Pumping a Core2 X6800 all the way up to 4GHz

IT'S INTERESTING that, for Intel's D975XBX mainboard, setting the "FSB Override" to 1333MHz instead of 1066MHz could seemingly be done, and the system would undergo a hard reboot after the change. But, still, the change is not implemented!

So, I had to stick with the "burn in" approach to push the FSB speed forward. After setting the Corsair DDR2-800 RAM voltage to 2 volts and running it at 4-3-3-10 at 800MHz, it was the time to proceed to the test run using the Thermaltake cooler, before I meddle with hot water (literally).

To make things simple and comparable, I used the same versions of usual PCMark05 and Sandra 2005 benchmarks, that I have used with my previous quickie test runs. The next round will be using the freshest versions, though - a fresh start with fresh architecture as a new reference.

Booting back at 3.46GHz 1066 FSB, not all benchmarks ran properly - Sandra ran fine, but PCMark05 CPU didn't complete its run. At 3.2GHz and 1.275 volts CPU input, everything ran fine, though.

I also run the benchmark at the same 3.2 GHz CPU speed, but with multiplier 10 and 1,280MHz overclocked FSB. Here's what we have, against a similarly overclocked AMD FX62 running at 3.2 GHz with the same DDR2 800 settings on Asus SLI32 mainboard - a neat quick clock-for-clock run!

Benchmark	AMD FX62	Core 2D FSB 1066	Core 2D FSB1280
PC Mark05 CPU	6380	8162	8196
PC Mark05 mem	4516	4548	4832
Sandra CPU int	23382	29711	29716
Sandra CPU FP	19397	20493	20506
Sandra memory int	9416	6828	7902
Sandra memory FP	9398	6822	7894

Interesting to note that the old PCMark memory test doesn't seem to favour the Athlon, even though it should win a raw memory bandwidth test anytime - as it rightfully does in Sandra, (after all, the integrated memory controller should win on this one).

With this done, I attached the Corsair Nautilus water cooler, my reference test liquid cooling system for on-off mainboard reviews. I put the whole mess under the air conditioner full blast (it was 33C outside, after all - this is Singapore, and yes, still cooler than London right now!), and made sure that both the fresh distilled water and the coolant are, well, refrigerated. Whether it helps long-term or not, is another question, but it does take away some CPU heat in the first few minutes, 'til the water heats up anyway.

I also put a water block on the North Bridge - Corsair does have an adaptor for Intel975XBX's irregular rectangle shape of North Bridge heat sink latches, so it went without a hitch. Now, the moment of reckoning - at 3.2GHz 1280 FSB, the BIOS went up fine, then Windows - the temperature of the CPU was only 38C! OK, go back to the BIOS, jump the multiplier to 12 with voltage to 1.3125 volts, and 3.84MHz Conroe XE boots! It went to Windows as well, and loaded it properly. The temperature shown was 43C, not a big deal. Benchmarks? Sandra complete, PCMark not (again).

Just for the sake of trying at that stage, I upped the "burn in" setting to 25 per cent instead of 20 per cent. This gives exactly 1.333GHz FSB and 4GHz CPU rate at this level. Windows loaded, passed the boot window, but froze at the user login screen. After a reset, upping the MCH and FSB voltage by a notch, and five-minute power-off rest, Windows did load to the main screen, and show the correct 4GHz clock with temperature of 46C! Tried to run Sandra, the benchmark run the integer part fine, but the FP part looked lower than usual - could it be throttling?

Anyway, to my dismay, after few benchmark runs, the system froze again. I repeated the process, and could boot the system and leave it to run for sometime, as long as there was nothing "very CPU/FPU intensive" running. Which, well, defeats the purpose of running a sped-up CPU, doesn't it? In summary, with a good board, good cooling (both the PC and the room), and some luck, you can get 4GHz out of Conroe XE. The overclocking margins are somewhat lower than on the Presler XE 965 (3.73GHz default clock, 4.32GHz air cooling at the 2.625 volts lowest voltage, and 4.8GHz with water cooler, running all the stuff). Over here, I will focus on getting the 3.84GHz fully stable first, which already is a third above the default X6800 clock rate. After this, let's see if 4GHz can really run stable without something even more exotic than water cooling. µ