Would sieving speed up with 64-bits

[pixl97]

I'll ask this question 'cause I've got a hold of my first opteron, and the fact that intel is coming out with a compatible x86-64 arch soon, if microsoft releases an OS anytime soon we'll be seeing 64-bitness everywhere.

What kind of speed up would be expected by 64 bit version of the sieve software (or factoring or sbclient even) if any? Will the larger amount of registers help also?

Heres my current output with proth sieve .42
Opteron 242 (1600MHz) 508kps(@277700) (linux 64 bit) 1024k
Athlon XP 2100 (1737MHz) 512kps (@277700) (linux 32 bit) 256k
Athlon XP 2500(1837MHz)520kps(@277700) (windows 2k) 512k

Thanks, peter

Oh, yea, I'll also add that I get slightly faster speeds using the cmov_sse2 client on the opteron versus the plain cmov version, that did seem odd to me for I've heard the opteron has a rather slow SSE2 implimentation compared to intel.

[mklasson]

A good kind of speedup would be expected for the sieve, and yes, more registers help.

As for the opteron's sse2 support, I think it's only "slow" compared to intel's because of the frequency difference. That is, it's not slow compared to doing the work with other instructions on an opteron processor.

[Death]

well, how about Macintosh client?

G5 is first 64-bit processor, and it blazing fast in d.NET project. So 64 bit can do a lot for 17oB!

[pixl97]

Originally posted by Death
well, how about Macintosh client?

It may be fast, but I dont see enough users for it to be feaseable for 17oB, for x86 on the other hand, a good portion of the processors next year will be 64-bit compatable.

[Death]

1. Verification of Riemann's Hypothesis http://www.zetagrid.net

Currently participating: 10,102 computers:
6,952 x86 on Windows
2,472 x86 on Linux
1,546 ppc on AIX
380 ia64 on Linux
81 ppc on Mac OS X
66 sparc on SunOS
11 alpha on Linux
10 ppc on Linux
7 sparc on Linux
5 s390 on Linux
3 amd64 on Linux
2 PA_RISC2.0 on HP-UX
1 x86 on SunOS
1 i386 on FreeBSD

well this is because of java client with complicated installation mechanism and lack of possibility to run it as service.

2. http://distributed.net OGR-24 blocks
X86 Win32 2000-07-13 2004-06-07 382,129,780
X86 Linux 2000-07-13 2004-06-07 68,804,992
Sparc Solaris 2000-07-13 2004-06-07 12,912,920
PowerPC Rhapsody 2000-07-14 2004-06-07 9,489,263
X86 FreeBSD 2000-07-13 2004-06-07 9,337,394
PowerPC Mac OS 2000-07-13 2004-03-16 6,227,522
PA-RISC HP-UX 2000-07-13 2004-03-28 2,581,870
PowerPC AIX 2000-07-14 2004-06-07 1,427,127

OGR-25 blocks
X86 Win32 2000-08-02 2004-06-07 24,606,231,515
X86 Linux 2000-08-02 2004-06-07 2,233,444,492
X86 FreeBSD 2000-08-02 2004-06-07 335,676,773
PowerPC Rhapsody 2000-08-02 2004-06-07 257,543,204
Sparc Solaris 2000-08-02 2004-06-07 213,945,351
PowerPC Mac OS 2000-08-02 2004-05-15 161,216,914
PowerPC AIX 2000-08-03 2004-06-07 69,247,652

RC5-72 blocks
X86 Win32 2002-12-03 2004-06-07 888,690,947
X86 Linux 2002-12-03 2004-06-07 149,926,790
PowerPC Rhapsody 2002-12-03 2004-06-07 123,164,629
X86 FreeBSD 2002-12-03 2004-06-07 32,117,504
Sparc Solaris 2002-12-03 2004-06-07 9,155,330
PowerPC Linux 2002-12-03 2004-06-07 1,932,801
X86 OpenBSD 2002-12-03 2004-06-07 1,609,179
X86 OS/2 2002-12-03 2004-06-07 1,474,725
X86 Solaris 2002-12-03 2004-06-07 1,220,709
PowerPC AIX 2002-12-30 2004-06-07 971,983

Well, how about
RC5-72 / Participant Listing by Overall Rank: 1 to 100

1 Diplomacy has gone ? 18-Dec-2002 17-Feb-2004 427 16,606,023
2 [PL#3042] The1kHill SubTeam 15-Jan-2003 07-Jun-2004 510 14,379,352
3 rc5-72@bruteforcesweden.nu 03-Dec-2002 07-Jun-2004 553 11,961,920
4 RRM (SNO3) 05-Jan-2003 07-Jun-2004 520 9,924,017
5 MU in Tokyo 26-Dec-2002 07-Jun-2004 530 9,542,515
6 rc5@apple.com 08-Dec-2002 07-Jun-2004 548 6,988,449

and after finishing OGR-24 I can switch nearly 20 GHz

[Keroberts1]

those 20 GHZ would do wonders if dedicated to sieving or factoring. I know you'd probably love to find a prime, i know i would too but I would also love to see this project finished in my life time. That is why I dedicate all of my resource to sieving for the time being. Once we have reached a level of save 5-6 hundred sieving will definatly no longer be necesary and then i as well will eb able to switch resources (besides those that are too slow to deal with PRP tests to PRPing. Of course right now sieving is the quickest way to eliminate tests and help the project progress.

[Death]

well, I CAN switch, but i need a client with all d.net vital features

no user attention, automatical installation and startup, automatic job assignment and reporting, etc......