Viewing page 5 of 5 pages.
Previous 1 2 3 4 5
TEST RESULTS
ON THE 6800GT
The cooler was first installed on an AOpen
Aeolus PCX6800GT-DVD256 — a medium-powered video card that features
a thermal sensor build into the GPU core. The accuracy of the sensor is unknown,
but it is good enough to make relative judgments between the various coolers.
Past tests of VGA coolers have also been installed on this card.
This GeForce 6800GT has stood up to a year's worth of abuse around the SPCR
lab.
Ambient conditions during testing were 16 dBA and 22°C. Tests were run
with the system fan at 12V, 9V, and 7V.
Thermal testing consisted of running two instances of CPUBurn and the
artifact scanner built into ATI Tool simultaneously to generate as much
heat as possible. An initial test was run with the system fan running at 12
volts, and then the fan was progressively slowed down to make the thermal environment
more difficult.
Once the temperature on the card had stabilized, the stress software was
left running for at least another 20 minutes while we watched the screen carefully
for visual artifacts that might indicate overheating. The last test, with
the system fan running at 7 volts, was left running for more than an hour.
Even during our most strenuous test, the Condor never allowed our test card
to get hot enough that there were visual artifacts on the screen.
|
Aerocase Condor Test Results (GeForce 6800GT)
|
|
System Fan Voltage
|
GPU
|
VGA Ambient
|
CPU
|
AC Power
|
System Noise
dBA@1m
|
|
12V
|
84°C
|
76°C
|
65°C
|
229W
|
26
|
|
9V
|
88°C
|
81°C
|
67°C
|
232W
|
25
|
|
7V
|
90°C
|
84°C
|
70°C
|
232W
|
24
|
|
Note: The PSU fan ramped up a bit under load to bring the system noise up ~2 dBA@1m over the 22dBA@1m minimum idle noise of this system with case fan at 7V.
|
Dropped into place on our testbed, the Condor did rather poorly in comparison
to other heatsinks that we've tested. This is excusable, since no
other passive heatsinks have been tested in this system. However, it's core
temperature that counts, and the Condor only barely kept our GeForce 6800
GT below 90°C. That may be good enough for our card, but it doesn't say
much for the Condor's potential to cool hotter cards like the X1900XTX, at least not in our low airflow case/system.
Our standardized test system doesn't really give the Condor a fair
chance. Passive cooling requires careful attention to system airflow, and
our test system is not designed to circulate much air in the bottom portion
of the case. The standard edition of the Condor that positions the heatsink
over the CPU cooler probably would have done better. We chose this "reverse wing" sample to ensure that the cooler could be used even with large tower heatsinks; in retrospect, perhaps both variants should have been tried.
ON THE X1900XTX
Aerocase's "Preview" Test
Aerocase seems to have fallen into the same trap that we did in their
test of the X1900XTX. That test used 3DMark05 to stress the card,
but our own testing showed that none of the tests in 3DMark05 managed to
stress the card any more than ATI Tool did. Tapping the full potential of
the X1900XTX required the SM 3.0 test in 3DMark06 — a much tougher
test. |
It was with some trepidation that we screwed the Condor onto the X1900XTX
— well known as the block heater of graphics cards and said to consume
over 120W peak under load. In addition to the Condor, two other heatsinks were
tested on the X1900XTX: The stock ATI cooler and our current favorite aftermarket
cooler, a Zalman VF900.
Midway through the first stress test, we realized that something was wrong:
The AC power consumption was too low. That's significant. Power consumption
is the best measure of how much heat is generated by a system, and our test
rig was consuming about ~230 watts — almost exactly the same as our previous
tests. We had the hottest gaming card on the market and it wasn't
delivering. What could be wrong? A careful examination of our test methodology
revealed that the software used to generate load — the artifact scanner
embedded in ATI Tool — did not stress the GPU enough to heat it up fully.
Several days of research and experimentation revealed a new candidate for
stress testing: The first SM 3.0 demo in 3DMark06.
This demo pushed the system's peak power consumption to almost 300W, although
the average during the test was typically 15~20W lower. Tests were run at
1280 x 1024 resolution (oddly, higher resolutions caused power consumption
to drop) with 16 sample anti-aliasing and 6x anisotropic pixel shading enabled.
The SM 3.0 demo was set to loop indefinitely during the test, and temperature
readings were not made until three loops in sequence generated the same peak
temperature.
This stress test is not continuous; each loop required about
20~25 seconds of load time when the card was not being stressed. As a result,
the core temperature fluctuated by almost 20°C between each loop. On average,
each loop took about two minutes to complete, about a quarter of which was
spend with the GPU in idle.
|
Aerocase Condor Test Results (Radeon X1900XTX)
|
|
Cooler
|
VGA Fan Voltage
|
GPU
|
VGA Ambient
|
Peak AC Power
|
CPU
|
Noise of cooler fan
dBA@1m
|
|
Aerocase Condor
|
–
|
105°C*
|
63°C
|
306W
|
68°C
|
N/A
|
|
92mm fan @ 12V
|
86°C
|
58°C
|
293W
|
67°C
|
24
|
|
Stock ATI Cooler
|
5V
|
105°C*
|
70°C
|
300W
|
65°C
|
37
|
|
12V
|
69°C
|
54°C
|
280W
|
62°C
|
51
|
|
Zalman VF900
|
5V
|
86°C
|
57°C
|
291W
|
65°C
|
20
|
*Test halted when ATI's thermal throttling
interrupted the test.
Note: The minimum system noise at low load with the CPU fan, case fan and PSU fan measured 22 dBA@1m. Under full load testing, the PSU fan ramped up to bring the system noise (without VGA card cooler fan) up to ~30 dBA@1m. Under such load, only the stock ATI cooler could be heard distinctly over the rest of the system noise. |
In the terrible thermal conditions of our test rig, the Condor was not able
to cool the X1900XTX to our satisfaction. In fact, ATI's thermal throttling
kicked in just as we were about to conclude that the temperature was stable at 105°C.
The power draw that we saw during the test was ~15W higher
than our other tests — a sign that the power circuitry on the board was
headed towards thermal runaway.
On the other hand, the stock cooler fared no better when undervolted to 5V
— and it was still far too noisy at 37 dBA@1m. Presumably, the fan voltage would
have been closer to 12V had we let the card control the stock fan. The 12V
test (51 dBA@1m) shows the kind of cooling that could be expected if this
were allowed to happen. The 20W (!) spread in power consumption between the
two fan speeds shows just how necessary that extra cooling is.
Some Extra Cooling
Aware of the limitations of our test system, we decided to try again, this
time with an additional fan taped on the floor of the case just in front of the Condor blowing towards
the back. We used a 92mm Scythe S-Flex fan, running
at 12V. This fan (which measured 24 dBA@1m) added very little noise to the system, especially under load
when the fan in the power supply ramped up.
The change was dramatic. The GPU core dropped 20°C to 86°C, equaling
the VF900 at 5V. More importantly, the card no longer overheated. That's right,
the hottest graphics card on the market, cooled by a passive cooler. Well,
not quite; the extra fan is cheating a bit.
CONCLUSIONS
Cooling an X1900XTX is a tough job under any circumstances, and
doing it quietly is an exceptional feat. We are confident that, if attention
is given to system airflow, the Condor is capable of cooling any card on the
market.
That said, we do not recommend using the Condor with an X1900XTX
because we do not recommend using a card that hot in any quiet system. The extreme
amount of power sucked up by the X1900XTX is almost certain to cause any power
supply fan to speed up. Even the Antec Neo HE in our test system (ducted with
fresh air to keep the fan speed down) ramped up; no other fan-cooled power supply would fare better.
There is little point in buying a passive cooler for a card that will increase
noise elsewhere in the system.
Because of the need for good system design, the Condor will
perform best in the hands a of meticulous, experienced user who understands how to
optimize airflow in a PC. It has excellent cooling potential, but it cannot just
be dropped into any system with good results. As we discovered in our early
tests, putting the cooler in an area with little airflow cannot result in good cooling performance. On the other hand, if the Condor can
share airflow with other system components, one less fan is needed and noise is reduced a bit (or a lot) further.
For all the customization and lack of mass production and distribution,
the price of the Condor is surprisingly modest. US$60 is not unheard of
for a high-end cooler. The asking price of the excellent Zalman
VF900 is as much as US$115, although the wonders of e-tailing keep
the average price around US$40. An extra $20 is not a bad price to pay for the
jump from low-noise to fully passive. On top of that, the unusual honesty and
frankness that is apparent on Aerocase's web site is impressive. Personally,
I am willing to pay a little extra for the knowledge that a company is well
run, and I'm probably not alone.
It's questionable whether the Condor will see the kind of sales
volume of well-established competitors like Zalman. Considering Aerocase's distribution and sales
model, it's uncertain whether they are pursuing that kind of goal. The Condor
is a niche product, and it fits that niche nicely.
Many thanks to Aerocase
for the sample of the Condor and the X1900XTX loaner, and to AOpen for the
samples of the VGA card and the motherboard in the testbed.
*
Editor's Counterpoint: If you wanted to quibble, you could ask where is the vaunted high technology and engineering of this aerospace design team? It's practical and it works well, but toothpicks?! The Condor is just two lengths of aluminum extrusion glued to heatpipes that join to a basic copper plate. It's the kind of thing that might show up as a DIY project by one of SPCR's dedicated silencers. As Devon mentioned earlier, it's a brute-force approach, hardly subtle... but admittedly still kind of elegant in its apparent simplicity.
My guess is that the company's response would be to say, "Go
ahead and DIY." I suspect it could cost you several times Aerocase's asking price in materials and time (especially the latter with trial and error runs), and if the final product doesn't work well, you may end up with a useless once-very-expensive video card with no one to blame but yourself. Also, the number of PC enthusiasts with the skills, tools and time to pull off such a cooler is extremely small.
*
SPCR Articles of Related Interest:
Fanless PCIe Graphics Cards
from Asus and Aopen
Gigabyte GV-N66256DP Fanless AGP video card
VGA Cooler Roundup: A
Thermalright, two Zalmans, and an Arctic Cooling
Sytrin KuFormula VF1 Plus Graphics Card Cooler
* * *
Discuss
this article in the SPCR Forums.
| Help support this site, buy from one of our affiliate retailers! |
|
