GIMPS HITS {insert number here} COMPUTERS

[Mystwalker]

Originally Posted by axn1

If we plot efficiency (Actual thruput/theoretical thruput) as a function of frequency, we'll see that efficiency drops as frequency goes higher. Does that answer your question? Linear scaling would mean that efficiency would remain constant irrespective of frequency.

One thing to add:

Assuming the computer is used for other things as well (what a crime! ), actual DC throughput will probably scale over-proportional in the beginning, as there is a basic load whose ratio gets smaller and smaller.
Of course, this effect dimishes with increasing frequency.

[axn]

Originally Posted by Mystwalker

Assuming the computer is used for other things as well (what a crime! ), actual DC throughput will probably scale over-proportional in the beginning, as there is a basic load whose ratio gets smaller and smaller.
Of course, this effect dimishes with increasing frequency.

Indeed. But I wonder if that is enough to compensate for the imperfect scaling at high frequencies (say going from 3.2 to 3.4 GHz?)

EDIT:- Of course, it is pointless to discuss this without first assuming how much the system is loaded. I was wondering about a lightly loaded system, say one used primarily for web surfing

[ninjabill]

The PC in question gets used for games, which use 100% of available processor. I was only considering throughput during times of no use.

Looking at the throughput of my p4 2.4 (533mhz fsb) and p4 3.0 (800mhz fsb) the throughput per Mhz seems almost identical

(0.079s per iteration on the 2.4, 0.063 on the 3.0)

regards

geoff

[jinydu]

Originally Posted by ninjabill

The PC in question gets used for games, which use 100% of available processor. I was only considering throughput during times of no use.

Looking at the throughput of my p4 2.4 (533mhz fsb) and p4 3.0 (800mhz fsb) the throughput per Mhz seems almost identical

(0.079s per iteration on the 2.4, 0.063 on the 3.0)

regards

geoff

That's not almost identical

0.063s is definitely a major improvement over 0.079s.

This is, of course, assuming that you're talking about similarly sized exponents on both computers...

[axn]

Originally Posted by jinydu

That's not almost identical

0.063s is definitely a major improvement over 0.079s.

This is, of course, assuming that you're talking about similarly sized exponents on both computers...

I suspect he is talking about relative thruput -- ie 2.4 GHz vs 3.0 GHz. Adjusted for clockspeed, they both show relatively the same thruput.

However, if the FSB for 3.0GHz was at 533 MHz, I think that the 3.0 would show less thruput.

[ninjabill]

Originally Posted by ninjabill

Looking at the throughput of my p4 2.4 (533mhz fsb) and p4 3.0 (800mhz fsb) the throughput per Mhz seems almost identical

(0.079s per iteration on the 2.4, 0.063 on the 3.0)

Hi there, The exponensts are similarly sized, I was talking about throughput per mhz, 0.079 * 2.4 / 3.0 = 0.0632 (which is farly close)

[petrw1]

Originally Posted by lpmurray

Great Internet Mersenne Prime Search
------- Aggregate CPU Statistics, P90 Units* -------

Last 7 Days Average Cumulative Today
from 08 Mar 2005 06h from 14 Mar 2005 06h
---------------------- ----------------------------------
Test Type CPU yr/day GFLOP/s CPU years CPU yr/day GFLOP/s
------------ ---------- ---------- ---------- ---------- ----------
Lucas-Lehmer 1405.406 16917.857 1339.370 1532.087 18442.799
Factoring 44.735 538.502 44.124 50.472 607.571
---------- ---------- ---------- ---------- ----------
TOTALS 1450.141 17456.359 1383.494 1582.559 19050.369



------- Internet CPU and Server Resources -------

Machines Applied on 55186 Accounts Server Synchronization 15 Feb 2005 03:47

TOTAL : 80030

28 months later:
7 day average has increased from ABOUT 1500 to ABOUT 2000 (33% increase).
Accounts decreased from 55186 to 44468 (20% decrease).
Machines decreased from 80030 to 67490 (15% decrease).

Our throughput is increasing even as our number of machines decreases. This is great news.

However per the post that asks if Moore's law will ever die: GIMPS is not keeping up with Moore's law. Even if we were to average throughput per number of machines we are still well under a 50% increase ... not the more than double Moore's law would suggest.

[jasong]

Originally Posted by petrw1

28 months later:
7 day average has increased from ABOUT 1500 to ABOUT 2000 (33% increase).
Accounts decreased from 55186 to 44468 (20% decrease).
Machines decreased from 80030 to 67490 (15% decrease).

Our throughput is increasing even as our number of machines decreases. This is great news.

However per the post that asks if Moore's law will ever die: GIMPS is not keeping up with Moore's law. Even if we were to average throughput per number of machines we are still well under a 50% increase ... not the more than double Moore's law would suggest.

IMAO, once the new Prime95 version is more stable and people start using the multi-core option to test exponents, throughput will increase dramatically. I wouldn't be willing to risk money on my opinion, but that's my best guess.

The throughput for 2 cores is about 1.93, which is only 3.5% difference. I don't know about anybody else, but when I download an exponent, the first thing I do is look at the projected end date. Having end dates that are almost halved will easily trump the little bit of loss from overhead. People will crunch more because credit can be gotten more often.

[davieddy]

Originally Posted by petrw1

28 months later:
7 day average has increased from ABOUT 1500 to ABOUT 2000 (33% increase).
....
However per the post that asks if Moore's law will ever die: GIMPS is not keeping up with Moore's law. Even if we were to average throughput per number of machines we are still well under a 50% increase ... not the more than double Moore's law would suggest.

My observations suggest the 7-day average is nearer 2400
so Moore isn't doing quite that badly.

[petrw1]

From http://www.mersenne.org/primenet/status.shtml today...........

------- Aggregate CPU Statistics, P90 Units* -------

Last 7 Days Average Cumulative Today
from 19 Jul 2007 06h from 25 Jul 2007 06h
---------------------- ----------------------------------
Test Type CPU yr/day GFLOP/s CPU years CPU yr/day GFLOP/s
------------ ---------- ---------- ---------- ---------- ----------
Lucas-Lehmer 1892.975 22787.061 697.385 2106.344 25355.533
Factoring 51.492 619.842 21.804 65.855 792.742
---------- ---------- ---------- ---------- ----------
TOTALS 1944.466 23406.902 719.189 2172.199 26148.275

The "Cumulative Today" average fluctuates wildly from 1600ish to 2300ish, but I have yet to see a recent "Last 7 Days Average" much over 2050. Are we looking at different stats?

[davieddy]

I was looking at GFlops/s and you were looking at
CPU years/day. Mystery solved I think