Author: K. Burcham
Date: 12:20:50 07/03/05
Go up one level in this thread
On July 03, 2005 at 12:17:20, Gian-Carlo Pascutto wrote: >I find these to be actually pretty interesting figures. There is also a direct >relation to Watts/knps assuming same size process technologies. Yes, I also find your statement interesting. We know that all things equal, if we load 25% of the transistors (call on them to draw current for switching) the current draw from the power supply is at a low value. If we load 100% of the transistors such as with a chess program, the current draw from the power supply would be near maximum (for a single task). Of course this maximum value is not refering to multi tasking such as burning a cd while running a chess program and then surfing, all at the same time. Each task will cause more current draw, even if a minimum amount. So I think what you are saying is true. I assume that the commercial servers run on 220/480 volt. Not sure about the large clusters. We have some large industrial computers at work that run on 3 phase 575 volt. Of course with this voltage the current potential can be very high if designed into the system. The T-leads going out of the cabinet are about 1 inch in diameter per phase. I would be curious to know what the current draw is for a cluster while running a single chess program, for 8 then 16 then 32. To get a true value of amps/kns or watts/kns, we could take a reading with a clamp-on, on one phase, record the value, and then record the second reading after the chess program was running drawing its maximum current. If we subtract the two we would have the actual amps/kns. Then like you said we could plug in the voltage into the formula and get our watts/kns. kburcham > >Does anyone have transistor counts / knps figures for modern ARM and PowerPC >CPU's? I would assume ARM to deliver easily the most bang-for-the-buck. > >Oh, and what about Itanium2? > >-- >GCP
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.