Bitspower NP60CS - click images to enlarge
Bitspower NP60CS
Taiwanese manufacturer Bitspower
are probably the best-known of the 3 seen in this article. They
make a big deal of their "skiving" technology, which
consists of folding thin fins up from a block of metal. This means
there is no physical junction between the base and the fins, which
is good for thermal conductivity into the fins. This particular
unit uses this technique to form the curved fins from the solid
copper base, making for a surprisingly hefty little cooler. The
5500rpm 60mm fan is mounted onto a metal shroud which is screwed
into the base. Noise-wise, the fan is louder than I would expect
for a 5500rpm unit. It's not as bad as a 7000rpm, but still a
fairly loud whine. The finish of the base is not bad, but not
the mirror finish seen on the Fanner units. There are fine machining
grooves in the base that will benefit from being filled with thermal
paste or more likely being lapped (this method would probably work fine). Bitspower
provided some generic silicone paste, but their website indicates
that they now provide Arctic Alumina paste instead. For our testing
we of course used Arctic Silver rather than any of the supplied
pastes. Attaching the unit to a socket is no great drama, again
providing a small flat-bladed screwdriver is used, but the clip
is extremely strong which can be a bit unnerving.. tight contact
is good for thermal transfer, of course.
Bitspower NP80D
Comprised almost entirely
of aluminium this large cooler is very light. The same "skiving"
technique is used to carve the curved fins from the solid base
which has a similar, but finer, machined surface. It's wider across
the socket and is the only cooler in this roundup to have a hands-only
mounting clip - there's no slot to use a screwdriver when attaching
it. I find the flexible end on this kind of clip very fiddly to
use, and on motherboards that have that end of the socket along
the top edge of the motherboard, in some cases the proximity of
the PSU makes it a real hassle to install or remove the cooler.
It uses the same fan as the previous Bitspower cooler and, again,
the sound is unusual for a 5500rpm - not really louder than normal,
but it has a high-pitched edge to the sound which can be quite
irritating. Perhaps the shape/density/proximity of the heatsink
fins is responsible for this, in the same way that the Delta 7000rpm
seems to be most annoying when fitted to a ThermoEngine. The bottom
surface again has light machining marks on it but is fairly smooth
and overall, as with all the coolers in this roundup, build quality
seems pretty good.
Results:
As discussed on Page 1, a
modified A7V was used. The under-CPU thermal probe was connected
to a MacPower DigiDoc5 and another probe was used to sample
the ambient air near the case. Testing consisted of a 3DMark2001SE
loop - 2 full runs of the benchmark at 640x480 resolution - one
in 16 bit and one in 32bit. This provided about 15 minutes of
CPU load for the cooler to reach a stable maximum temperature.
To make sure no CPU cycles went idle, I also had the Folding@Home console client running in the
background. Ambient was between 22.5 and 23.2C for the duration
of testing and the exact ambient temperature was recorded at the
same time as the temperature of the under-CPU probe at the end
of the two 3DMark2001SE runs.
The CPU used was a lowly Duron 700MHz, but running at 900MHz with a hefty 1.9v (reported by MBM5) core voltage. Why not a faster CPU? We've seen more than our fair share of CPU's killed during cooler testing (see here and here for examples) so I hope you'll forgive me if I don't break out the tray of 2100+'s for this comparison. Once we work out the thermal efficiency of each cooler, we can calculate the approximate temperature that higher CPU's would reach.
Here's the real-world results:
Mathematical Manglings
The sophisticated readers of today want to see C/W in their cooler
reviews. Historically we haven't provided these because we haven't
had any incredibly accurate ways of calculating them. Our testbed
have always been fairly "real-world" ones, with the
vagaries and margin for error that that implies. I'm comfortable
reporting my experiences and hoping that the audience will receive
them as projections to be referenced against their own experiences,
rather than absolutely accurate and unquestionable facts. However
this time I thought I'd experiment along the C/W road and see
where it leads me.
C/W is, of course, a measure of the thermal efficiency of a cooling system. Simply, it's the temperature rise in Celcius (C) for each Watt (W) that the object being cooled is emitting. You calculate it by taking the recorded temperature of the cooled object, minus the ambient temperature (so, the change in temperature, or Delta T).. divided by the power in Watts that the cooled object is putting out. Note that there is huge potential for error in the way we are gathering the information for this calculation, so our results are to be taken as an approximate projection only. Radiate, a program which calculates these things and can be grabbed from here, reckons that my Duron is putting out about 56.9W of heat. I have no way to verify this, so everything from here on in has an even greater probability of error than before. The temperature is measured as described, which of course isn't exactly the temperature of the cooled object.. but fairly close to it. And the ambient is measured, not inside the case where the cooler is sucking air in, but from the outside of the case. The case is so oriented that hot air can rise directly out of it, but again there'll be a small variation there. A final note is that our testbed uses the mounting clip for the cooler, whereas many other websites use an artificial testbed which doesn't use the mounting clip to hold the CPU on. That gives them a more accurate picture of how efficient the actual heatsink is, but our way gives a more accurate idea of how the cooler will perform in the real world. Clamping pressure has a big effect on cooling performance and the most efficient cooler in the world isn't going to do much for your CPU if it's held on with the equivalent of a couple of old rubber bands. So, disclaimers aside, this is what we get for C/W:
Those numbers seem awfully low. I'm thinking either Radiate is wrong (too high) about how much power the CPU is putting out, or we are really inaccurately recording the core temperature, and getting it way too low. One major factor in our getting lower temps than other sites is, I guess, that we're measuring from the back of the CPU, so there's a chunk of insulating ceramic between our probe and the heatsink/core. Also, I know 3 decimal places is overstating the accuracy, but otherwise Excel has the top two values rounded to .46 but the bars at different lengths, which looks weird. So anyway. Let us persist with our increasingly irrelevant and hypothetical exercise, and try to project an approximate temperature from a higher-wattage CPU. To do this we'd take our prevously calculated C/W, multiply it by the CPU's heat output in W, then add the ambient temp in C. According to AMD's Thermal, Mechanical and Chassis Cooling Design Guide, the most power (as at March 2002) a SocketA CPU should put out is 72W. Assuming an ambient temperature of 23C and our previously-calculated C/W values, this gives us:
These first struck me as being too low, but when you consider that the ambient temp used in this projection is quite low, it's perhaps not too unrealistic. Also, this is a projected temp for the back of the CPU, so the core will of course be higher. But remember, this is only a projection - so don't place too much faith in it. However as a general guide I think it lets us draw some conclusions about the respective coolers..
Conclusions:
Of these coolers, I'd only really recommend the Bitspower NP80D
and the Sibak AC-02-625B for high-end CPU's. Mid-range CPU's (up
to about 1.4GHz) would probably be served well by any of them.
Be aware that the Sibak, while out-performing them all, also has
the most beefy and thereby most annoyingly loud fan of the lot.
It does seem to be the only one that would cope with a high-end
CPU under sustained load, in a range of normal ambient temperatures.
However, I have to confess I'm growing tired of the "take
a chunk of metal and whack a huge fan onto the top" mentality
of cooler design lately. I know the combination of high-wattage
CPU's, weight/space limitations and cost point towards this approach,
but I'd really like to see a little more imagination - and less
noise - from heatsink vendors. However, all three manufacturers
represented today seem to have produced quality products, each
with features that make them if not unique, at least unusual and
showing promise for future designs. Bitspower's skiving technique
seems a great way to remove the thermal junction between base
and fins. Sibak's coolers, while perhaps borrowing on ideas we've
seen before, are experimenting with non-traditional designs with
some success. Fanner's use of low-profile, quieter fans suggests
they share my concerns about cooler noise. I think they've all
played a strong opening hand and I look forward to seeing more
from them in the future.