I just finished running my own little test on 2 computers. One computer has a 550ti card the other a 650ti card. Times for the 2 cards are not that different.
Running 1 Perseus WU it took about 4 ½ hours. But on both computers the gpu was only running at 70% – 75%. If this could be bumped up to 90+% could probably shave off another hour. I don’t overclock and don’t know why on 2 computers the gpu only ran at 70 – 75 percent.
Running 2 Perseus WU’s it took about 8 hours. It was a little faster than 1 WU because now the gpu is running at about 95%.
Running 3 Perseus WU’s it took about 10 ½ hours but was slowing down the CPU, so the Wu’s that ran on the CPU were taking a lot longer. It really bogged down the cpu.
It might be different for your computer and gpu. I only did approximations. I'm going to stick with running 2 wu's. Hope this helps.
My GPU has been crunching a Perseus Arm Survey work unit for just over 24 hours. Is this expected? Seems a bit long.
Indeed, too long. Your current valid results list has one job listed that took only about 2 h!
One other thing that I noticed in the output of your tasks is this :
Quote:
[18:39:05][31212][INFO ] Application startup - thank you for supporting Einstein@Home! parse_init_data_file: no end tag
18:39:05 (31212): Can't parse init data file - running in standalone mode
This is not normal, and it's not good! I would recommend to update the version of BOINC you are using (the file that the app complains about is written by the BOINC client software).
Hope this helps
HBE
Downloaded latest BOINC from boinc website. It would not run: missing shared libraries. Looked/searched for the libraries under the FC17 distro but was not able to resolve. The boinc webpage talks about 3 required packages and they are installed. I have run into this before where BOINC is compiled against a specific distribution with specific flags which results in a binary not compatible with your distribution.
Downloaded latest BOINC from boinc website. It would not run: missing shared libraries. Looked/searched for the libraries under the FC17 distro but was not able to resolve. The boinc webpage talks about 3 required packages and they are installed. I have run into this before where BOINC is compiled against a specific distribution with specific flags which results in a binary not compatible with your distribution.
Ughhhhh....
I feel your pain!
There's a lot to like about Linux, but that kind of problem isn't one of them. ;-)
I compile a recent BOINC version from source on a Linux host that does not have X libraries or other extra libraries installed. This way I get a BOINC client with minimal dependencies that can run on a barebones Linux system or Linux system that does not have all the extra libraries. I connect to the Linux hosts from boinccmd, BOINC Manager on a Windows host, or my one Linux host that does have all the libraries installed for running BOINC Manager.
IOW's you can't have it both ways. You're implying if you ran them one at a time the 7970's would magically start cranking them out in 3000 seconds.
I can tell you for a FACT, they won't.
i'm not implying, i've performed some test.
single WU on 7970 runs for 5600 sec.
2 WUs in parallel completed in 7800 seconds.
4 WUs in 13200 sec.
and so on.
Actually, they do not run in parallel to each other. Your GPU can only run one task at a time, so what happens with running 10 is this:
1. The CPU will translate a part of the data from task 1 into kernels that the GPU understands, the CPU will transport this data to the GPU, the GPU will run it. The CPU transports the result data back to main memory, translates it back into something the human can work with, save it to disk.
2. The CPU will translate a part of the data from task 2 into kernels that the GPU understands... etc.
Rinse repeat for all 10 tasks.
When all 10 tasks have had part done, the CPU will translate a further part of the data from task 1 into kernels that the GPU understands and transport it to the GPU. And then task 2, and task 3, and task4, etc.
[18:39:05][31212][INFO ] Application startup - thank you for supporting Einstein@Home! parse_init_data_file: no end tag
18:39:05 (31212): Can't parse init data file - running in standalone mode
This is not normal, and it's not good! I would recommend to update the version of BOINC you are using (the file that the app complains about is written by the BOINC client software).
As I posted on Seti:
The APP_INIT_DATA.xml file keeps track of the elapsed time that a task took. Seeing the the BOINC Wiki, it looks to be a problem with the science application, or at least with the BOINC API. Not saying it still can't be a problem with BOINC, but problems with it were fixed in 6.13/7.0.3.
I just finished running my
)
I just finished running my own little test on 2 computers. One computer has a 550ti card the other a 650ti card. Times for the 2 cards are not that different.
Running 1 Perseus WU it took about 4 ½ hours. But on both computers the gpu was only running at 70% – 75%. If this could be bumped up to 90+% could probably shave off another hour. I don’t overclock and don’t know why on 2 computers the gpu only ran at 70 – 75 percent.
Running 2 Perseus WU’s it took about 8 hours. It was a little faster than 1 WU because now the gpu is running at about 95%.
Running 3 Perseus WU’s it took about 10 ½ hours but was slowing down the CPU, so the Wu’s that ran on the CPU were taking a lot longer. It really bogged down the cpu.
It might be different for your computer and gpu. I only did approximations. I'm going to stick with running 2 wu's. Hope this helps.
RE: RE: My GPU has been
)
Downloaded latest BOINC from boinc website. It would not run: missing shared libraries. Looked/searched for the libraries under the FC17 distro but was not able to resolve. The boinc webpage talks about 3 required packages and they are installed. I have run into this before where BOINC is compiled against a specific distribution with specific flags which results in a binary not compatible with your distribution.
RE: Downloaded latest
)
Ughhhhh....
I feel your pain!
There's a lot to like about Linux, but that kind of problem isn't one of them. ;-)
Funny. Even the BOINC website
)
Funny. Even the BOINC website advises using your own distro's version of BOINC.
RE: So what? That still
)
again, 10 tasks runned in parallel took 30600 seconds on my machine. or 3060 seconds per a single task.
RE: RE: So what? That
)
And again, so what that they ran in parallel?
They still took 30600 seconds to run in aggregate. So if you ran the run SEQUENTIALLY it would have taken 10 times as long (more or less).
IOW's you can't have it both ways. You're implying if you ran them one at a time the 7970's would magically start cranking them out in 3000 seconds.
I can tell you for a FACT, they won't.
I compile a recent BOINC
)
I compile a recent BOINC version from source on a Linux host that does not have X libraries or other extra libraries installed. This way I get a BOINC client with minimal dependencies that can run on a barebones Linux system or Linux system that does not have all the extra libraries. I connect to the Linux hosts from boinccmd, BOINC Manager on a Windows host, or my one Linux host that does have all the libraries installed for running BOINC Manager.
RE: IOW's you can't have it
)
i'm not implying, i've performed some test.
single WU on 7970 runs for 5600 sec.
2 WUs in parallel completed in 7800 seconds.
4 WUs in 13200 sec.
and so on.
RE: again, 10 tasks runned
)
Actually, they do not run in parallel to each other. Your GPU can only run one task at a time, so what happens with running 10 is this:
1. The CPU will translate a part of the data from task 1 into kernels that the GPU understands, the CPU will transport this data to the GPU, the GPU will run it. The CPU transports the result data back to main memory, translates it back into something the human can work with, save it to disk.
2. The CPU will translate a part of the data from task 2 into kernels that the GPU understands... etc.
Rinse repeat for all 10 tasks.
When all 10 tasks have had part done, the CPU will translate a further part of the data from task 1 into kernels that the GPU understands and transport it to the GPU. And then task 2, and task 3, and task4, etc.
All until all tasks are done.
RE: RE: [18:39:05][31212
)
As I posted on Seti:
The APP_INIT_DATA.xml file keeps track of the elapsed time that a task took. Seeing the the BOINC Wiki, it looks to be a problem with the science application, or at least with the BOINC API. Not saying it still can't be a problem with BOINC, but problems with it were fixed in 6.13/7.0.3.