Singeing a GPU with LL work?

[Rodrigo]

Originally Posted by ckdo

The manual results form will accept that line. It means: It was found that M859433 is prime using a 512k FFT.

Thanks, ckdo, that's good to know. I should be pretty well set now to start CUDALucasing in earnest.

Can I delete the "c" and "t" files as each exponent is finished, or should I keep them? Not that they take up THAT much room, but I can see how they could start cluttering up the directory listings.

Rodrigo

[LaurV]

Originally Posted by ckdo

Not "might" but "will".

Will not, if there was any communication with the server. Or, say like that, If you job was a double check, it will not, for sure. If your job was a first time check, then it might, but after you reported the result from manual test (by GPU), the one on CPU (P95) might start and be reported as double check, depends on your settings (if "whatever makes more sense" is selected). If you select "first time tests" or "world records", then P95 will communicate with the server and see that the exponent is reported, and it will automatically assign a new one.

That is the same as you go to your "assignment" page on mersenne.org (you must be logged in for the link to work) and "unreserve" an exponent. You mark the checkbox for the exponent you want to "unreserve", then mark the checkbox under the table, then click "unreserve" button. In this case you don't have to do anything else, like editing the "worktodo" file or so. P95 is quite clever and it will get a new assignment first time when it check up with the server, and it will cancel the old assignment. Things go the same way if you have reported the results for an ("first time test"-option selected) exponent that is not started in P95. Next time when P95 checks with the server it will get a new exponent.

Just to make sure, you can browse to the "assignment" table after you reported you GPU result and see if the exponent is still on the list. It will take a bit to update the table, it is not really real time. But if the exponent still on the list, then check the two boxes and unreserve it. Yo do not have to do any "woktodo" editing tricks.

Things should not be differently if the exponent is already started on P95, in fact it can not continue and report the result as "double check", because for double checks the initial LL step is shifted with a couple of bits. For a double check it has to start from the beginning, otherwise P95 will deal with the same data, and it will replicate any possible error that eventually appeared at the first test.

[ckdo]

Originally Posted by LaurV

Will not, if there was any communication with the server. Or, say like that, If you job was a double check, it will not, for sure. [...]

This is all gibberish. Prime95 will start processing the exponent unless the GPU result has been submitted and Prime95 has been trying to update the server-side ETA for the assignment. Further restrictions apply.
So yes, by all means, it should be removed from the worktodo file when you move it into the GPU's queue.

[kjaget]

Originally Posted by Rodrigo

Thanks, ckdo, that's good to know. I should be pretty well set now to start CUDALucasing in earnest.

Can I delete the "c" and "t" files as each exponent is finished, or should I keep them? Not that they take up THAT much room, but I can see how they could start cluttering up the directory listings.

Yes, you can delete the old checkpoint files. All they are there for is to save your progress along the way so when you restart you don't have to start from scratch. Once you've finished the exponent there's no use in keeping them.

I thought CUDALucas was supposed to remove them that automatically, but if not you can do it yourself.

[Rodrigo]

Originally Posted by kjaget

Yes, you can delete the old checkpoint files. All they are there for is to save your progress along the way so when you restart you don't have to start from scratch. Once you've finished the exponent there's no use in keeping them.

I thought CUDALucas was supposed to remove them that automatically, but if not you can do it yourself.

kjaget,

You're right, it did remove the files automatically. I did one test LL on a small exponent and let it run to the end, and the next time I checked, the checkpoint files for that had been deleted. I still have checkpoint files for a larger exponent that I stopped "in flight", but then that's exactly what you'd want. So it all makes sense.

Rodrigo

[Christenson]

Originally Posted by ckdo

The manual results form will accept that line. It means: It was found that M859433 is prime using a 512k FFT.

Hey, assuming my exponent is *NOT* prime, do I get a residue to report ?

[Rodrigo]

Aha, I knew something was up with the title of this thread!

At first I was scratching my head, thinking that when I started the thread maybe I'd mistyped the title "Singing up a GPU," but tonight I got a new notice of more posts and now it says that the title is "Singeing." Hmm... so I went back to the previous e-mail notifications, and sure enough -- I typed it right the first time!



Very funny, I love it!

Rodrigo

[KingKurly]

Originally Posted by Rodrigo

Aha, I knew something was up with the title of this thread!

At first I was scratching my head, thinking that when I started the thread maybe I'd mistyped the title "Singing up a GPU," but tonight I got a new notice of more posts and now it says that the title is "Singeing." Hmm... so I went back to the previous e-mail notifications, and sure enough -- I typed it right the first time!



Very funny, I love it!

Rodrigo

Thank you; I thought I'd lost it.

[Rodrigo]

OK, I've been LLing a "real" exponent now for a day and a half, and I'd like to compare notes.

The GT 430 is doing 10K iterations every 10:12 or so. At this rate it will finish the 54M exponent in just under 39 days. I'm not familiar with the formula for calculating these things, but I'm estimating that this would put its daily throughput at something less than 3 GHz-days/day.

At this rate, if I'm calculating this correctly, the GT 430's contribution amouints to slightly less than a notional third core of a Pentium Dual Core E6700. Does this gibe with what others are getting from their GT 430?


Two more notes:

1. I also have LLs running on the two CPU cores in Prime95. The per-iteration time appears to have been completely unaffected on one core, and to have gone up by 0.01 seconds (from 0.054 to 0.055) on the other one.
2. The CPU temperatures reported by CoreTemp prior to starting CUDALucas were 64 and 62 C. With CUDALucas running, now they're at 66 and 64. This doesn't seem to be much of a difference -- with the GPU running all the time, I'd braced myself for a much bigger jump in temperature inside the case. And unless the GPU fan is REALLY quiet, I don't hear anything different coming out of the box. What am I missing?

Rodrigo

[Rodrigo]

Originally Posted by Rodrigo

I also have LLs running on the two CPU cores in Prime95. The per-iteration time appears to have been completely unaffected on one core, and to have gone up by 0.01 seconds (from 0.054 to 0.055) on the other one.

Typo alert -- that difference is 0.001 seconds, of course.

Rodrigo

[S34960zz]

Originally Posted by Rodrigo

The CPU temperatures reported by CoreTemp prior to starting CUDALucas were 64 and 62 C. With CUDALucas running, now they're at 66 and 64. This doesn't seem to be much of a difference -- with the GPU running all the time, I'd braced myself for a much bigger jump in temperature inside the case. And unless the GPU fan is REALLY quiet, I don't hear anything different coming out of the box. What am I missing?Rodrigo

Take a look at GPU Observer (Windows 7 desktop Gadget) for GPU temperature / performance info (I use this occasionally) [in German, auf Deutsch]:
http://blog.orbmu2k.de/sidebar-gadge...adget#more-203

or CPUID's Hardware Monitor (I have this open at all times on my machine):
http://www.cpuid.com/softwares/hwmonitor.html
http://www.cpuid.com/downloads/hwmonitor/1.18-setup.exe

Other options exist, too, but I've been happy with these. Thanks to whomever here posted about them earlier.