PrimeNet 5.0 Upgrade

[garo]

I think that is to not encourage ECM too much. It doesn't really contribute to the main aim of GIMPS - to find Mersenne primes.

[RMAC9.5]

Yesterday, one of my V4 PCs which does only TF work reported 3 completed tasks to the V5 server through the V4 interface. Here are the Work Results Details:
v4_computers 70081909 F-PM1 2009-07-03 07:27 61.4 879094071157651347121 5.1624
v4_computers 70049131 NF 2009-07-03 07:26 23.2 no factor to 2^70 3.3604
v4_computers 70130579 NF 2009-07-03 07:26 30.2 no factor to 2^70 3.3565

Note, two of these results were given TF credit and one was given P-1 credit by the V5 server. Is the P-1 credit intentional or is this a bug?

[Kevin]

Originally Posted by RMAC9.5

Yesterday, one of my V4 PCs which does only TF work reported 3 completed tasks to the V5 server through the V4 interface. Here are the Work Results Details:
v4_computers 70081909 F-PM1 2009-07-03 07:27 61.4 879094071157651347121 5.1624
v4_computers 70049131 NF 2009-07-03 07:26 23.2 no factor to 2^70 3.3604
v4_computers 70130579 NF 2009-07-03 07:26 30.2 no factor to 2^70 3.3565

Note, two of these results were given TF credit and one was given P-1 credit by the V5 server. Is the P-1 credit intentional or is this a bug?

IIRC, the client does nothing to indicate to the server what method was used to find a factor, so the server guesses based on the size of the exponent and factor.

[petrw1]

Originally Posted by RMAC9.5

Note, two of these results were given TF credit and one was given P-1 credit by the V5 server. Is the P-1 credit intentional or is this a bug?

I had this happen last fall too ... TF reported as P-1

[cheesehead]

I had factors credited as having been found by ECM even though I'd never performed ECM.

[lfm]

Originally Posted by RMAC9.5

Yesterday, one of my V4 PCs which does only TF work reported 3 completed tasks to the V5 server through the V4 interface. Here are the Work Results Details:
v4_computers 70081909 F-PM1 2009-07-03 07:27 61.4 879094071157651347121 5.1624
v4_computers 70049131 NF 2009-07-03 07:26 23.2 no factor to 2^70 3.3604
v4_computers 70130579 NF 2009-07-03 07:26 30.2 no factor to 2^70 3.3565

Note, two of these results were given TF credit and one was given P-1 credit by the V5 server. Is the P-1 credit intentional or is this a bug?

I'd say it means it is time to upgrade to v25.

[lfm]

Originally Posted by petrw1

My 3.4 Ghz PIV can complete a P-1 on 50M in just under 2.5 days for about 4.3 points. : about 1.75 PPD.

ECM-F (F24) just over 3 days for just over 3 points with 768M RAM: 1 PPD.

LL/DC is almost 2 PPD.

The Pentium 4 seems to have particularly good match for LL tests. A combination of the fine tuning of the FFTs software in p95 and the speed ratio between float and integer instructions make the P4 more than almost any other cpu faster for LLs. (Obviously not in an absolute sense but in the relative sense of speed of various work units types)

I'm not entirely certain about the ECM code but it indicates to me that ECM is not quite as heavily dependent on float speed as the LL tests. ECM probably has some extra integer or address computation that shows up on the P4 speeds.

The reverse is true for certain models of AMD (especially in 64 bit mode) where the float/integer speed ratio leans much more to the integers resulting in much better PPD for TFs on those machines.

[petrw1]

1. How does one be on these reports and yet have 0.000 points?

2. When you compute percentiles are these 0 point people included? The way I interpret percentiles is, if there are 1000 people on the list then each 10 people is 1 percentile. The top 10 are in the 100th percentile, the last 10 are in the 1st percentile. End even if someone in this bottom 10 with 0 points is declared in the 1st percentile then they should be counted in the total number for computing percentiles. So for example the LL list has 2745 names on it yet the user summary report shows my ranking as out of 2704.

[lfm]

Originally Posted by petrw1

1. How does one be on these reports and yet have 0.000 points?

2. When you compute percentiles are these 0 point people included? The way I interpret percentiles is, if there are 1000 people on the list then each 10 people is 1 percentile. The top 10 are in the 100th percentile, the last 10 are in the 1st percentile. End even if someone in this bottom 10 with 0 points is declared in the 1st percentile then they should be counted in the total number for computing percentiles. So for example the LL list has 2745 names on it yet the user summary report shows my ranking as out of 2704.

I think there are some results that get 0.0001 to 0.0004 ghz-days that round of to zero with 3 digit fractions.

yes I expect they are counted.

[petrw1]

Originally Posted by lfm

I think there are some results that get 0.0001 to 0.0004 ghz-days that round of to zero with 3 digit fractions.

That is a good possibility in the TF list but there are no LL tests in the last 365 days near the 0.0004 ghz-days range. That would have to be an exponent under 151,000. Way back in 1997 when this formal project started they were nearing 1,000,000.

[petrw1]

Dec 24, 2008
Today's Numbers 
 Teams 121 
 Users 5791 
 CPUs 40240 
 TFLOP/s 38.349 
 GHz-Days 19174.259

Today (July 20, 2009) almost half a year later:
Today's Numbers 
 Teams 237 
 Users 24121 
 CPUs 132926 
 TFLOP/s 44.165 
 GHz-Days 22082.610

So over 6 months the CPU count has gone up 230%
Users count up 317%
while the TFLOPS/s rate has ONLY increased 15%.

Can anyone offer an explanation why these extra 92,000 CPUs appear to be adding very little to the thruput???

I'm not sure these can all be V4_Computers since I never saw a count much above 75,000 Computers in V4.