Trial Factor Bit Depth

[lavalamp]

I am sure this must have been discussed before with regards to OBD, but I could not find it.

I have just made a post in which I extended the table of trial factor depths found on The Math page of mersenne.org all the way up to billion digit numbers.

Here's the post:
http://www.mersenneforum.org/showpos...1&postcount=70

I have included the methodology used to attain the numbers and I would appreciate any comments on it's validity.

If the numbers do turn out to be correct, then it looks like the OBD candidates need to be trial factored all the way to 87 bits, and currently there are 10 (soon to be 13) exponents at 81 bits.

In terms of levels, that's high enough to take the project to level 21.01, which at the minimum would require 22 candidates at 87 bits.

[ET_]

87 bits on one OBD exponent would require at least 4 months of 24/7 activity on a GXT275. I guess it's premature to reach that bit-depth with actual methods...

By the time GIMPS extends 6 times its bounds, I think we'd better check fornew factors than for bit-depth.

Luigi

[CRGreathouse]

Originally Posted by ET_

By the time GIMPS extends 6 times its bounds, I think we'd better check fornew factors than for bit-depth.

That's how I feel. Anyone crunched this comparison?

[lavalamp]

Originally Posted by ET_

87 bits on one OBD exponent would require at least 4 months of 24/7 activity on a GXT275.

I dunno, it might be nice to have 1 candidate taken all the way, call it a figurehead. 4 months is quite a while, but it's better than the 8 years and 5 months my overclocked i7 would take. A GTX 480 would take less time of course, but still a couple of months most likely.

Even so, it's good to know what the ultimate goal is. 87 seems tantalizingly close as the project has several candidates at 81 bits now. We all know of course that 87 bits requires another 63 times the work to reach, but that's not as much of a carrot as saying, "JUST 6 MORE!"

Originally Posted by CRGreathouse

Anyone crunched this comparison?

What do you mean by this?

[ET_]

Originally Posted by lavalamp

I dunno, it might be nice to have 1 candidate taken all the way, call it a figurehead. 4 months is quite a while, but it's better than the 8 years and 5 months my overclocked i7 would take. A GTX 480 would take less time of course, but still a couple of months most likely.

Maybe I'll give it a try as soon as winter approaches, we actually have 96/98 degrees in Rome. definitely too hot to run a GPU search 24/7.
But I don't recommend following my example!

The problem here is that, even using MORE_CLASSES, each class from 86 to 87 bits would last more than 3,5 hours, and interrupting the search to use the (otherwise unusable) computer would waste approximately 2 hours of calculations.

Originally Posted by lavalamp

Even so, it's good to know what the ultimate goal is. 87 seems tantalizingly close as the project has several candidates at 81 bits now. We all know of course that 87 bits requires another 63 times the work to reach, but that's not as much of a carrot as saying, "JUST 6 MORE!"

In 4 months' work we could bring 63 exponents to 80/81 bits...

Luigi

[mdettweiler]

Originally Posted by ET_

Maybe I'll give it a try as soon as winter approaches, we actually have 96/98 degrees in Rome. definitely too hot to run a GPU search 24/7.
But I don't recommend following my example!

The problem here is that, even using MORE_CLASSES, each class from 86 to 87 bits would last more than 3,5 hours, and interrupting the search to use the (otherwise unusable) computer would waste approximately 2 hours of calculations.

If the computer has integrated graphics available, you could always try running the monitor off that instead of the GPU. AFAIK, that will still leave the computer usable while the GPU is crunching.

[ET_]

Originally Posted by mdettweiler

If the computer has integrated graphics available, you could always try running the monitor off that instead of the GPU. AFAIK, that will still leave the computer usable while the GPU is crunching.

I have a motherboard with no integrated video card. It consumes less...

Luigi

[lavalamp]

Originally Posted by ET_

In 4 months' work we could bring 63 exponents to 80/81 bits...

Ah yes, but have you considered JUST 6 MORE!

[CRGreathouse]

Originally Posted by lavalamp

What do you mean by this?

Has anyone done a theoretical comparison of sieving to that depth vs. searching for small factors by other means?