LLRNET Times

[masser]

I was wondering what times LLRNETers are getting for the prp tests on various machines. I think exponents and processor type will be enough information... thanks in advance.

[tnerual]

here are results for my different computers.

centrino 1400

• 341552*5^193658-1 is not prime. Time: 6193.459 sec.
• 285728*5^512506-1 is not prime. Time: 51441.080 sec.

centrino duo 1666

• 285728*5^733582-1 is not prime. Time: 94179.191 sec.
• 204394*5^1995051-1 is not prime. Time: 782868.893 sec.

athlon 2500+ (+15% overclock)

• 341552*5^187538-1 is not prime. Time: 4854.982 sec.
• 285728*5^748894-1 is not prime. Time: 103441.989 sec.

sempron 3000 (+15% overclock)

• 285728*5^311806-1 is not prime. Time: 11799.970 sec.
• 285728*5^770626-1 is not prime. Time: 82267.028 sec.

amd x2 4200+ (+10% overclock)

• 1552*5^188006-1 is not prime. Time: 3043.420 sec.
• 285728*5^789274-1 is not prime. Time: 64976.979 sec.

duron 750

• 285728*5^800074-1 is not prime. Time : 333308.0 sec.

[masser]

tnerual,

Were those results from LLRNET or PRP?

Here are some of my LLRNET timings:

On an Intel Xeon 3.06 Ghz machine (with HT; but just running one instance):
256702*5^174731-1 ....Time: 1644.373 sec
193786*5^174735-1 ....Time: 1792.452 sec

[tnerual]

results are from llrnet (home network llrnet server)

timings are higher than yours because there is a fft size change around n=185000 (and my computers are slower )

on one of the two athlon 2500:

341552*5^184982-1 is not prime. Time: 3920.526 sec.
341552*5^185198-1 is not prime. Time: 4511.780 sec.

[rogue]

For those interested in using Phil Carmody's phrot program:

This is on a 2.5 GHz PowerPC G5.

48394*5^175883-1 is composite LLR64=50FE60D030A03630. (t=853.97s)
283076*5^310170-1 is composite LLR64=2D96FE2AA13464F9. (t=4081.90s)
139196*5^511667-1 is composite LLR64=D45D02044A5FE779. (t=11885.75s)
88444*5^801949-1 is composite LLR64=2D695E143B370B86. (t=37481.93s)
92936*5^1209620-1 is composite LLR64=EEF754705F6574CA. (t=89317.65s)
250216*5^1611717-1 is composite LLR64=2ACF5E1C96CA38CC. (t=129794.06s)
304004*5^1999948-1 is composite LLR64=B21B0888C052BEE5. (t=188094.46s)

[tnerual]

Originally Posted by rogue

For those interested in using Phil Carmody's phrot program:

times looks good but the problem with phil's phrot is that it's not "dumb friendly".

i tried everything to get a results.txt file as output, it was impossible for me.

another problem is that i have a very unstable electrical line. so the computers resets on a regular basis (2-3 times a day). with llrnet, they start automatically at the last saved checkpoint. it's not the case with phrot.

in fact, if phrot was able to communicate with a llrnet server and to handle checkpoint, i would have switched a long time ago.

[rogue]

Originally Posted by tnerual

times looks good but the problem with phil's phrot is that it's not "dumb friendly".

i tried everything to get a results.txt file as output, it was impossible for me.

another problem is that i have a very unstable electrical line. so the computers resets on a regular basis (2-3 times a day). with llrnet, they start automatically at the last saved checkpoint. it's not the case with phrot.

in fact, if phrot was able to communicate with a llrnet server and to handle checkpoint, i would have switched a long time ago.

There are three enhancements that I am working on for the next release of phrot:

1) Create phrot.ini file so that it can track its progress through an input file.
2) Add a -o option to write test results to results.out. PRPs will still be written to Phrot.prp.
3) Unroll the modular reduction code. This will of most use on PowerPC and other register rich architectures. It gives a 10% performance boost on PPC.

Of note, I did consider adding checkpoint logic, but I wanted to complete the current round of testing before I tackle that enhancement. It won't be in 0.51, but possibly in 0.52.

FYI, I will not release source code without Phil's blessing.