Graph over time of what kind of machines run GIMPS (PrimeNet) - Page 2

GP2 · 2003-11-17, 00:57

I added hbock's data, and two new interesting things stand out.

- Possible wave in trial factoring data?

- The decline (in total machines, total accounts, and PIII and Athlons) following the post-M39 peak is practically a straight line downhill.

See http://opteron.mersenneforum.org/ for links to all the graphs.

QuintLeo · 2003-11-19, 15:15

I'll point out that some CPU types are still not recognised by GIMPS, and some of the waves might be accounted for in part by newer CPU types being recognised by GIMPS as "unknown type" - like when the Thoroughbred Athlons showed up, it took George a few months to get around to updating Prime95 to recognise those, they were an unknown CPU type for a while - and the K5 is *still* an "unrecognised" type, abet in probably pretty low numbers outside of my factoring farm....

GP2 · 2003-11-19, 15:44

"Unspecified type" was in the thousands way back in 1999, but in early 2000 dropped below 1000 and ended the year around 450. It's never been above 1000 since (though it jumped a bit in Nov 2001 like all the other CPU types), nearly always in the low hundreds.

In any case, the total machines graph already includes "unspecified type" as part of the total.

Some of the odd effects in 1999 are undoubtedly artifacts (in the CPU types graph): at one time Pentium II and Pentium Pro were combined into a single type, P3s and Celerons weren't recognized, until about Oct 12 1999. So their gradual climb from zero (and accompanying gradual decline in "unspecified type", not shown in the graph) is undoubtedly artificial.

But I'm fairly sure the waves in later years are genuine. "Unspecified type" doesn' t have the numbers to affect the graphs much except up to early 2000.

Maybeso · 2003-11-19, 21:53

The waves don't appear to be annual -- it's hard to tell when each mark is 3 months. But the pattern is strongly annual.

Perhaps holidays, school years for students, pc purchase times, and internet hookup times, each have an influence.

2003-12-13, 05:29

Looking at the numbers got me thinking about investment and expenses. Some back of the envelope figures:
23K Athlons at $1.5K, 17K P4 at $1K, 7.5K P3 at $200 and 4K Celerons at $200 is 51, 500 machines with a market (not new cost but market) value of $53.8 million. Running at 150 watts each (probably more with those 400W supplies in use) is ~7700 kilowatts. Electrical cost per hour at 5 cents per kWhr is around $390 per hour.
Just some rough estimates. They don't reflect your machine being used for other purposes.
There's market segment data here too. Interesting about the Athlons outstripping P4

QuintLeo · 2003-12-14, 23:41

150 Watts is reasonable for an Athlon or a P4, but probably on the high side for anything else - and WAY high for P5/K5 and even K6/P3-Celeron class machines.

Some will be higher, some lower, even wihin a class - depends on how "tricked out" a given machine is.

Then there's dualy machines - which *usually* should eat a little less power per CPU....

To put things in perspective - my K5 boxes eat anywhere from 55 to 65 watts, depending on CPU speed and specific motherboard/HD/vid card. My K6s eat 60-ish to 85, depending on specific configuration and speed. My Athlons are all gaming boxes, and a lot more tricked out than the K5 or K6 machines, and eat 135-150 watts, except for the Tyan Tiger dualy that eats a little less than 300. My only P4 box eats about 130 (Shuttle SS51 box).

I suspect a lot of folks would LOVE to see your 5 cents per watt, though - I've got that here, but I'm on a special "time of day" rate.... 9-)

nucleon · 2003-12-17, 10:36

Quote:

Originally posted by tinydog
Looking at the numbers got me thinking about investment and expenses. Some back of the envelope figures:
23K Athlons at $1.5K, 17K P4 at $1K, 7.5K P3 at $200 and 4K Celerons at $200 is 51, 500 machines with a market (not new cost but market) value of $53.8 million. Running at 150 watts each (probably more with those 400W supplies in use) is ~7700 kilowatts. Electrical cost per hour at 5 cents per kWhr is around $390 per hour.
Just some rough estimates. They don't reflect your machine being used for other purposes.
There's market segment data here too. Interesting about the Athlons outstripping P4

You don't have an equivalent tflops figure of the collective 'gimps machine'?

From the top500 computing pages. The Earth Simulator needs 3000kW, and gives 35Tflops.

I rekon it would be an interesting comparision between gimps as a whole and some of the super computer farms around the place.

-- Craig

GP2 · 2003-12-17, 15:30

Quote:

Originally posted by nucleon
You don't have an equivalent tflops figure of the collective 'gimps machine'?

At
http://www.mersenne.org/primenet/
the figure given is 8739 gigaflops (this changes slightly from day to day).

2003-12-13, 05:29	#16
tinydog 2×7×269 Posts	Dollars and cents Looking at the numbers got me thinking about investment and expenses. Some back of the envelope figures: 23K Athlons at $1.5K, 17K P4 at $1K, 7.5K P3 at $200 and 4K Celerons at $200 is 51, 500 machines with a market (not new cost but market) value of $53.8 million. Running at 150 watts each (probably more with those 400W supplies in use) is ~7700 kilowatts. Electrical cost per hour at 5 cents per kWhr is around $390 per hour. Just some rough estimates. They don't reflect your machine being used for other purposes. There's market segment data here too. Interesting about the Athlons outstripping P4

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
GIMPS website looks kind of weird now	ixfd64	PrimeNet	1	2014-10-07 03:33
Wich kind of Gimps task's better fit this?	Damian	Information & Answers	19	2011-01-06 01:55
Machines Missing from Primenet Summary	edron1011	PrimeNet	2	2008-11-17 19:16
Graph of leading edge of LL testing (and double-checking) over time	GP2	Data	10	2003-11-17 14:55
Alienware now makes fastest GIMPS machines out there?	E_tron	Hardware	11	2003-09-03 03:08

2003-11-17, 00:57	#12
GP2 Sep 2003 2×5×7×37 Posts	I added hbock's data, and two new interesting things stand out. - Possible wave in trial factoring data? - The decline (in total machines, total accounts, and PIII and Athlons) following the post-M39 peak is practically a straight line downhill. See http://opteron.mersenneforum.org/ for links to all the graphs.

2003-11-19, 15:15	#13
QuintLeo Oct 2002 Lost in the hills of Iowa 111000000₂ Posts	I'll point out that some CPU types are still not recognised by GIMPS, and some of the waves might be accounted for in part by newer CPU types being recognised by GIMPS as "unknown type" - like when the Thoroughbred Athlons showed up, it took George a few months to get around to updating Prime95 to recognise those, they were an unknown CPU type for a while - and the K5 is still an "unrecognised" type, abet in probably pretty low numbers outside of my factoring farm....

2003-11-19, 15:44	#14
GP2 Sep 2003 2·5·7·37 Posts	"Unspecified type" was in the thousands way back in 1999, but in early 2000 dropped below 1000 and ended the year around 450. It's never been above 1000 since (though it jumped a bit in Nov 2001 like all the other CPU types), nearly always in the low hundreds. In any case, the total machines graph already includes "unspecified type" as part of the total. Some of the odd effects in 1999 are undoubtedly artifacts (in the CPU types graph): at one time Pentium II and Pentium Pro were combined into a single type, P3s and Celerons weren't recognized, until about Oct 12 1999. So their gradual climb from zero (and accompanying gradual decline in "unspecified type", not shown in the graph) is undoubtedly artificial. But I'm fairly sure the waves in later years are genuine. "Unspecified type" doesn' t have the numbers to affect the graphs much except up to early 2000.

2003-11-19, 21:53	#15
Maybeso Aug 2002 Portland, OR USA 100010010₂ Posts	The waves don't appear to be annual -- it's hard to tell when each mark is 3 months. But the pattern is strongly annual. Perhaps holidays, school years for students, pc purchase times, and internet hookup times, each have an influence.

2003-12-14, 23:41	#17
QuintLeo Oct 2002 Lost in the hills of Iowa 111000000₂ Posts	150 Watts is reasonable for an Athlon or a P4, but probably on the high side for anything else - and WAY high for P5/K5 and even K6/P3-Celeron class machines. Some will be higher, some lower, even wihin a class - depends on how "tricked out" a given machine is. Then there's dualy machines - which usually should eat a little less power per CPU.... To put things in perspective - my K5 boxes eat anywhere from 55 to 65 watts, depending on CPU speed and specific motherboard/HD/vid card. My K6s eat 60-ish to 85, depending on specific configuration and speed. My Athlons are all gaming boxes, and a lot more tricked out than the K5 or K6 machines, and eat 135-150 watts, except for the Tyan Tiger dualy that eats a little less than 300. My only P4 box eats about 130 (Shuttle SS51 box). I suspect a lot of folks would LOVE to see your 5 cents per watt, though - I've got that here, but I'm on a special "time of day" rate.... 9-)