Really depends on the cache hit and miss rate, plus the cost of backing up the pipeline and the cost of a stall even with OOO execution. Generally benching with SPEC2000 should get you the data you need to see which is more important.

But on another point it also really depends on the program being run too but me thinks SPEC2000 should give you a rough conjecture of what's best.

if i recall correctly

a quick link for the comparison.. even on a divider tight timings is better

http://www.dfi-street.com/forum/showthread.php?t=27875

This post has been edited by babyelf: Dec 29 2005, 03:53 PM

Does it implement on benching of two diffrent processors or just 1 regarding the "cache hit and miss rate, plus the cost of backing up the pipeline and the cost of a stall even with OOO execution"?

sorry, quoted wrongly. Meant that answer for the thread starter.

Simply said, it really depends on the chip. Like for the P4, it's insane miss rates due to replay really made low latencies redundant in certain situations.

Depends on what you need really, but benching the processor's cache can show the architecture's strengths and weaknesses and allow you to tune your overclock to either the timings or raw cycle speed.

lol "if i recall correctly" is for antonio

he asked for iirc..

regarding your post that's another reason why intel do ok with high timings

but baby....

the test were made simply with 2.5-4-4 not 2.5-3-3....so in other words...not the exact score u will get with 255-260 2-2-2-5 (BH)/300 2.5-3-3-7 (TCCD)settings....

http://www.xbitlabs.com/articles/cpu/display/netburst-1.html

The great guide to netburst will explain XD

But really, Netburst only shows it's true power when pushed beyond 4GHZ and beyond.

that was just an example

http://xtremesystems.org/forums/showpost.p...58&postcount=23

this better?

netburst has alot of potential.. intel see a huge potential for it but the problem with it is when they try to raise their clock speed the amount of heat generated is just ridiculous..

netburst is coming to an end now.. last cycle of its life

Sad but true.

But the new benches of Yonah are impressive, A64 matching performance WITHOUT the internal GMCH. Intel is showing that it's still in the enthusaist game, and once the play the performance per watt game, the next gen chip battle is going to get toasty

the 955XE looks good.. but FX60 coming soon too

impressive as it is i still think AMD will hold the dual core arena.. well my personal opinion.. putting high hopes on FX60..

at least the battle is close this time round

price wise i think intel will have the lead due to their die shrink

much better...but a few compromises in benchmark/HTT....

tight timing usually wins over ram mhz ler....
here is another good example
http://i4memory.com/showthread.php?t=1662

This post has been edited by sup3rfly: Dec 29 2005, 09:49 PM

Assuming clock speed of the CPU is similar, 300HTT of 3-4-4-8 TCCD chipset registers a lower synthetic benchmark value in Sandra compare to 260 2-2-2-5 BH5. But the difference in value is not that significant. This has been discussed at great lengths with all the benchmark figures in xtremesys back then. It shows that at some point MHz is better and at some point timing is better. It depends on the applications, hardware and some other factor as well.



It depends on the gap also, say you have 200HTT 2-2-2-5 and 250HTT 2.5-3-3-6 with both setup and the same CPU speed. Obviously it's a no brainer, the latter would perform better.



Good points, one of the factor needs to be taken into consideration as well.

100% sure ,,,,tested myself!

higher fsb is always better than tight timming with lower fsb!

your comparison are made with same processor speed ??

I do feel that timing is important.....

there is 1 time i run a system at default setting and the ram timing was 3-4-4-8. My boss just said, hdd is too low speed lar, loading game so slow...so i go into bios and set a new timing, 2.5-3-3-7, and no more complain lor....

since you're going to do with little overclocking, i believe that lower fsb with 2225 is better than a bit higher fsb with 3xx8.

example,

210 @ 2-2-2-5 vs 220 @ 3-x-x-8 if i assume that you're only overclock a bit.