be possible to also run other sub-projects even with app_info.xml, by copying the relevant parts about other applications from your client_state.xml and adding it to your app_info.xml

zombie67

10-12-12, 10:10 PM

Haven't tested it, but it should be possible to also run other sub-projects even with app_info.xml, by copying the relevant parts about other applications from your client_state.xml and adding it to your app_info.xml

Absolutely true. If you can write or copy one. I haven't seen any yet. Just GPU-only.

nanoprobe

10-12-12, 11:51 PM

Hey, nanoprobe. Try this one. I also included the Nvidea version of this app_info. This one works for me as well and was posted by glennaxl here (http://www.dslreports.com/forum/r27614814-).

BINGO! We have a winner. With this app_info I just quadrupled the out put of my 7970. Thanks a ton, I owe you one.1147

BTW glennaxl is a team mate. Really sharp on Nvidia stuff.

zombie67

10-12-12, 11:55 PM

<sarcasm>Guys, slow down on the GPU stuff. Credits are great and all, but crunching via GPUs may complete the work before I get my blue badge.</sarcasm>

STE\/E

10-14-12, 06:09 AM

I guess WCG doesn't want to ruin it's Reputation for being stingy with the Credits, seems like I get less & less for the Wu's. Started out with the BETA Wu's getting around 59 Credits, then when the regular Wu's started flowing I was getting about 39 Per Wu, now I see I'm only getting about 29-31 Per Wu. wonder where it'll fall to before it stops, could be something to do with Credit New ???

Rattledagger

10-14-12, 09:56 AM

Duke of Buckingham

10-14-12, 10:06 AM

Creditnew? !/*%#$&£€?

http://m5.paperblog.com/i/11/111882/how-twitter-is-destroying-your-mind-L-02AjIT.jpeg

Mr. Hankey

10-14-12, 02:51 PM

cant seem to keep the WUs flowing very well either.. they trickle outs in batches of 12 or 13..

zombie67

10-14-12, 05:17 PM

My guess I that this is a learning experience for the project. They've never had to feed such volumes of tasks before. I run it along side other GPU projects, so I don't run out of work.

Rattledagger

10-14-12, 08:36 PM

cant seem to keep the WUs flowing very well either.. they trickle outs in batches of 12 or 13..
WCG is using fairly low limits on tasks per scheduler-request, task in progress and starting daily quota, but atleast personally I've had no problems getting enough work. But seeing I'm on the low end of the scale with roughly 5 minutes/wu and only one GPU, anyone with multi-GPU and using 1.5 minute/wu is more likely to run into problems.

The low limit on task in progress is to a large extent due to buggy v6-BOINC-clients. Since WCG delivers a customized BOINC-client, to automatically attack to WCG, more users than normal is still running v6.10.58. Since WCG does various internal tests before releasing an upgraded client it takes much longer time than normal before a new release is out. While many users does run the WCG-version, it's of course possible to run the normal BOINC-client and many users is doing this.

WCG is the 2nd. largest DC-project based on #active users and the largest based on #active computers. SETI@home has more active users but fewer active computers. So while GPU-crunching is new to WCG, handling so many tasks per day isn't really anything new to WCG. While looking on the short-term stats, it seems GPU gave an increase from 600 k/day to 1.1 M/day. But, looking more long-term, start of September the production was around 900 k/day, and going even longer back around April the production peaked at 1.76 M/day. The low output in October is due to the two sub-projects with lowest average run-time/wu was ending, and due to a file-system problem most other wu's was increased in size. The problem with file-system have been fixed and they also added more memory to their servers.