| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Active Research
Forum: OpenPandemics - COVID-19 Project
Thread: OpenPandemics - GPU Stress Test |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 781
|
|
| Author |
|
|
ca05065
Senior Cruncher Joined: Dec 4, 2007 Post Count: 328 Status: Offline Project Badges: 
|
@erich56
After I completed the final GPU work unit in my queue, the work request changed from: Requesting new tasks for CPU and NVIDIA GPU [World Community Grid] [sched_op] CPU work request: 3798671.04 seconds; 0.00 devices [World Community Grid] [sched_op] NVIDIA GPU work request: 341697.74 seconds; 0.00 devices to: Requesting new tasks for CPU [World Community Grid] [sched_op] CPU work request: 3660133.20 seconds; 0.00 devices [World Community Grid] [sched_op] NVIDIA GPU work request: 0.00 seconds; 0.00 devices I have not been able to find any explanation. The graphics card section of the device settings are still set to request GPU work units. |
||
|
|
jeffwy
Advanced Cruncher Taiwan Joined: Nov 17, 2004 Post Count: 77 Status: Offline Project Badges: 
|
I did not see the posts about a GPU stress test so did not join, but what I did notice was that the GPU WUs were not really utilizing a lot of my GPU, the fans on the GPU weren't even spinning during the days the project was running. But it would cause my CPU fans to constantly change speeds where other projects including both GPU and non-GPU projects typically would not do.
----------------------------------------In addition, was that it? A week of GPU units and we're finished for now?  |
||
|
|
DennyInDurham
Cruncher USA Joined: Aug 4, 2020 Post Count: 23 Status: Offline Project Badges: 
|
I did not see the posts about a GPU stress test so did not join, but what I did notice was that the GPU WUs were not really utilizing a lot of my GPU, the fans on the GPU weren't even spinning during the days the project was running. But it would cause my CPU fans to constantly change speeds where other projects including both GPU and non-GPU projects typically would not do. In addition, was that it? A week of GPU units and we're finished for now? About 30,000 GPU WUs about as fast as possible. It found a hiccup or two and melted the file system on the server when they packaged some of the results, along with some client SSDs.  GPU WUs are back to trickling out now. |
||
|
|
erich56
Senior Cruncher Austria Joined: Feb 24, 2007 Post Count: 295 Status: Offline Project Badges: 
|
...along with some client SSDs. I saw this coming; an in order to avoid it, I changed to Ramdisk early enough  |
||
|
|
jeffwy
Advanced Cruncher Taiwan Joined: Nov 17, 2004 Post Count: 77 Status: Offline Project Badges:
|
I am certainly hoping that you are being sarcastic about SSDs dying and GPUs overheating...
----------------------------------------I store all my WUs on a HDD not SDD anyway. lol. |
||
|
|
squid
Advanced Cruncher Germany Joined: May 15, 2020 Post Count: 56 Status: Offline Project Badges: 
|
|
||
|
|
Crystal Pellet
Veteran Cruncher Joined: May 21, 2008 Post Count: 1324 Status: Offline Project Badges: 
|
About 30,000 GPU WUs about as fast as possible. Not 30,000 WU's, but 30,000 batches. Don't know how may WU's are in 1 batch. |
||
|
|
goben_2003
Advanced Cruncher Joined: Jun 16, 2006 Post Count: 146 Status: Offline Project Badges:
|
OK, that makes more sense but a GPU core is not equivalent to a CPU core and you are only using 70% of 24 cores. There must be a compromise position but a 24x markup in your example is excessive. Using that basis a GTX1660 running a single WU would have about a 1400x markup which is ridiculous. Sure, then maybe 1 hr * 24 core * (70/100) = ~17 would be fine? I am not sure I agree that 24 is excessive, but I do think it should be more than 2(It uses a cpu core too). I would like to note some things: In my testing, there was more total science done(jobs/hr) when running just the iGPU without CPU units. So I ran just iGPU on the machines without discrete graphics. My time credit earned per day went downhill, but the amount of science done went way up. I ran it this way anyways since I consider the science to be more important than time credit. I still do like time credit though. Now for discrete graphics though, maybe counting all 1920 cores would be excessive(using the nvidia rtx 2060 as an example). Maybe per warp group like poppinfresh suggested would be fine? That would still be a very large number. So in this example 1920 / 32 * 1 hr * % load? One major problem is that this is all much more complicated to implement in practice. One problem is that I do not think load% is sent. It also gets more tricky with how many WUs are run at a time. The max that are set to run at a time is sent to the server, I do not think the number actually running at a time is. Also, while my Nvidia card does have the warpSize(32) reported, it does not appear to report the number of cores(1920) or the number of groups of 32 there are(60).  |
||
|
|
DennyInDurham
Cruncher USA Joined: Aug 4, 2020 Post Count: 23 Status: Offline Project Badges:
|
About 30,000 GPU WUs about as fast as possible. Not 30,000 WU's, but 30,000 batches. Don't know how may WU's are in 1 batch. Yes, you're correct. |
||
|
|
jlrobins58@gmail.com
Cruncher Joined: Jan 2, 2021 Post Count: 5 Status: Offline |
Kind of miss the GPU WUs. Other than it blocked an African Rainfall Project unit from running.
I am new and just a casual supplier of CPU/GPU power. But the GPU test increased my throughput scoring to over a million a day!?! That is up from a few hundred per day before and since that. My laptop did run noticibly hotter, though. I have both an Intel HD630 GPU and an NVIDIA GeForce GTX 1050 and it appeared that both were grinding through WUs. |
||
|
|
|