| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Completed Research
Forum: Help Conquer Cancer
Thread: HCC GPU APP HAS LAUNCHED. |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 363
|
|
| Author |
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
When they come up with more work, the best way to get it done (in addition to serializing the work units) will be to engineer multiple work units to run on a card but not using the CPU at the same time. If left up to each user, the only way to do that is to throw a lot of CPU cores at it and hope to saturate the GPU. Then, they seem to settle down to a steady-state wherein they are not all running on the CPU at once. But that is hardly practical in most cases, especially since the app_info precludes other projects, unless specifically allowed. Maybe they can address that in some subsequent enhancement? Why do I get the feeling that the next-wave of GPUs from AMD and nVidia will be the best of the GPUs as we know today plus an integrated-8-core-CPU? Well, keep those present problems coming in and we could get our hands on that next-wave of GPU hardware sooner.  ; |
||
|
|
Jim1348
Veteran Cruncher USA Joined: Jul 13, 2009 Post Count: 1066 Status: Offline Project Badges: 
|
Yes, new hardware solves all problems, especially around Christmas.
|
||
|
|
mmstick
Senior Cruncher Joined: Aug 19, 2010 Post Count: 151 Status: Offline Project Badges:
|
When they come up with more work, the best way to get it done (in addition to serializing the work units) will be to engineer multiple work units to run on a card but not using the CPU at the same time. If left up to each user, the only way to do that is to throw a lot of CPU cores at it and hope to saturate the GPU. Then, they seem to settle down to a steady-state wherein they are not all running on the CPU at once. But that is hardly practical in most cases, especially since the app_info precludes other projects, unless specifically allowed. Maybe they can address that in some subsequent enhancement? This would be the perfect method of going about it, although I would go a step farther and have the single work unit use all cores. However, this would require a rewrite of the application, but the software would go something like this: Work unit contains many images > All cores but one are sentenced to crunching the initial phase thread > After completing the first phase the image is sent to the GPU worker thread to be fed to the GPU, and a new image is sent to the initial phase thread. After the GPU worker has completed a work unit it is sent to a final phase thread. As it is now, I do use an app_info to run 16 work units at the same time on my FX-8120 so I can achieve a WU return rate of one every 22 seconds. I have found benefit in increasing work units to double the number of actual cores. |
||
|
|
Amr Adam
Advanced Cruncher Egypt Joined: Aug 13, 2012 Post Count: 74 Status: Offline Project Badges: 
|
Then with the GPU enabled for the HCC project, it would be more efficient for the over-all WCG production that people with only CPUs to pull out from HCC and redirect their effort to the other projects, right ?
|
||
|
|
Jim1348
Veteran Cruncher USA Joined: Jul 13, 2009 Post Count: 1066 Status: Offline Project Badges:
|
Then with the GPU enabled for the HCC project, it would be more efficient for the over-all WCG production that people with only CPUs to pull out from HCC and redirect their effort to the other projects, right ? I think that is unarguably correct. I think the CPU version is kept going mainly for the benefit of people who are hunting for badges (based on total time, not points), or those who just prefer HCC over other projects but don't have a GPU to give to it. But the other projects may never have GPU versions, and would love to have the CPU help, so I would give it to them in that case and let the GPUs handle HCC. Maybe in the future new projects will be developed with GPUs in mind. |
||
|
|
mmstick
Senior Cruncher Joined: Aug 19, 2010 Post Count: 151 Status: Offline Project Badges:
|
When they come up with more work, the best way to get it done (in addition to serializing the work units) will be to engineer multiple work units to run on a card but not using the CPU at the same time. If left up to each user, the only way to do that is to throw a lot of CPU cores at it and hope to saturate the GPU. Then, they seem to settle down to a steady-state wherein they are not all running on the CPU at once. But that is hardly practical in most cases, especially since the app_info precludes other projects, unless specifically allowed. Maybe they can address that in some subsequent enhancement? Why do I get the feeling that the next-wave of GPUs from AMD and nVidia will be the best of the GPUs as we know today plus an integrated-8-core-CPU? Well, keep those present problems coming in and we could get our hands on that next-wave of GPU hardware sooner. ; That is, essentially, what AMD is aiming for with HSA. A GPU that is also a CPU, it has CPU components integrated inside the GPU, like an APU but in reverse. As for NVIDIA, I don't know if they even have plans for some HSA like architecture. It won't be the next wave though, HSA is set for 2014, so HD 9000 series at earliest. |
||
|
|
Jim1348
Veteran Cruncher USA Joined: Jul 13, 2009 Post Count: 1066 Status: Offline Project Badges:
|
Nvidia is planning to integrate an ARM processor onto their GPU chip, which should show up in Maxwell, their next consumer product after Kepler.
http://www.xbitlabs.com/news/cpu/display/2012...icroprocessor_Report.html |
||
|
|
robertmiles
Senior Cruncher US Joined: Apr 16, 2008 Post Count: 443 Status: Offline Project Badges:
|
Then with the GPU enabled for the HCC project, it would be more efficient for the over-all WCG production that people with only CPUs to pull out from HCC and redirect their effort to the other projects, right ? I think that is unarguably correct. I think the CPU version is kept going mainly for the benefit of people who are hunting for badges (based on total time, not points), or those who just prefer HCC over other projects but don't have a GPU to give to it. But the other projects may never have GPU versions, and would love to have the CPU help, so I would give it to them in that case and let the GPUs handle HCC. Maybe in the future new projects will be developed with GPUs in mind. I've tried giving my GPU mostly to HCC and my CPU partly to other WCG projects and partly to non-WCG BOINC projects. The scheduler chooses to give nearly all of the WCG time to the HCC GPU workunits. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Has anybody a ATI 7850 card running ? I have real problems get the card working with boinc
|
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Let's keep this problem discussion in 1 thread, Marco. Replies please here: http://www.worldcommunitygrid.org/forums/wcg/...322_lastpage,yes#lastpost
thx |
||
|
|
|