Glaciator, Swiftech 462A, Thermalright SK6 Performance Tests

SUMMARY: A different approach to heatsink testing – how well does a heatsink “push the envelope”.

One issue that comes up in the testing we do is how a user can replicate our test results. The short answer is you really can’t unless you buy some expensive gear – not a real option.

I had played around with the idea of a “performance” test a while ago and spent a lot of time this week trying it out. Basically, the approach is to find the highest stable speed a given CPU/mobo combo can run with watercooling, and then see how close an aircooled heatsink can come to it. The premise is that watercooling will give the best cooling performance which will result in the highest stable speeds.

This becomes more pertinent as heatsinks converge on C/Ws in the 0.15 to 0.20 range – at these C/Ws, the absolute temperature difference among various heatsinks are down to 5 C/9F at 100 watts. Considering the variety of systems and local temperature variations among users, differences among users may not be that apparent.

Current heatsink technology is hitting a wall – don’t expect to see much better temperature performance from heatsinks than a C/W of about 0.15 – there are limits. New technologies may push this towards 0.10 – MAYBE – but even at that level, we’re still looking at possible CPU case temps of 45-50C for “normal” users.

To ascertain heatsink performance “at the limit”, I ran a series of tests with using the Glaciator, Swiftech’s 462A with the Sanyo fan and the Thermalright SK6 with a Delta 38 fan. I think these three represent the best that are out there, taking three different approaches to CPU cooling.

The Glaciator is designed for high cooling performance at relatively low noise levels, the Swiftech 462A high performance at moderate noise levels, and the Thermalright for high performance at high noise – all too typical of the “brute force” approach we are seeing. It will take another generation of heatsink design to couple high performance with tolerable noise levels.

For this test, I used a T-Bird on the WIll KK266*. I first ran Prime 95 at the highest possible speed using watercooling. My test criteria was that Prime 95 had to run for one hour without failing.

Then I installed the heatsinks and started at the watercooling speed, running Prime 95. If Prime 95 failed after repeated attempts to run it for one hour, I then backed the speed off by backing down FSBs one MHz at a time, repeating the testing until I could get a stable Prime 95 for one hour.

From this table, you would conclude that, all things being equal, the Thermalright SK6 performs better than the Swiftech and Glaciator. The actual performance difference is minimal – 2% in FSB speed (36 MHz), not noticeable to the user.

I recorded all temps with the following results:

¹ This is the temp measured in the base of the block; CPU case temps are higher by an estimated 5 C.

Interpreting C/W: For every watt the CPU radiates, the heatsink will cool the CPU Case Top by the (C/W x watts) plus ambient temp. For example, at an ambient temp of 25 C, a C/W of 0.25 with a CPU radiating 50 watts means that the CPU case temp will be 50 x 0.25 = 12.5 C over ambient temp, or 37.5 C. The lower the C/W, the better.

The results from temperature testing correlate to the performance results shown above. However, the difference between all three is 4 C, even though MBM temps show a 6 C swing. Users performing the same tests in their own systems may in fact show different rankings due to variations in their own system. For example, a system with “back of the motherboard cooling” could well see a different pattern.

What I found most interesting was the difference in CPU back temps – the Thermalright showing the smallest variance (7.1 C) with the Glaciator and Swiftech showing larger variances (11.9 and 15.2 C respectively) (CPU back Temp was measured by a thermocouple on the CPU’s center back). Large variances from the front to back temps have been associated with less stability.

Performance testing of these three high-performance heatsinks vividly illustrates how close each is in delivering exceptional CPU cooling performance. The perceived difference at the user’s desktop is narrowing such that, for all practical purposes, closely rated heatsinks are interchangeable.

The diversity of these three heatsink designs impacts user’s purchasing decisions:

• If performance at any price or noise level is your objective, the Thermalright SK6 tops the list;
• If performance at tolerable noise levels and mounting security are your objectives, the Swiftech 462A tops the list;
• If performance at reduced noise levels is your objective, the Glaciator tops the list.

You can build your own list and determine what best suits your needs. However, I think you will find more heatsinks in the 0.20 to 0.15 range hitting the market, and purchase decisions will be more bounded by a range of features than performance measures. Users will also find that local differences in factors such as case design, case airflow, computing intensity, etc., will impact observed results.

*I thought this board died, but apparently what happened is the board got so hot in this testing that it tripped a thermal protection circuit. It takes some time to reset, and after a day it was OK. First time I ever had that happen!

Email Joe