PI: 680i VS 975X

[lopri]

OK I spent some time after eating half a turkey testing the mysterious Super PI numbers on this board. My conclusion: 680i will be eventually faster than 975X, provided NV will perfection the BIOS. This board has so much potential in it and right now the buggy BIOS really rains on the parade this board deserves. See below my test results (and comparision with 975X) and you be the judge.

CPU: Xeon 3060 (E6600) @3.60GHz
Memory: Crucial 10 Year Anniversary DDR2-667 2GB (1GBx2)
Motherboard: P5W-DH (975X), EVGA NF 680i (680i)
OS: Windows XP (Base Priority set to 'High')

SP 8M

975X | DDR2-800 (3-3-3-8-2T) | 2m 54.860s

975X | DDR2-1000 (4-4-3-9-2T) | 2m 51.562


680i | DDR2-800 (3-3-3-8-1T) | 2m 53.922s

680i | DDR2-800 (3-3-3-8-2T) | 3m 02.922s

680i | DDR2-1100 (4-4-4-9-2T) | 2m 57.672s


SP 32M

975X | DDR2-800 (3-3-3-8-2T) | 13m 58.875s

975X | DDR2-1000 (4-4-3-9-2T) | 13m 47.641s


680i | DDR2-800 (3-3-3-8-2T) | 14m 28.187s

680i | DDR2-1000 (4-4-3-9-2T) | 14m 01.188s

680i | DDR2-1100 (4-4-4-9-2T) | 13m 56.078s


When I first got this board, I couldn't understand why the PI times were so slow on this board. It turns out the board's subtimings are way loose, not to mention that the BIOS is quite messy. But as you can see, this board has a huge advantage when it comes to memory. Not only it clocks memory better than 975X, but also gives much more freedom for users. To maximize performance, one can;

1) Overclock memory to sky high, or
2) Configure memory timings as tight as possible should the memory not be capable of high frequencies.

This board has a lot to be explored and has a lot going for it, IMO. I can't wait to see what DFI will bring to the table with the same chipset!

[Dumo]

Nice comparo there lopri
680i also fast for 3d.

[lopri]

Granted I don't think 680i will excel 975X under exact same frequency/timings. 680i's chance to beat 975X comes from its flexible nature. (e.g. by doing 1200MHz where 975X can only do 1000MHz, etc.) Also the biggest boost will definitely be 1T, IMO. From my tests above, 800MHz/3-3-3-8-1T beat 1100MHz/4-4-4-9-2T.

[Dumo]

It will shine with 1333FSB cpu and >1000EPP memory

[erwinz]

nice work..

[JoeBar]

Thanks for the results!

[HKPolice]

Is there any way to adjust the sub timings?

[dinos22]

from your results there i would conclude that 975X has an edge over 680i

[gundamit]

Very nice results and consistent with what we've seen elsewhere. As of this moment 975x has the edge on Pi all clocks being the same. But does anyone doubt that most CPUs will see a higher OC on the 680i chipset? I fully expect 680i to carry my very average 6600 to the speeds it saw on the 965 chipset, with about a 200mhz advantage over 975x. That should be enough to push it past my best 32M times on the Intel chipsets. At least I hope thats the case and I am prepared to be incredibly bitter if it is not.

[lopri]

Is there any way to adjust the sub timings?

Here:


and here:


Be careful though. If you mess something up you can corrupt your OS installation.

[bingo13]

Lopri,

Asus Striker Results -
Crucial 10th Anniversary 2GB - 2.325V
E6600-
8800GTX
XPSP2 - Full Load with Sound enabled

3-3-3-6-1T 32M
tWTR- 6 (5 will lock this board)
tRD - 2
tRC - 10
tWR - 3

Still tuning it but this 0505 BIOS is very 1T friendly so far.

[bingo13]

Lopri,

Asus Striker Results -
Crucial 10th Anniversary 2GB - 2.325V
E6600-
8800GTX
XPSP2 - Full Load with Sound enabled
BIOS -0505

3-3-3-8-1T 32M
tWTR- 6
tRD - 2
tRC - 11
tWR - 3

[Gautam]

Seems like 680i 1T is indeed catching up to i975x 2t...compared to the stuff we've seen before, a 13 second gap is quite respectable.

[Spacemaster]

So 975X is beter when we go 1:1?

[Blacklash]

Granted I don't think 680i will excel 975X under exact same frequency/timings. 680i's chance to beat 975X comes from its flexible nature. (e.g. by doing 1200MHz where 975X can only do 1000MHz, etc.) Also the biggest boost will definitely be 1T, IMO. From my tests above, 800MHz/3-3-3-8-1T beat 1100MHz/4-4-4-9-2T.

The part I bolded got my attention. Interesting. Maybe I'll try the 680i.

[guess2098]

i would siad it is depends on FSB and ram speed. i has more test coming up tonight

Seems like 680i 1T is indeed catching up to i975x 2t...compared to the stuff we've seen before, a 13 second gap is quite respectable

[lopri]

Bingo,

Thank you for sharing your settings. I will definitely try them out. I am having a problem with 1T stability on my board. The most I can do is 8M. Anything more than that the system will crap out.

Is the 3 sec difference between the 2 runs coming from different tRAS (and subsequently tRCT)? If so, it's interesting in that tRAS normally has zero impact on Intel chipsets.

I found the 680i chipset responds well to voltages, as long as I can keep it cool. Swapped the stock heatsink/heatpipe with Thermalright HR-05-SLI and it's like the first time ever I've feel it (the Thermalright) getting hot. On most other boards, it either stayed cool or became mildly warm.

Will NV try to improve memory performance via BIOS update? From what I can see, the most urgent fixes (when it comes to overclocking) are 1) Filling the FSB holes, 2) Easier 1T.

[.p4nzer]

My opinion is: CPC must be enabled... 2t is too slow.