Intel QX6850 is Kentsfield's last small step forward

BEFORE ITS YORKFIELD 45nm, and AMD's Phenom quad-core chips come out, Intel has done its final tuning and one last refresh of the 65nm Extremely Expensive Kentsfield, the QX6850.

It rates at 3GHz CPU and FSB1333, the fastest for a single-socket quad core right now. The chip is the same G0 revision as the last batch of QX6800 (2.93GHz / FSB1066) and X3230 (Xeon 2.66GHz / FSB1066) processors we had a look at over the past few months.

So, any real improvements in the clock speed headroom - with air only, first? After all, come on. No one would buy this to run it at the default rated speed. We ran it on the Asus P5K3 Deluxe mainboard using Intel P35 chipset with 2 x 1GB of fast and hot OCZ DDR3-1333 Platinum memory. Zalman 9700CNPS high-end copper heat sink / fan took care of the heat.

The CPU booted to BIOS - in and out of it - at 4.02GHz / FSB 1600 (multiplier x10) without much problem at 1.45 volts, whereas the QX6800 could muster about 3.86GHz on the same board with the same cooling and settings. So, there we go with a 4 per cent improvement - not much, but still a bit.

The stable operation in Windoze XP - i.e. the usual Sandra, Everest, 3DMark06 CPU stuff - could be sustained at 3.8 GHz flat with FSB 1520, compared with 3.7 GHz on the QX6800. The OCZ memory ran at DDR3-1520 in sync as well, with quite impressive 7-8-7-14 latency for an early DDR3 part at 1.7 volts. The Everest CPU and Sandra memory shots illustrate the 'achievement'.

In summary, the QX6850 is similar fare to the 6800 of a few months ago, with a bit more fine tuning added for a small final speed step up. This baby should ran fine at around or above 4.2GHz with decent US$150-class water cooling, and probably 5GHz with cascaded dual cryo-freeze units. This is as high as you'll go on Intel till the 45nm Yorkfield Extreme Edition surfaces some time soon after Intel's Fall IDF.

I hope AMD somehow comes out with a fast desktop quad-core Phenom too around that time, as it would make for an interesting pre-X'mas match - any delays there may result in Intel not exactly rushing to give us faster Penryn clocks at launch..

Oh by the way, one complaint: the CPU multiplier doesn't go above 11 - a problem if you want sky-high overclocks without sky-high FSB, where adding a notch, to say 12, wouldn't be a bad idea. How about that?