PDA

View Full Version : which would give better performance?



Amoeba Assassin
07-11-2003, 10:52 AM
Right now I have a 2.6c at 276 fsb. If I got some pc3500 hyper x....

1. 2.6c @ 3.59, ddr442

or

2. 2.8c @ 3.5 with ocz gold at ddr500

which would be better? I'm looking for a trade.

Amoeba Assassin
07-11-2003, 01:43 PM
I'm thinking #2 would

Pepper
07-11-2003, 02:11 PM
Unbelievable :)

Thats just what I wanted to ask!! Just got in from work and I was gonna post it till I saw yours.

I don't know whether to get a 2.4 and watercool it or 2.8 and air cool it,costs about the same I reckon but which has the better ram performance?

I reckon I'll go 2.8 air cos I've got the better cpu and I can always go to water later and hit 4 gig hopefully!!!!

Pep

Amoeba Assassin
07-11-2003, 04:40 PM
Yeah I agree. Does anyone know which would be better, #1 or #2?

Sick0fthelies
07-11-2003, 04:44 PM
Im thinking the higher ram frequency wuld be better, its only 90mhz youre sacrficing.

Amoeba Assassin
07-11-2003, 06:10 PM
yeah & you never know, maybe it will run at ddr502, or 504 etc...:D

Zroc
07-11-2003, 09:09 PM
Except that that 90MHz of clockspeed you'd sacrifice is worth a lot more than that 58MHz of memory frequency.

13x276=3590 5:4 DDR442
14x250=3500 1:1 DDR500

-The overall hit from 1:1 to 5:4 is roughly 2%.
-Raw FSB typically returns roughly 20% actual performance.
-Raw clockspeed returns roughly 70% actual performance

276 is 26 FSB higher than 250, so 26/250 = 10.4% higher FSB.
10.4% * 20% = 2.08%.

3590MHz is 2.6% higher clockspeed than 3500.
2.6% * 70% = 1.82%

So...in real performance, that 2.6 at 3590 will be +2.08% +1.82% -2% (the 5:4 hit from 1:1) = 1.9% faster.

Basically, if all other settings are the same, the 2.6@3590 scenario will be roughly 2% faster. The x factor here is, you will most certainly be able to run better timings at DDR442 than at DDR500, assuming the same RAM, and you could pick up an extra 1-3% there, depending on how much more aggressively you can set it. This is pretty important folks...you can entirely make up (or even surpass) the small 1:1/5:4 hit with better timings, depending on how wide the timings gap is.

A couple keys to understanding modern P4 overclocking:

-Realizing that the P4's at current clockspeeds have an over-abundance of memory bandwidth to deal with. This is why you're not seeing huge penalties when using the 5:4 or even 3:2 dividers (the hit from 1:1 to 3:2 is roughly 5%).

-FSB isn't quite as important as it was with 533 procs...still important, but not quite as important. 20 FSB to a 533 proc is a 15% improvement...20 FSB to an 800 proc is 10%. 20 FSB more when you're already at 250 is only an 8% improvement. There is also now a bit of an FSB bandwidth glut compared to the clockspeed...you will see FSB (and memory bandwidth) becoming a bit more important as the clockspeed rises in the future (think 5GHz) and can take advantage of it.

-Clockspeed is king right now in P4 800 overclocking.

Hope that helps...

JeffPH
07-11-2003, 09:11 PM
zroc d man :)

charlie
07-11-2003, 09:32 PM
Zroc,
OK but what if it's identical clocks?
12 x 300 = 3600 5:4 ddr480 @ cas2, 3,2,5
13 x 277 = 3600 1:1 ddr542 @ cas3, 4,4,8 ((pretend)
Who wins????????
C

Major
07-11-2003, 09:52 PM
go read Zroc's thread, its all covered there !!! excellent reading !!! excellent job !!

http://www.xtremesystems.org/forums/showthread.php?s=&threadid=12734

charlie
07-11-2003, 10:45 PM
Yeah, but i'm Lazy
Hey Slaughter!
:bday:

Kunaak
07-11-2003, 10:49 PM
the tricky part is always the same.

who ever has the most stable OC, wins.
banzi runs are nice for screenshots but never get high score or good times on PiFast.

stability is still something to consider.

;)

Major
07-11-2003, 10:55 PM
Thanks Charlie !!! hehe I've been buying myself some presents for the last couple of weeks ! lol

I agree Kunaak 100%, It's cool to see how far you can push it just to bench and stuff, but for me it has to be 24/7 to "count"

Zroc
07-12-2003, 02:13 AM
Originally posted by charlie
Zroc,
OK but what if it's identical clocks?
12 x 300 = 3600 5:4 ddr480 @ cas2, 3,2,5
13 x 277 = 3600 1:1 ddr542 @ cas3, 4,4,8 ((pretend)
Who wins????????
C
LOL! Lazy...yoda would be angry ;)

That one's easy. You've got a 23 FSB difference...23 divided by 277 means that's an 8.3% raw FSB increase, which is 20% effective, so that's 8.3% times 20% = 1.66%. Subtract the 2% hit...theoretically, the 12x277 is faster. However, those better timings will make up the difference and then some.

Either way, they're gonna be so close it wouldn't matter ;)

Happy B-Day, Slaughter :toast:

I'm also in that same school of though, Kunaak. I'm not the bonzai 3DMarker...I just want to maximize my money while remaining 100% stable.

saaya
07-12-2003, 09:39 AM
ocz will get you much higher and the P4 doesnt gice a §§§§e about timings....

Amoeba Assassin
07-12-2003, 01:40 PM
damn zroc good point

anyways thanks for all the input, I think I might just have to stay with this