GIMPS now has a Fermat prime number of members

When George Woltman was asked whether this caused a conflict of interest, he replied,"Who the hell are you? Get away from me, and quit asking me stupid questions."

Not sure if others will be as entertained reading that as I was writing it.

Anyway, quick question. I'm on a Q6600, which was one of Intel's first quad-cores, basically 2 Core2Duos glued together. Since each Core2Duo has shared cache, I'm running one number on each sub-processor(or whatever you want to call it, 2 numbers, 1 to each Core2Duo on my Core2Quad). Anyway, my question is should I use both cores on the number, or just assign one core to each number? Either way, they'll only be running Prime95, but I'm not sure if the other core will aid the first one, or just "fight" with it.

I'll await your answers with bated breath. (I tried baited breath, but worms don't taste very good)

Quote:

---

Originally Posted by jasong

Anyway, my question is should I use both cores on the number, or just assign one core to each number? Either way, they'll only be running Prime95, but I'm not sure if the other core will aid the first one, or just "fight" with it.

---

Using more than one core on a number will speed up an individual test but you lose some efficiency so in the long run it is better to run one number on each core (so more than 1 number being tested at the same time).

Quote:

---

Originally Posted by Digital Parasite

Using more than one core on a number will speed up an individual test but you lose some efficiency so in the long run it is better to run one number on each core (so more than 1 number being tested at the same time).

---

Yes, but the cache is shared, plus I like to jump from project to project. So, on the one hand, I don't think we'd get much improved efficiency by running 4 numbers, and on the other I won't go stir-crazy waiting for numbers to finish so that I can go to another project. ;)

Anyway, it looks like I'm running two first-pass tests that are very close to each other. Since they're both close to each other and also have an exponent less than 40M, my guess is that someone abandoned both these numbers and I was lucky enough to accidentally do some "garbage" collection.

Oh, and they seem to be running in lock-step with a predicted end time of July 5th 8am, which I assume is expressed in Greenwich time, which means 3am here.

Quote:

---

Originally Posted by jasong

Yes, but the cache is shared, plus I like to jump from project to project. So, on the one hand, I don't think we'd get much improved efficiency by running 4 numbers, and on the other I won't go stir-crazy waiting for numbers to finish so that I can go to another project. ;)

---

With the Q9660 like you said it is two dual-cores sandwiched together. The cache is only shared between two of the cores in each of the packages. So if you are running 2 tests on a quad-core and they are running on different packages then you aren't sharing caches between the tests.

Even still, I think the shared cache is large enough to fit 2 copies of a test inside. Whether you are using 2 threads for 1 test, or 2 threads for 2 tests, you still are going to be using more cache than a single thread for 1 number since each thread has to run its own code for doing the FFT. By doing multi-threaded you don't get everything for free.

If you are really concerned, stop everything else on your computer, and try running in the 2 different modes and see which is faster. The multi-threaded will finish a test faster, but using separate tests might finish the 2nd number faster than doing 1 and 2 more quickly since the separate tests would be going at the same time.