Nforce and Kentsfield don't love each other

READER feedback came in fast: within a few hours after the 680i vs 975X review, and despite the net cables cut due to the ferocious Taiwan quake, a couple of interesting comments emerged.

That included one about the poor experiences by a chap who tried to use Kentsfield quad-core QX6700, Intel's current top of the line desktop CPU, with EVGA Nforce 680i board.

Basically, the thing overclocked over 20% poorer than on his Intel-based Asus board - 2.9 GHz vs 3.6 GHz (water cooled), and the FSB also couldn't move much beyond the standard 1066 MHz. A bit disappointing for a top-end "enthusiast CPU" on a top-end board "made for enthusiasts", isn't it?

The experience implied that Kentsfield and NF 680i don't really feel 'heat' for each other (except for the CPU and North Bridge heat output, of course), but in fact it could be a problem between the inconsistent QX6700 overclocking achievements, unit by unit, and the same on the NF 680i side - not to mention the early board and BIOS quirks.

As mentioned in my own reviews - if talking about 100% stable long-term operation where everything, and I mean everything, runs as it should - my particular QX6700 sample was limited to 3.2 GHz / 1066 FSB on both Intel and Nvidia boards.

The various website's mileage with overclocking the Kentsfield generally doesn't vary by much: no one seems to have obtained stable operation with any sample at a FSB above 1333, even if using those nitro-cooled technique.

A few weeks ago, at an Nvidia plugfest here in Singapore, our brave colleagues from VR Zone faced the gas poisoning (i.e. nitrogen vapours) for the really powerful CPU cooling, and CO2 from dry ice for the Nvidia GPU cooling, and put a truly huuuge heat sink/fan combo on the Nforce 680i North Bridge. Yes, after all the effort in which three grown men fed the overclocked monster (see photo) they got that 4.65 GHz CPU clock and completed all the 3DMarks with it - yet they could only get a bit above FSB 1200!

The performance potential of 4 cores running at 4.65 GHz (which, by the way, is a 75 double precision GFLOPs peak performance supercomputer on its own!) would surely get brutally strangled by a narrow 9.6 GByte/s FSB like this.

During the Nvidia pre-launch Asia Pacific press discussion, I was told that NF 680i should easily handle 2.1 GHz FSB on the Conroe, and up to 1.8 GHz FSB on Kentsfield. I didn't see anyone yet achieve anything close to them in practice. More work needs to be done to get these numbers out, now - the RD600 is coming, and the rumours about its capabilities are there. µ