sr2sieve client

[KriZp]

After switching to Fedora 7 x86_64 running sr2sieve 1.5.19, I get around 730 kp/s on my athlon64 and 780 kp/s per core on my Opteron, compared to 500 kp/s and 520 kp/s respectively with jjsieveSSE2 in windows XP 32 bit.

[VJS]

720 to 935 switching to linux???

That's crazy!!!

I'm currently getting 915 on each of my 4 processors q6600 at 3.3Ghz.

That would mean that switching to linux would yeild almost 1200 kps???

Humm... might try a dual boot.

[KriZp]

Originally Posted by VJS

720 to 935 switching to linux???

Switching from 32 bit windows to 64 bit linux that is. It would be interesting to know the performance on 64 bit windows.

version 1.6.4 on 64 bit linux is slightly faster than 1.5.19, 800 vs 780 kp/s on my Opteron 165@2.7GHz and 760 vs 730 kp/s on the 2.5GHz Athlon64.
Good job Geoff!

[mgpower0]

Originally Posted by VJS

720 to 935 switching to linux???

That's crazy!!!

I'm currently getting 915 on each of my 4 processors q6600 at 3.3Ghz.

That would mean that switching to linux would yeild almost 1200 kps???

Humm... might try a dual boot.

Should give it a go thats running llrnet on the second core too. Running two sieve clients it runs at around the 950 mark

Put it this way, 64bit linux will handle anything I can throw at it, at the moment I am running psp sieve (at a respectable 902kp/s) I have an llrnet client running for riesel, am running muon1 in a windows 32bit VM and also have 4 EON clients running and i am still surfing and posting on this rig.

[Lennart]

C2Q 2.4Ghz v1.5.21 sr2sieve Win64XP 1239-1245 000 p/sec/core
C2Q 2.4Ghz v1.6.4 sr2sieve Win64XP 1044 000 p/sec/core
C2Q 2.4Ghz v1.6.6 sr2sieve Win64XP 1208 000 p/sec/core

==================================================
C2Q 2.4Ghz v1.6.9sr2sieve Win64XP 1475 000 p/sec/core
==============================================



==================================================
C2Q 2.4Ghz v1.6.10sr2sieve Win64XP 1475 000 p/sec/core
==============================================


Lennart

[geoff]

For those testing the speed of the x86-64 version, speed might start to reduce as the sieve depth nears 2^51 (about 2250T). This is caused by the increasing number of corrections required as the SSE2 floating point operations approach their limits of precision. Beyond the depth of 2250T the program will switch over to using the higher precision x87 FPU and speeds will start to increase again.

On some x86-64 machines it might even be worthwhile turning off the SSE2 optimisations before the sieve depth reaches 2250T. This can be tested by using the --no-sse2 switch and checking whether p/sec speed increases. On my Core 2 Duo the break-even point seems to be about 1500T-1800T.

This doesn't apply to the x86 version, it always uses the x87 FPU and its speed should steadily increase as the sieve depth increases.

[umccullough]

For the hell of it, I threw Ubuntu 7.10 x86-64 on my Core2Duo E6750 and I'm getting ~1460 kp/sec/core on SoB/PSP combined sieve - niiiice.

Looks like Linux64 is about to become more popular on my machines.

[em99010pepe]

sr2sieve-1.7.5 is out...

[VJS]

Any speed advantages???

I havn't seen a change report from version to version but I'm always happy with the result.

[geoff]

Originally Posted by VJS

Any speed advantages???

For this project, older pentium 4's (Prescott and earlier) should gain up to 15% from version 1.7.6 compared to version 1.6.18, but results are a bit mixed on other machines. Any comparisons with version 1.6.18, especially on 32-bit machines, would be welcome.

(Beware that 1.7.x versions currently use elapsed time instead of CPU time, so for a direct comparison run version 1.6.18 it is necessary to use the -e switch or else to make sure that nothing else is running on the same CPU).

[VJS]

Thanks Geoff,

On a dual or quad core CPU is there any reason to believe that specifiying affinities to particular CPU core for each instance is superior to simply running them with no affinities?
Or identifying half the cache for core2duo's for each instance etc...

Thanks, sometimes it's just easier to ask then try it. :-)