As we mentioned, our previous best on air cooling was an impressive 3.8GHz at 1.55V; this was our starting point using the Vapochill LS. Changing nothing other than the cooling system yielded a healthy 4.1GHz at about the same voltage. A 300MHz gain with a cooling improvement alone is no small feat! Unwilling to stop there, we forged ahead.
After a few hours of tinkering and some frustrating moments, the best we managed was 4220MHz at about 1.64V. This was a Prime95 large FFT-stable result. Though 4.3GHz seemed well within reach, the Phenom II required Vcore increase that would have overburdened our Vapochill; it was clear that 1.65V was the best the LS could do without diminishing returns. Using AMD’s nifty Overdrive application, we were able to further increase the core frequency to about 4.5GHz before we’d get spontaneous reboots. We’d guess that it would be possible to run most 3D benchmarks at around 4.3GHz and some single core benchmarks at about 4.4GHz.
When all was said and done, the Vapochill LS boosted our clock speed by another 420MHz over our best on air. Interestingly, we saw an almost identical core scaling on an old AMD Opteron 148 back in 2006.
The Vapochill LS certainly had a much hotter load on it than we saw back in 2006. Seeing the evaporator dip well into the -30s was an unfamiliar sight. Single core Athlon 64 processors from yesteryear simply didn’t shed this kind of heat.
Overclocking Observations
- Hyper Transport frequency had little impact on the system’s overclockability. We reduced it as low as 1000MHz (5x) and as high as 2000MHz (10x) with seemingly no impact. There have been reports that the HTT frequency needs to be dropped at very low temperatures (LN2 and DI potentially). This was not a requirement on single stage phase change.
- Northbridge frequency had little impact on the system’s overclockability. Once increased beyond a certain point, the system would not post, but we had no issues running a 2200-2300MHz clock speed. Lowering the northbridge multiplier did not yield better results.
- Voltage above 1.65V appeared to create a heat load that negated its overclocking benefit on the Vapochilll LS.
- Increasing the reference clock by a few points seemed to yield a slightly better overclock than using multiplier overclocking alone. This is counter intuitive, but it did make a positive improvement.
- Northbridge and HTT voltage increases did not improve the system’s overclockability.
- The MSI board we used frequently did not POST and required that we clear the CMOS. We hope that future bios revisions will correct this issue.
Conclusion
Booting the Vapochill LS after all these years certainly brought a smile to our faces, and we definitely feel that it can breathe some life into just about any processor. Through excellent condensation prevention measures and new socket kits, Asetek has kept the Vapochill potent and relevant over time. While we feel we could maintain our clockspeeds indefinitely, we’d love to see Asetek produce a newer unit tuned for modern quad core processors.
Although we realized a pretty significant gain through the use of our Vapochill LS unit, phase change is definitely not a magic bullet for the Phenom II. To be perfectly honest, we were expecting something more significant than our respectable 420MHz. All the news of 6GHz Phenom IIs on liquid nitrogen has had a hand in spreading unrealistic expectations. The truth is that these are extreme results with almost 2V on the core and load temperatures nearing -190°C. The processors used to attain these results are also very likely hand-picked for maximum results. Buyers need to remember that realistic 24/7 clock speeds are very different from bench-stable clocks.
Chips like the Phenom II keep clocking if they can be kept cold enough and fed enough Vcore. We said it before and we’ll say it again: AMD got 45nm right the first time, and the Phenom II is a great overclocker. In comparison to some of the first Phenom processors, the Phenom II is simply a monster overclocker!
Special thanks go out to Astek for providing us with the socket AM2 kit that made this testing possible!
Pages: 1 2