"Xeon" Kentsfield overclocks better than desktop chip

SINCE THE first Core 2 Extreme QX6700 landed in my lab, I had that persistent problem with its overclockability.

If you count full reliable operation across all tested apps, it was stuck at 3.2 GHz CPU/1066 FSB on all mainboards I used, except the Asus Striker Extreme Nforce 680i gadget, which managed to complete the benchmarks at 3.33 GHz / 1333 FSB. It's a dual-die chip package, after all, and a three-load bus can't be an overclocking wonder from the sheer engineering point of view.

Then I got a hold of another fine chip fitting into the exact same socket, called Xeon X3220 - basically an UP server-oriented version of that same Kentsfield (actually it is also known as Clovertown UP. Its default specs are 2.40GHz CPU clock, 2x4MB L2 cache, and 1066Mhz FSB using 9X or lower multiplier (yes, it is locked unfortunately). I plugged it into both my other Asus board, the P5N32-E SLI Plus (also based on Nforce 680i, but without some overclocker gadgetry and less heat piping) as well as MSI 975X Neo Platinum PowerUp - both boards were limited to 1066 FSB with the old Kentsfield for full, reliable benchmark operation.

Guess what, both booted WinXP at 3.2 GHz CPU and 1420 MHz FSB (1.325 V CPU, 1.4 V NB / FSB), the maximum I could go due to the 9X fixed multiplier lock limitation - all that using Asus aluminium-copper air cooler (OK, a bit heavy one, as the photo shows)! It ran Sandra at 7.5 GB/s memory performance (Corsair and G.Skill CL3-3-3-5 at DDR2-710), and gave me 4,762 CPU 3DMark06 score - not bad at all... These numbers are quite a bit nicer than the same Kentsfield with It seems that this new stepping from Intel has vastly better FSB headroom - I'm now eagerly awaiting the X3230 - 2.66GHz / 1066MHz FSB version, and the X3240 - 3.00GHz, 1333MHz FSB flavour. Overclocking that one to FSB1600 would be the target. In the meantime, I'm maxing out the chip on Asus Striker Extreme first... read about that in a day.

Why all that? When AMD Barcelona chips come, they will have a big advantage on the memory performance. Intel knows this and, of course, it is doing its best to reach the impossible and drastically improve these FSB speeds as an interim fix till the CSI comes in Nehalem. With Penryn-class chips, Intel may have about 25% clock-speed advantage on its dual-die quad-core compared to AMD native quad-core late this year (3+ GHz vs 2.5 GHz official speeds, but FSB / memory performance uplift is needed to keep that advantage in as many apps (and more importantly, benchmarks) as possible. µ