View Full Version : Given a choice, what would you pick?
Tonight, to settle a "what would win" question i've been asking myself, i'm going to be benching SPi 32M at 400x9, with the following RAM setups:
1:1- 400MHz 3-3-3
4:5- 500MHz 4-4-4
2:3- 600MHz 5-5-5
Which one do you guys think will give me the best time?
straps will be the same for all, subtimings will be too I reckon. I wont be playing any fancy tricks- I just wanna know. I have an idea, but im gonna see how it pans out in the real-world.
K
EDIT: yea, I know theres other options, that might well work out better e.g. 450x8 with 450Mhz or 562MHz RAM...humour me!
Look here: http://forum.pclab.pl/index.php?s=&showtopic=287314&view=findpost&p=4160012 - from i965 1:1 to P35 3:5 @ FSB 400, x8 [I played more time with Yorkfield than with x9 multi Conroe :p:], 16M.
I should've known I wouldnt be the first person to think of this...I wont click on the link- im gonna run the tests anyway :) I'll read it tomorrow or something :)
Latencies make a big difference, enough to level out a DDR200MHz edge on RAM. ;)
I would run 1:1 FSB, that gets best results and lowest 400 timings, OR better is 514 x 7 1:1 4-4-4-4 if possible. It'll be quicker than 450 x 8 563 5-5-5-5 but I'm not sure how it pans out vs 514FSB runs. PL makes a good amount of difference, keep it at 6 or better, below if you can.
Which do I think is better? I haven't ran 600 5-5-5 yet on P35 to answer that. 540 4-4-4 should win 600 5-5-5 in my prediction.
massman
11-20-2007, 02:07 PM
Difference between Cas4 500Mhz and Cas5 600Mhz is almost nothing, in favor of the Cas4.
If I where you, I'd go for the Cas3 memory :)
Gautam
11-20-2007, 03:47 PM
I disagree entirely. :p:
600 5-5-5 > 500 4-4-4 > 400 3-3-3
I'm about 95% certain of this, let's see what the outcome is. :D
Reefa_Madness
11-20-2007, 03:54 PM
I disagree entirely. :p:
600 5-5-5 > 500 4-4-4 > 400 3-3-3
I'm about 95% certain of this, let's see what the outcome is. :D
Gautam's opinion is supported here...
http://www.xbitlabs.com/articles/memory/display/core2duo-memory-guide_4.html#sect0
Running higher memory clocks, with looser timings, results in better performance. The above linked article does not test at the exact same clocks as the OP requested, but the results in the article still support Gautam.
filmbot
11-20-2007, 04:11 PM
I disagree entirely. :p:
600 5-5-5 > 500 4-4-4 > 400 3-3-3
I'm about 95% certain of this, let's see what the outcome is. :D
Im not saying I disagree with you, but I figure you have tested this and what Im wondering is, does the stress/length of the test also come into play? That might sound weird but basically what I mean is, would running 1M at 400 3-3-3 show more benefit then 600 5-5-5? And running 32M the opposite, 600 5-5-5 over 400 3-3-3?
I disagree entirely. :p:
600 5-5-5 > 500 4-4-4 > 400 3-3-3
I'm about 95% certain of this, let's see what the outcome is. :D
Yep
Although Hipro did some 2k1 tests and if you can run 1T with low timings (3-2-2) then on 975 chipset 420-430 1:1 is best
Gautam
11-20-2007, 04:50 PM
Im not saying I disagree with you, but I figure you have tested this and what Im wondering is, does the stress/length of the test also come into play? That might sound weird but basically what I mean is, would running 1M at 400 3-3-3 show more benefit then 600 5-5-5? And running 32M the opposite, 600 5-5-5 over 400 3-3-3?
I've tested 500 4-4-4 vs 400 3-3-3 and it was a landslide victory for the former. The 600 is just a hunch, but an educated one. :p:
I think usually the length doesn't matter. 1M and 32M seem to show how things stack up the same. Sometimes 32M will show differences that 1M doesn't, but I've never seen them disagree. What's faster for 1M is faster for 32M as well.
However for 3D, 450MHz 3-3-3 is faster than 600 4-4-4, but in any calculation of SuperPi, its the other way around easy.
3D is a completely different ballgame where 975X dominates everything but very aggressive settings on P35/X38 and/or DDR3 and tight timings generally have a tendency to blow past everything else.
I haven't ran 32M much because my RAM faulted before I could but 1M for me was slower at ~563, 600, 620 5-5-5 than 540 4-4-4 without a doubt. It gained some ms, but my PL was different (6 vs 7) so I can't be too sure why. I reckon the longer the calculation time the more the speed will make an impactual difference. You can gain over someone else by just 1MHz, pretty easily.
massman
11-21-2007, 12:16 AM
I disagree entirely. :p:
600 5-5-5 > 500 4-4-4 > 400 3-3-3
I'm about 95% certain of this, let's see what the outcome is. :D
Sorry, my bad :). 600MHz Cas5 would indeed be faster in superpi (did some tests a few days ago, should have look to the graphs before responding :D )
massman
11-21-2007, 12:17 AM
Another thing of interest is the Performance Level. If you're using 2:3, you should be able to use PL=5, no?
I ran the tests anyway- 400MHz 3-3-3 was a writeoff because I could only run it on a 333/1333 strap.
I ran 500MHz 4-4-4 and 600MHz 5-5-5 on a 266/1066 bus and there was about 7 seconds advantage using.... 600MHz 5-5-5.
Will try and get the pictures, but they were on EP...which has now died :(
I was surprised that 20% increase in clocks with a 25% increase in latency didnt give a performance loss overall.