Author: Robert Hyatt
Date: 07:59:33 01/16/03
Go up one level in this thread
On January 16, 2003 at 03:26:13, Aaron Gordon wrote: >On January 15, 2003 at 21:29:34, Robert Hyatt wrote: > >>On January 15, 2003 at 19:04:36, Aaron Gordon wrote: >> >>>On January 15, 2003 at 18:14:50, Jorge Pichard wrote: >>> >>>>On January 15, 2003 at 17:43:47, Jorge Pichard wrote: >>>> >>>>>On January 15, 2003 at 15:21:00, Robert Hyatt wrote: >>>>> >>>>>>On January 15, 2003 at 12:56:07, pavel wrote: >>>>>> >>>>>>>Robert, >>>>>>> I must have missed this in your earlier discussion, but how much >>>>>>>speedup are you getting on this Xeon? Did you replace your older ones with the >>>>>>>new ones already? >>>>>>> >>>>>>>Cheers, >>>>>>>pavs >>>>>> >>>>>> >>>>>>I simply replaced my quad 700 with a dual 2.8. It is somewhere around 2x >>>>>>faster, >>>>>>overall... >>>>> >>>>> >>>>Therefore, the 8x 1000 Mhz used vs Kramnik is roughly almost the same speed as >>>>the newer dual Xeon 2.8 Ghz. >>>> >>>>PS: I'm NOT comparing these two systems Mhz Per Mhz since 8x1000 = 8000 Ghz, >>>>whereas 2 x 2.8 = 5.6 Ghz, but the performance difference should be roughly >>>>equal using the Newer Dual Xeon 2.8/533. >>>> >>>>Pichard >>> >>> >>>I highly doubt that. You seem to be forgetting that the P3's are faster MHz for >>>MHz than the Pentium 4 chips. It roughly takes a P4-1.6GHz to equal one P3-1GHz. >> >>That isn't true in my case. My quad xeon 700 was a PIII-based xeon >>box. My dual 2.8 is almost exactly twice as fast, which is what the >>clocks say should happen. > >Actually, figuring the speed up into it... 4 p3 chips have a speedup factor of >3.1x. 700 * 3.1 = 2170MHz and according to you (I remember the message well) you >stated with the Intel C compiler your Quad got 1.6mn/s. Dual 2.8GHz w/ SMT >should get a speedup of around 2.21 over 1 cpu w/o SMT. This is (1+(2-1)*0.7) * >1.30 (30% boost from SMT). Now, 2800 * 2.21 = 6188MHz. This isn't the "total >MHz" of course, whats actually being used from Crafty. Anyway, as I said in the >previous message every post I've seen you're showing numbers around 2mn/s for >the dual 2.8... now.. The problem with the above is that you are calculating the speedup for _crafty_. Did you specifically want to use crafty? Or was your question "which machine is really the fastest assuming that the algorithms used are optimal?" The answer to the latter question is different from the first question's answer. The number I quoted was raw NPS, which is simply a raw performance indicator. It doesn't necessarily mean that the dual is faster or slower than the quad for chess, although, in reality, they seem to be equal because the dual _still_ has to use four threads to take advantage of SMT... >P3's @ 2170MHz = 1.6mn/s >P4's @ 6188MHz = 2.0mn/s >If you don't add the SMT as MHz you still get.. >4760MHz, 2.0nb/s (2800*1.7) >119% more cycles for what, 25% increase in performance? > >Now, the clocks say what again? ;) > >>>Also, about the "400/533mhz" bus business, it's not actually 400mhz or 533mhz. >>>It's just quad pumped (four data transfers per cycle). Thus a "400MHz" p4 bus is >>>actually 100MHz and a "533" is 133. Same goes for DDR (double pumped, two >>>transfers per cycle). I prefer DDR over RDRAM. DDR is much more efficient >>>(actually does what it's rated for). The PC1066 RDRAM only pulls around 3.2gb/s >>>actual but 240MHz(480) DDR hits 3.7gb/s easily (on an Nforce2 board). ;) >> >>If you think about it, does it matter? 1 transfer 400M times a second, or >>4 transfers 100M times a second? It is simply semantics, not practical speed. >> >>BTW my xeon box (2.8 x 2) is DDR ram, not rambus... I'm not a fan of rambus >>at all except for certain kinds of streaming memory applications... > >It does to me, I like everything to be right. I run my (DDR) bus at 183MHz, if I >said I was running a "366MHz fsb" I would be what? Wrong.
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.