| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Community
Forum: Chat Room
Thread: What is the ideal work load? |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 11
|
|
| Author |
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Good evening,
I have an E8400 (45NM dual core, 3.0 GhZ) and an Nvidia 9800 GTX. I can crunch a standard unit in about 5 hours on each core. Reading up on GPU's, I realized that their throughput is about 180 GFLOP/sec, which is greater than the standard CPU 10-20 GFLOP/sec. To get the best bang for each unit of time donated, I signed up on GPUGrid as well and noticed that the CPU still handles two work units, whereas the over CPU load has fallen from 100% to 60% (B/C the GPU is now taking the work). However, the time to completion is no longer accurate as it gradually counts up and then counts down. Reminiscent of a moving sand dune... My question is this, what is the most optimal grid computing choice for my computer? SHould I run 1 GPUGrid work unit and 1 CPU based work unit? Or, should I disable GPUGrid and let the CPU do 2 CPU work units? Please note: The system will not run two GPU work units at a time (B/C the board is a single chip with multiple cores known as 'Pixel Shaders'. Any thoughts, ideas, or suggestions are appreciated.  |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
I think you should first consider what work is being done on each of the projects you are considering participating in and use that as a base for your decision. I am partial to WCG but what you crunch is really up to you.
----------------------------------------I may have misread what you wrote but I think you may not be understanding the relationship between GPU and CPU in the crunching sense. Even though GPUGrid utilizes your GPU, your GPU must interact with your CPU in order to do things like write to/ from memory/ swap file, hard drive, etc. and it appears in your case it is fairly disruptive as your CPU utilization has been reduced by 40% (seems high, maybe post in the GPUGrid forums to see if this is the expected overhead) . This means your CPU is no longer as effective for crunching WCG tasks as it use to be. Likely you will see your average task time get a little bit longer if you continue to crunch GPU. That being said you may still want to crunch both if the electricity bills and the heat don't convince you to only go with one project. In the last couple of days I have started crunching at GPUGrid myself so I can get a feel for what it is like, how much I need to really know about graphics cards, overclocking them, temps, etc. I am trying to get in shape for when there is a CUDA project on WCG so I can help other people who want to crunch with both CPU and GPU. No, I don't know of anything about any projects or when this might happen. I have a GTX 295 and the CPU overhead to keep it's 2 GPUs fed properly is only about 25% of 1 core. Because I have an i7 920 I have not really noticed all that much of a change but I may try to bump my overclock on the CPU to help make up some of the difference. Edit: I am not even going to talk about the measuring differences between processing potential. I will say that programming for a CPU has been around for a long time and is very flexible. Programing for crunching on GPUs on the other hand feels like all we have is a screw driver and everyone thinks they can build a house with it. Don't get me wrong, it is very good at what it does, but it just isn't quite ready for the big time yet (still wrapping duct tape around the handle of that screwdriver). That is another reason I am on GPUGrid ... continual beta, so when it comes time for WCG to host a CUDA project the technology should be a bit more mature.  [Edit 1 times, last edit by Former Member at Apr 7, 2009 9:17:08 AM] |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
Hi,
----------------------------------------the timekeeping in the 6.4 clients is highly inaccurate. The latest 6.6.x alpha clients have now in fact switched to showing Elapsed time (aka wallclock last I looked), so I'm running a special cooked one-off version that shows both CPU and Elapsed time in 2 columns. GPUGrid I understood now to only use very little CPU%, so you have to clarify what you mean with 60%. If u use the WCG default preset profile [Standard Settings], 60% average will be the CPU use. Snow Crash is though right. Ask at GPUgrid, for we know nothing here about CUDA crunching configurations. edit: inserted preset profile name!
WCG
----------------------------------------Please help to make the Forums an enjoyable experience for All! [Edit 1 times, last edit by Sekerob at Apr 7, 2009 9:23:57 AM] |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
I bet Sekerob has it exactly right ... check you device profile for the artificail 60%.
|
||
|
|
mwgiii
Advanced Cruncher United States Joined: Aug 17, 2006 Post Count: 131 Status: Offline Project Badges: 
|
Try the beta version of Boinc. I was unable to get 2 workunits plus a gpugrid workunit running consistantly until I upgraded my Boinc client to a beta version.
----------------------------------------  |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
mwgiii - Edgeonsite is running as many WUs as possible on that PC. The question was centered on the "optimal" mix of project types, namely WCG and GPUGrid.
|
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Yeah, I wasn't very elaborate and that is my fault.
So, I run WCG and because I have a dual core, it can run one project per core. I removed all artificial limits when I first installed it, so the moment my computer boots up, its full bore on the crunching and CPU usage on both cores is at 100%. I notice that my computer operates just fine as I read or do a little web browsing. However, when I enable GPUGrid, my computer effectively works on 1 WCG project and 1 GPUGrid project. In this scenario, the second core is effectively not being used and the usage of both cores drops to 60% (I watch this in the task manager). So, the GPU project has effectively offloaded all of the work from one of the cores. Steve is correct, one core must remain dedicated to work with the GPU. In effect, one core is now the GPU's secretary. Writing and filing stuff and fetching new data as needed. Due to limits of system architecture, its simply not possible to allow this core to work on its own project. So, what I am describing is a tradeoff. Am I getting more done with 1 core and the GPU or am I better of using 2 cores and no GPU? GPUGrid on NVIDIA chipsets is done with their CUDA code, which appears to be the best on the block (I admit, I haven't looked at anything else though). Check out some of the white papers on it. http://www.gpugrid.net/science.php http://www.nvidia.com/object/cuda_what_is.html Check out the three pdf articles on the above site. As for overclocking: On my CPU, I do not do it. On my GPU, it is a 9800 GTX, which is overclocked at the factory and so it is already at its peak performance. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
What beta version are you running and how did you get 2 CPU projects and 1 GPU project to run on your computer at the same time? If I can do that, then that would be great!
|
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
First I think you should check your preferences to make sure in the basic options section: Use no more than: 100 % of processor time
----------------------------------------and in the processor section: On multiprocessors, at most use = 16, On multiprocessors, use: 100% I would suggest make any changes necessary on the WCG website. After you do that you should open your local preferences from the Advanced menu and click the Clear button just to make sure your local settings are not taking priority over what you just set up on the web. Now go to the Projects tab and click the Update button. I am running the 6.6.20 beta client with just WCG and GPUGrid and am not having any major problems. Sekeob has posted about sometimes it displays military time and at others it displays local which I don't really mind. I have heard some bad things about hung WUs from SETI (not that I am particulary interested, I figure if the aliens wanted us to know about them, they would tell us) so do yourself a favor and stay away until they get themselves straightened out. You don't really need to have a dedicated CPU core to crunch GPU, it will *borrow* what it needs but that really should be a fairly small percent of your total CPU time. The only other thing is that GPUGrid will only download 1 WU per CPU core so you will never get more than two WUs from them but that is OK. I would suggest that you set your cache in the Work Unit Cache Settings section of the device profile (same place on the web where you made the changes above) to equal the number of days it takes to process 2 GPUGrid WUs. This will reduce unecessary communication with the GPUGrid server. That's it. No special magic, it just works. For the record I am crunching 8 WCG and 2 GPU at the same time but that is the max defined by the hardware I have, you should have no problem running 2 WCG and 1 GPUGrid. {edit: added 100% cpu note 1st para.} [Edit 1 times, last edit by Former Member at Apr 8, 2009 11:24:11 AM] |
||
|
|
mwgiii
Advanced Cruncher United States Joined: Aug 17, 2006 Post Count: 131 Status: Offline Project Badges:
|
What beta version are you running and how did you get 2 CPU projects and 1 GPU project to run on your computer at the same time? If I can do that, then that would be great! 6.6.15 Until I tried that one, I was having the same problem as you. |
||
|
|
|