Dual core hits 4.8 GHz, Presler XE 3.73 on Asua jobbie First INQpressions Corsair Nautilus 500 Cooler
By Nebojsa Novakovic: Monday 05 June 2006, 23:55
WHILE WATER can only do so much to cool down your hot CPUs, GPUs and so on. Phase change and special liquids or freon-like stuff are next, I guess, if you want -30 C.
In the meantime, there still is space for improvement - for instance, easier installation, more efficiency, and less pipe clutter.
Corsair's third-generation water cooler, the Nautilus 500, tries to accomplish exactly that, and I put it through the paces with one of the toughest PC platforms to cool - a heavily overclocked Presler XE 3.73 GHz dual-core CPU, also known as the "Netburst' Last Outburst" before Cointreau (sooooory, Core 2 Duo) drinking becomes the new habit.
The box is as big as the predecessor, Corsair COOL, but this time, the cooling stuff is moved outside the casing, to sit on top in that 'fasionable' round case. The cooling blocks and the pipes are basically the same (and mostly interchangeable) as the COOL, but the larger external reservoir and radiator should help, on top of drastically reduced installation time of slightly less than 30 minutes in my case (to be precise, 'lucky' 28 minutes according to Chinese numbering), including North Bridge cooling.
This time, instead of Intel D975XBX, I used an Asus P5WDG2-WS mainboard, a drastically redesigned workstation-grade upgrade of their previous (not exactly stellar) 975X board. It has even two PCI-X slots for my old UltraSCSI320 drive set, and seemingly an improvement in the power circuitry for smoother overclocking. The memory was Corsair 5400UL, the old trustworthy set that can do CL 3227 at DDR2-667 speed.
So here we go: Default 3.73GHz at 1.3V, HT (HyperThreading, not HyperTransport) set on, 1066MHz FSB - fine! CPU temperature tops at 57 C in Intel monitor, stress test with no errors. Same at 4.26GHz, same voltage, HT on again, 61 C temperature. Then, rather than going step-by-step, I took a gamble, upped the voltage to 1.35V, kept the 16X multiplier, but set the FSB to 300MHz QDR or 1.2GHz throughput, and the memory to CL 3-2-2-6 at DDR2-600 dual-channel.
Well, it worked - at 4.8GHz! However, during the Intel stress test, the CPU test failed. I restarted the system with HT disabled, and all was fine - the tests were passed, and, well, the temperature was at 68C tops during the stress test. Good cooling job, it seems.
And, now, ladies and gentlemen, my (probably) last run of Sandra on an Netburst chip - look at the results... 4.8 GHz dual core, FSB 1200, in stable non-HT operation:
CPUint 54538 MIPS
CPUfp 64372 MFLOPs
MMint 25563 it/s
MMfp 12257 it/s
memInt 7537 MB/s
memFP 7538 MB/s
Now, well, other hacks have managed to push the 65nm NetBurst well beyond 5GHz and, in fact, some have pushed Conroe to almost the same GHz that I am running the hot Presler XE here. I am happy to say that, after few days with the Nautilus 500 prior to the Computex run, the system still runs fine, and I'll run more benchmarks on it, if it survives that long - in particular comparing it to the Conroe and AM2 stuff.
And the power bill? Ed, can we have an extra power supply bill allowance every time we test NetBust? Also, anyone else's experiences of this type? Let us know. µ