[Shaktai]

Since D2OL has caught on with several folks now, I thought I would post some benchmarks. These are based upon a 7 day average because the size and speed of D2OL work units varies dramatically. In order to compare, be sure to crunch for at least a week. A single day's work will not give an accurate picture (My PM 800 can very from 39 to 55). Memory on all machines is 256 mb except the PM 800 which has 512 mb.

iBook 600, OS X 10.2 - 32
PowerMac 800 (single CPU - no level 3 cache), OS X 10.2 - 46
Celeron 1.3 ghz, Win XP - 46
Celeron 1.3 ghz, Mandrake Linux 8.2 - 76.5
Athlon 1900 (1.6 ghz), SDRAM, Mandrake Linux - 87.5
Athlon 1900, (1.6ghz), DDRAM, Mandrake Linux - 93
Athlon 2200, (1.8ghz), DDRAM, Mandrake Linux - 105

Team MacNN is now #2 in daily production and we should make #9 over all in about 6 days, and #8 overall in about 9 days. Less if we get another boost in production. From then on the competition gets tougher but we might make the top 5 by years end with just our current production. Also note how much faster the Celeron above is under Linux compared to Windows XP. You can also see that A Power Mac can crunch on par with the Athlons running Linux on an equal megahertz basis.

[jswayze]

FWIW, I've averaged the data from my two 1GHz nodes (Dual 1GHz CPU - about 8 CPU-days) and I come up with 72.5 condidates/day. I've got to believe that a dedicated CLI and/or Altivec-enabled client would produce MUCH better results.

[Shaktai]

Originally posted by jswayze:
FWIW, I've averaged the data from my two 1GHz nodes (Dual 1GHz CPU - about 8 CPU-days) and I come up with 72.5 condidates/day. I've got to believe that a dedicated CLI and/or Altivec-enabled client would produce MUCH better results.

Is that 72.5 per CPU or combined?

[jswayze]

That's per CPU. Here are my current stats:

Code:
Rank	ID	   Cands.	Confs.	Avg Cands./Day	 Best	           Total Docking Time
1862 10426	388	   7,760	 14.37	  	       nci_1160_1915 		5 days, 10 hr, 06 min, 08 sec
1920	11071	369	   7,380	 46.12	  	       nci_1152_1533 		5 days, 04 hr, 06 min, 44 sec
4359	11174	40	    800	   10     		       nci_1039_0063 		3 days, 01 hr, 27 min, 05 sec
4702	11246	28	    560	   28	     	       nci_1114_0063 		19 hr, 29 min, 48 sec

Their "Avg Cands./Day" is way off... wierd.

-J

[Shaktai]

Originally posted by jswayze:
That's per CPU.

Their "Avg Cands./Day" is way off... wierd.

-J

Per CPU, that is actually very impressive numbers. Apparently the L3 cache provides a big benefit for Macs. (My PM 800 doesn't have the L3 cache) That is even better then the Athlon's running Linux on an equal megahertz basis and blows away everything on windows. D2OL does use a vector java library, but it is hard to tell exactly how much that contributes. While technically not "altivec enabled" G4's with L3 cache do very well. They can match or better the Athlon's with Linux, and will be typically from 30 to 70% faster then the Windows boxes on an equal megahertz basis.

As to "their" avg cands/per day, it is based upon total time since the node was first registered. If you have any down time or time when you aren't crunching, then that gets factored in as well.