- Joined
- Oct 13, 2003
- Location
- Hiratsuka Japan
Maybe I have been living in a bunker for several years, and can't overclock to the max worth snot. I do distributed computing via BOINC (SETI, Einstien, ClimatePrediction) so stability matters on my farm of 9 computers, consisting of 1 266fsb nF1, 1 nF2 200mhz single-channel, 3 200fsb nF2 dual-channel, 2 Via Kt-600's at 166 and 200 FSB, a Tyan tiger AMD 762 dual cpu board, and a lonely 910mhz P3, soon to be given away to the kid's day care center as a game/first computer machine (anyone have a freebie modern Japanese OS?).
My normal overclock efforts start at boosting cpu multiplier, finding stable, then back down and do minor adjustments to the RAM FSB. Obviously a poor low-hanging-fruit approach, given a test of Anandtech's mention of using memtest86's bandwidth test and concluding tRas 11-12 is "optimal" for nF2 boards, which indicates I could do much higher FSB speeds and lower multipliers by "loosening the RAM timings" from 3-3-3-8 to get the same "CPU speed in MHZ". Tried it on one A7n8X-X board single-channel, and watched the Einstien "time to completion" indicator steadily drop. My my, 'tis faster.
Several questions. Is Anandtech close to correct in scrapping tight mem timings, which back in the days of P3 overclocking with Crucial 2-2-2-5 PC133SDRAM used to be important? Whut da fook is bandwidth, and why should I care for my applications, which are large, exceeding L2 cache, running in memory, and swapped in/out as necessary every hour or so? Should FSB be upped on all boards/lower multipiers to get better performance? What tRas settings are you folks using, stable?
Thanks
My normal overclock efforts start at boosting cpu multiplier, finding stable, then back down and do minor adjustments to the RAM FSB. Obviously a poor low-hanging-fruit approach, given a test of Anandtech's mention of using memtest86's bandwidth test and concluding tRas 11-12 is "optimal" for nF2 boards, which indicates I could do much higher FSB speeds and lower multipliers by "loosening the RAM timings" from 3-3-3-8 to get the same "CPU speed in MHZ". Tried it on one A7n8X-X board single-channel, and watched the Einstien "time to completion" indicator steadily drop. My my, 'tis faster.
Several questions. Is Anandtech close to correct in scrapping tight mem timings, which back in the days of P3 overclocking with Crucial 2-2-2-5 PC133SDRAM used to be important? Whut da fook is bandwidth, and why should I care for my applications, which are large, exceeding L2 cache, running in memory, and swapped in/out as necessary every hour or so? Should FSB be upped on all boards/lower multipiers to get better performance? What tRas settings are you folks using, stable?
Thanks