To determine the effects of modifying cpu Mhz, system bus speed (and thus ram speed), and ram timings upon the processing speed of a UD task.
Using UD_agent, restore a backup task.
wait till the initial molecule first appears and start timing.
Stop timing when the final molecule has finished all its conformers.
An arbitary series of 10 molecules with a varying number of conformers was chosen.
Molecules were part of a Long Molecule Task.
The memory timing tests were done twice to ensure that the results i were getting were consistant. times never varied by more than 0.5 of a second (max accuracy of stopwatch)
CPU: Athlon 1200C
Mobo: iwill KK266r-plus
Ram: 2x256mb PC150 kingmax SDRAM.
OS: Win2k, sp2.
Apps running: UD, UD_agent.
Timing device: stopwatch.
Error: +/- 0.5 seconds.
Think: ver 1.03g(2617)
Target protein: Cyclooxygenase-2
Molecule range used for testing: 1-643-5-3-1 to 1-643-5-3-10
Task Priority: Normal
View: for timing Think was left on the 'Application' window.
A. FOR RAM TIMINGS ONLY:
(all ram tests done with 1350@150 (9x150)
A. changing ram interleave has negligable short term effect.
B. Going from 6-3-3 cas3 to 5-2-2 cas2 gives 1.4% better performance. small, but still detectable given my testing method.
C. 1200 -> 1350 is 12.5% Mhz increase.
At 133/135 bus speed that gave us 8.8% reduction in time taken (26.7 sec).
at 150 bus speed it gives us 9.9% reduction (29.4 sec).
D. Keeping total Mhz fixed but varying fsb (aint unlocked cpu's handy?)
At 1200Mhz, 133fsb -> 150fsb = 12.5% increase in system bus speed AND ram speed,
which gave us only a 1.9% reduction in time taken (5.8 sec)
But at 1350Mhz, 135fsb -> 150fsb = 11.1%,
we got a time reduction of 3.1% or (8.5 sec)
Thus i conclude that for the AMD athlon & duron range of CPU's Mhz has the greatest effect on numbercrunching time, followed by system bus speed then ram timings.
1350@135 (10x135) = 276.0 sec and
1200@150 (8x150) = 296.9 sec.
I can see having DDR system would substantially improve timing scores, particularly the newer faster DDR motherboards based on the sis-735 or kt266a chipsets.
Can also see dual channel RDRAM having a BIG effect, but it being offset by the differing architecture (poor FPU) of the pentium4.
if anyone is interested and if i can be bothered i may do longer timings and more of them. see if i could get some scale happening...
i.e 1200@150 = what@133?
Athlon 1200C OC-ed to 1350 @ 150 & 1.86V (1.94V max).
Enermax 550W PSU!
Swiftech MCX462 HSF! (commin soon)
512Mb Kingmax PC150 SDRAM @ CAS 2.
40Gb IBM 60GXP.
16.8Gb IBM 16.8GP.
SB Live Player.
Leadtek Winfast GeForce 2 Pro 32Mb oc'ed to (215/460).
Pioneer 106S 16x dvd, 40x cdrom.
12x10x32x Ricoh CD-RW
OEMs selling "High End"PCs with integrated video will be forced into Q3tournaments using a TNT2M64!