specific question about HTT, 1:1, divider...

[CHaynes112]

I am trying to figure out which of the following 3 configurations would yield me the best performance... This will help me decide if I need the 11x multi (or just the 10x), and if I need higher HTT (vs tighter timings)... I am not asking about stuff in general... I am asking only about this specific situation :)

1. CPU - 2800 (280x10)
RAM - 280 (9:10) = 252 @ 2.5-3-2-7

2. CPU - 2805 (255x11)
RAM - 255 (1:1) = 255 @ 2.5-3-2-7

3. CPU - 2800 (280x10)
RAM - 280 (1:1) = 280 @ 3-3-3-7

#1 has a higher HTT on the CPU than #2, however, since the RAM is only running 252, perhaps the higher HTT of the CPU means nothing. #2 has 3MHz more or the RAM, but a lower HTT on the CPU, however, it does run 1:1. And #3 has the same high HTT as #1, however, it has 25MHz more on the RAM than #2 and also runs 1:1. A lot of times, I see threads (at DFI Street for example) and the end result is always "It doesn't matter because the difference is so small and you would only see a SMALL difference in certain benchmarks". Well, I am interested in those 'small' differences :) I know comparing #2 to #3 is the classic "Timings vs. MHz" battle... but what about #1 vs. #3... even though a divider is used, and the RAM MHz is lower, do the tighter timings justify it?

One more thing... do some benchmarks 'like' higher MHz and others 'like' tighter timings (ex: SuperPI, Sandra, 3DMark, etc.) or do they all usually benefit from the same configuration? If certain benchmarks do like certain things, what are a few examples?

-Thanks

[geoff2k]

> I am trying to figure out which of the following 3 configurations would yield me the best performance...

Best performance in what? Benchmarking? Pure RAM bandwidth? 3D gaming? What are your priorities?

[cx323]

just test them all in what you want max performance in and use the one that does better in that task

[Major]

My guess would be 280x10 1:1 3-3-3

But as pointed out above you really need to run benchmarks at the various configurations and make an educated decsion about your system performance.

[STEvil]

1 may not be faster than 2 since your HTT multiplier is not listed.

[ozzimark]

My guess would be 280x10 1:1 3-3-3

But as pointed out above you really need to run benchmarks at the various configurations and make an educated decsion about your system performance.
agreed.

also, this might be interesting to some of you.. taken from a little disagreement i had with a guy over at [H]:

all right. some numbers

settings for 300x10:

http://eclipseoc.com/phpwcms_filestorage/1/1_153.png


and settings for 375x8:

http://eclipseoc.com/phpwcms_filestorage/1/1_154.png

take note of one thing: the max async latency. i had to raise this to 11ns to boot.



now, the numbers

memtest
300x10 -> 2860mb/s
375x8 -> 2860mb/s

sPI 1m
300x10 -> 27.812s
375x8 -> 27.812s

sandra
300x10 -> 7988/7895mb/s
375x8 -> 7981/7911mb/s

everest
300x10 -> 7996mb/s read
2600mb/s write
36.5ns latency
375x8 -> 8006mb/s read
2599mb/s write
36.4ns latency

cinebench
300x10 -> 62.3s
375x8 -> 62.0s



i think it's safe to say that they're the same damn thing ;)

stevil, ht link at 3x and 2.5x for 300mhz and 375mhz respectively

[STEvil]

and of important note is also that HT speeds dont seem to matter (very old thread on here somewhere about it), although I have not seen them retested now that we have dualcore cpu's ;)

[omga14]

here's a whole thread and test done with dividers down to 140: http://www.dfi-street.com/forum/showthread.php?t=27875

[CHaynes112]

> I am trying to figure out which of the following 3 configurations would yield me the best performance...

Best performance in what? Benchmarking? Pure RAM bandwidth? 3D gaming? What are your priorities?

Benchmarking and 3D Gaming are my top Priorities :D


cx232 & Major,

I can't run benchmarks to see which of these would be faster... as stated above... I am trying to get a 'preview' of what would be faster, before I buy another CPU and RAM... I was hoping this would help me decide between the Opteron 170 vs. 175, and the Plat EB vs. Ballistix :)

ozzimark,

That is great... I have been searching for a comparison like that forever... according to that, if all else is equal (CPU MHz, RAM MHz, and RAM timings) it doesn't matter if the CPU has a higher HTT or not. It appears to only be of benefit when the CPU HTT and RAM MHz are ran together at 1:1... but if the CPU HTT is raised WITHOUT the RAM MHz been raised (hence using a divider) then no gains are seen... which would make since, because the RAM MHz would be the 'bottleneck' here, and raising the CPU HTT all day with other dividers, wouldn't see any performance gains.

So to sum it all up... it seems that the higher CPU HTT in my example #1, won't really do anything since the RAM MHz is still them same... and possibly #2 would be SLIGHTLY faster, because of the 5MHz more CPU MHz and the 3MHz more RAM MHz... But, as Major and ozzimark pointed out, it seems that example #3 may be the better option of the 3 anyways... I have always heard that SuperPI and 3DMark seem to like higher RAM MHz... I am just curious as to if there are any common benchmarks that take a hit with 3-3-3 vs. 2.5-3-2? And one more thing... why is everyone so obsessed with OCZ Plat EBs when they can usually only do 3-3-2 upto 260-270 or so stable... wouldn't Ballistix at 280-290 @ 3-3-3 be better?

I think I have learned (for my application), an Opty 170 would be fine... really no need for the 11x multi... now I'm just trying to decide on the Plat EBs vs. Ballistix :)

-Thanks

[uOpt]

None of that fiddling matters for real-world performance.

[CHaynes112]

None of that fiddling matters for real-world performance.

As I stated above...

"A lot of times, I see threads (at DFI Street for example) and the end result is always "It doesn't matter because the difference is so small and you would only see a SMALL difference in certain benchmarks". Well, I am interested in those 'small' differences :)"

[uOpt]

As I stated above...

"A lot of times, I see threads (at DFI Street for example) and the end result is always "It doesn't matter because the difference is so small and you would only see a SMALL difference in certain benchmarks". Well, I am interested in those 'small' differences :)"

Well in that case :)

http://cracauer-forum.cons.org/forum/crabench-memory.html

http://www.cons.org/cracauer/crabench/memory.user_narrow.html

[CHaynes112]

:worship: :D

[ozzimark]

why is everyone so obsessed with OCZ Plat EBs when they can usually only do 3-3-2 upto 260-270 or so stable... wouldn't Ballistix at 280-290 @ 3-3-3 be better?
i personally like the ballistix more myself ;)

also, in a similar note to what uOpt posted, there's more such info that portrays similarly small differences, and includes 2T in the testing :D
http://www.hardforum.com/showthread.php?t=917047