sieving speeds for Intels

[jasong]

I've started a new thread in the hopes of convincing people with Intels that it's worthwhile to sieve. I have a dual-core 2.8GHz Pentium-D and I can sieve at about 11.9 kp/s per core, which sounds slow, but I'm knocking down candidates at about 2m30s a candidate, which is significantly faster than even the slowest test.

I would implore people with Intels to consider sieving a small, 5G range, to help the project. I feel it would speed up the overall project, and make the primes per test ratio a lot denser. We'll also probably find the actual primes a lot faster if people would favor sieving over PRPing for a while.

Just my opinion.

[paulunderwood]

Interesting. Would a 64-bit sieve with a 64-bit operating system on one core a 64-bit Pentium "D" out perform a sieve running on a 32-bit Athlon?

[Citrix]

May be some one could set up an exchange, where someone with an athlon would donate time sieving and inexchange you, with a p4, would PRP numbers for him. Credit for all PRP work goes to the athlon owner and factors belong to the P4 owner.

What do you think?

[rogue]

Originally Posted by paulunderwood

Interesting. Would a 64-bit sieve with a 64-bit operating system on one core a 64-bit Pentium "D" out perform a sieve running on a 32-bit Athlon?

I'm tempted to get one of the new Mac Pros (in addition to my G5, not to replace it) so I could find out. Unfortunately, I doubt I could convince my wife to spend the money on one.

[axn]

Originally Posted by rogue

I'm tempted to get one of the new Mac Pros (in addition to my G5, not to replace it) so I could find out. Unfortunately, I doubt I could convince my wife to spend the money on one.

The apple site shows that Mac Pro uses Intel Xeon "Woodcrest". If so they are based on the Core 2 architecture rather than Net Burst.

Net Burst (P4) is pretty bad in sieving. Core 2 generally rocks -- being comparable to Athlon 64s clock-for-clock (maybe even better). Of course, they're even better at PRP than P4!!

[rogue]

Originally Posted by axn1

The apple site shows that Mac Pro uses Intel Xeon "Woodcrest". If so they are based on the Core 2 architecture rather than Net Burst.

Net Burst (P4) is pretty bad in sieving. Core 2 generally rocks -- being comparable to Athlon 64s clock-for-clock (maybe even better). Of course, they're even better at PRP than P4!!

I'm more or less interested in the 64-bit capabilities of it. That it can be both great at sieving and PRPing is nice, but I can do PRP tests for other projects (PIES and GFN) on my G5. What I haven't seen yet is a 64-bit Intel ASM manual.

[axn]

Originally Posted by rogue

I'm more or less interested in the 64-bit capabilities of it. That it can be both great at sieving and PRPing is nice, but I can do PRP tests for other projects (PIES and GFN) on my G5. What I haven't seen yet is a 64-bit Intel ASM manual.

Aah! My bad. Are you looking for instruction timings/latency etc., or just a reference? If it is the latter, AMD64 reference can be obtained from AMD's site.

[rogue]

Originally Posted by axn1

Aah! My bad. Are you looking for instruction timings/latency etc., or just a reference? If it is the latter, AMD64 reference can be obtained from AMD's site.

Timings and latencies are nice, but the reference is more useful. The NASM manual is a start, but it doesn't have 64-bit instructions. Do Intel and AMD use the same 64-bit instructions?

[axn]

Originally Posted by rogue

Timings and latencies are nice, but the reference is more useful. The NASM manual is a start, but it doesn't have 64-bit instructions. Do Intel and AMD use the same 64-bit instructions?

More or less. Here are a bunch of differences

[geoff]

Originally Posted by jasong

I've started a new thread in the hopes of convincing people with Intels that it's worthwhile to sieve. I have a dual-core 2.8GHz Pentium-D and I can sieve at about 11.9 kp/s per core, which sounds slow, but I'm knocking down candidates at about 2m30s a candidate, which is significantly faster than even the slowest test.

So a 1GHz P3 is as fast as a 2.8GHz P4 :-(

Finding a single prime eliminates roughly 25,000 candidates at the moment. It would take your P4 something like a month to eliminate this many candidates by sieving. I can only guess, but I think if you spent a solid month PRP testing the smallest candidates you would be unlucky not to find a prime.

There are other ways to look at it.

[jasong]

Originally Posted by geoff

So a 1GHz P3 is as fast as a 2.8GHz P4 :-(

Finding a single prime eliminates roughly 25,000 candidates at the moment. It would take your P4 something like a month to eliminate this many candidates by sieving. I can only guess, but I think if you spent a solid month PRP testing the smallest candidates you would be unlucky not to find a prime.

There are other ways to look at it.

You may be right.

A while back, I used an odds figuring program to calculate the odds of a random number, of the form k*2^n-1, being prime, and came up with a simple formula which I felt Riesel Sievers could use to figure out where each computer would best help the project, statistically. Unfortunately, there isn't a base-5 version of the program, so I've been forced to guess at what would be best.

Btw, the program is called pr_prob, and it's on a yahoo group which has to do with prime numbers. Unfortunately, I forget the precise name of the group.