Xeon X3220 overclocked from 2.4GHz to 3.4GHz

AS PROMISED, prior to my regular China flight, I rushed to get that wondrous but multiplier-handicapped X3220 2.4GHz/FSB 1066 quad-core Xeon variety up and running on the Asus Striker Extreme, the current wunderkind of Nforce 680i boards. At least when it's updated with 0701 or newer BIOS for FSB 1333 glitch problems.

Since even the old QX6700 Kentsfield did fine at 3.33GHz CPU/FSB1333 on this board, I wondered how far the newer, more polished Xeon variety would go on this board. Ater all, it booted Windows and ran Sandra & 3DMark06 CPU with FSB1420 on 975X (MSI Neo Platinum), P965 (Asus Commando) and non-OC NF680i (Asus P5N32-E SLI Plus), neither of which could run the above mentioned Kentsfield at more than FSB1066 (all bench done well) or FSB1200 (just Windows boot).

The problem was, I only had a bit under two hours to do all this - so, all the same components - Corsair XMS6400CL3 and G.Skill DDR2-1066CL4 memories, as well as Asus V60 silent Cu-Al heat sink - were used to save time. Initial run was all Auto overclocking and voltage settings with fixed clock multiplier ratio of nine. The 3GHz/FSB 1333 worked right out of the box, Sandra & 3DMark CPU done. The only problem was, Nvidia Monitor utility in WinXP was showing vastly different voltages than the BIOS voltage monitor! The CPU voltage (1.32 V in the BIOS monitor) was seen as 1.45 V in Windows, while FSB 1.4 V in BIOS was shown as 2.25 V in Windows - huh, that's quite a difference.

Next step: set FSB to 355 x 4, i.e. 1420 throughput, same as all other boards, for 3.2 GHz CPU clock - also worked fine, tests passed, all settings still on Auto. The bold step after, FSB at 375 x 4 = 1500 MHz throughput, or 3.38 GHz CPU. The BIOS CPU voltage shown was 1.33 volts, and the FSB & North Bridge were at 1.45 volts (red band). The heat sink wasn't hot yet on the CPU, but, despite 2 fans mounted on the heat pipes, the NorthBridge could fry an egg! Nevertheless, both tests completed.

Well, time was running out, so I went straight to FSB 400 x 4 = 1600. System booted, but no Windows boot unfortunately, both WinXP and WinXP 64 crashed few seconds after boot start. I first tried to go higher (up to 412 x 4 FSB, or 3708 GHz CPU clock, see photo), as there are FSB 'black holes' at certain frequencies where you might just pass if you get the FSB a bit UP rather than down - CPU booted, but, again, not the Windows. So, I went down, back to 388 x 4 FSB settings, Windows booted, but benchmarks crashed on run. At 383 x 4, Sandra ran, but not 3DMark - same at 380 x 4.

Finally, the setting where everything worked was 378 x 4, or FSB 1512 - or 3.41 GHz CPU clock. This time, he BIOS CPU voltage monitor displayed 1.34 volts, and the FSB & North Bridge were both at 1.5 volts (really flaming red band). CPU temperature in Nvidia Monitor after running both benchmarks was 46 C, not bad at all. After that, I tuned the memory to CL 3-3-3-5 a DDR 756 at 2 volts, it worked at 1T command time even with all four DIMMs installed, Corsair & G.Skill together.

Here are the results, using Sparkle Calibre P880 GeForce 8800GTX graphics (the one with Peltier Junction):

3DMark06 CPU 4971
3DMark06 UXGA 10928
Sandra CPU int/fp 64013 / 43802
Sandra media int/fp 380557 / 205970
Sandra memory int/fp 7831 / 7846

So, last night before I left, the system at this setting completed the above benchmarks without a hitch - the heat sink was getting a bit warmer, but still far from anything uncomfortable. The point is, in an hour, I went from the default 2.4GHz clock to above 3.4GHz (seemingly) reliably overclocked in a hot non airconditioned room with just air cooling on both CPU and North Bridge - a whopping 42% overclock! Not bad at all, knowing that, at the end, this is not a single die, but a Multi Chi(m)p Module - as one of the readers said in reply to yesterday's story, Intel might really be pulling out all stops to tune the parts to the hilt before AMD Barcelona sees the (public) light. Or, is it just that the real best parts always go under the server SKUs? Whatever the case, I look forward to the next batch of these 'UP Xeons', and new BIOS revs on the Asus Striker Extreme.

One point though: Asus should fix the size problem with its heat sinks on its own boards - as you can see on the photo, Asus V60 heat sink doesn't fit very well into the Asus Striker Extreme mainboard, if you want to use Asus vertical fans on the Asus heat pipes on this Asus high-end board product. For heaven's (or more importantly, users) sake, didn't anyone tests putting these together first when designing the stuff? After all, these vertical fans are out on Asus boards since over a year ago. Basically, I had to twist & turn and mount the two fans tilted, under quite an angle, compared to their 'natural' upright position, and waste 10 minutes of time in the rush. Mounting at an angle can be painful, huh.