Soltek QBIC EQ3901 SFF barebones system

Complete|Mobile Systems
Viewing page 6 of 6 pages. Previous 1 2 3 4 5 6

TESTING

The following tools were used during testing:

Testing consisted of an examination of how the system behaved acoustically in response to changes in thermal load. No performance benchmarks were run; performance in a SFF system depends mainly on the components it is tested with, with rarely more than ~3% difference in performance between models if the same components are used.

Ambient temperature during testing was 20°C. Ambient noise level was 19 dBA/1m. Cool 'n' Quiet was enabled during all testing.

FAN (MIS)BEHAVIOR

The fan controller in the EQ3901 did not impress us. Rather than the quasi-linear fan control that we've become used to in other SFF systems, each fan has just two modes: Low and high. Transitions between the two modes are not smooth and occur almost instantaneously. The time lag that exists is probably a result of air resistance, not a gradual change in input voltage.

There are four possible noise signatures depending on how fast each fan is spinning. How frequently each one is heard depends on the fan settings in the BIOS.

Setting the Blower fan threshold for 40°C kept the blower in the low state permanently, even under full CPU load. However, with the CPU under load, we were unable to keep the CPU cool enough to prevent the heatsink fan from ramping up.

It is inevitable for the heatsink fan to increase in speed under heavy load. Leaving the CPU under load revealed another undesirable characteristic of the fan controller: The heatsink fan cycled between high and low speed as the CPU temperature changed in accordance with greater or lesser airflow. The cycle could be delayed by raising the thermal threshold, and the period of the cycle could be increased by increasing the Fan Tolerance value, but we were never able to prevent the cycling effect altogether.

An example: With the fan threshold set to 60°C and the fan tolerance set to 1, placing the CPU under load did not affect fan speed until the CPU temperature hit 62°C. At this point, the CPU fan sped up to full speed. It remained at full speed the temperature dropped to 58°C. With these settings, the CPU temperature would cycle between 58°C and 62°C, with the fan changing speed roughly once every two minutes.

Changing the fan tolerance value to five caused the temperature to cycle between 54°C and 66°C, thus increasing the period of the cycle to approximately one change every five minutes.

Theoretically, if the loosest settings were used to control the fans — a 70°C threshold and a tolerance of 5 — the cycling could be eliminated if the CPU temperature could be kept below 76°C at all times. This would only be possible with a cooler CPU or a more effective heatsink.


CPU temperature — and heatsink fan speed — cycles under load according to the BIOS fan settings

NOISE MEASUREMENTS & ANALYSIS

Soltek QBic EQ3901-300P Noise Characteristics
Fan State
CPU Heatsink Fan
1300 RPM
3900 RPM

Exhaust
Blower

1200 RPM
33 dBA/1m
39 dBA/1m
1800 RPM
41 dBA/1m
43 dBA/1m

Both fans low: At idle, the Soltek is far from quiet. Some modification is definitely required before this would be acceptable to SPCR regulars. Fortunately, reducing the idle noise would likely be an easy mod: The primary source of noise is the blower, which is easily removable. The blower produces a low frequency buzz that is quite loud. A low hum from the heatsink fan and some background chassis resonance is also present.

Blower low, HSF high: This configuration is the second quietest. However at 39 dBA/1m, second quietest is only a nominal honor; this is noisy, there's no way around it. Because the heatsink fan is spinning at almost 4000 RPM, noise comes from all sources: The motor gives off a highly resonant mid-band drone. Turbulence noise is also evident.

Blower high, HSF low: The blower maxed out at a relatively low 1800 RPM, and the buzz that was so apparent at idle is now higher in pitch and a full 10 dBA/1m higher in volume. For those of you unfamiliar with the decibel scale, subjectively, that's roughly double the noise.

Blower high, HSF high: This level is achieved under load whenever the exhaust fan threshold is anything other than 40°C. Describing the noise at this level is moot; suffice to say that the system is noisy. Combine the noise quality from each of the two previous two configurations and you should have a good idea of what it sounds like.

At no time did the power supply become a major source of noise. Admittedly, it would need to be quite loud to be heard over the other two fans, but given that the majority of the heat produced in the system was exhausted through the power supply, we were pleased to note that it didn't ramp up enough to add to the overall noise level.

AUDIO RECORDINGS OF THE TEST SYSTEM

MP3 recordings were made with the system in each of the four possible noise states. One recording features the transition between two states: The CPU fan can be heard ramping up at the beginning of its cycle.

MP3: Soltek QBic EQ3901-300P, Blower Low, HSF Ramping from Low (33 dBA/1m) to High (39 dBA/1m)

MP3: Soltek QBic EQ3901-300P, Blower High, HSF Low: 41 dBA/1m

MP3: Soltek QBic EQ3901-300P, Blower and HSF both High: 43 dBA/1m

Comparisons

MP3: Arctic Cooling Silentium T2 with test system (3.5" HDD suspended) at idle - 23 dBA/1m

MP3: Arctic Cooling Silentium T2 (3.5" HDD suspended) at max load - 34 dBA/1m

MP3: Shuttle XPC SB86i with Samsung Notebook Drive, Idle: 29 dBA/1m

MP3: Shuttle XPC SB86i with Samsung Notebook Drive, Load: 33 dBA/1m

SPCR MP3s: HOW TO LISTEN & COMPARE

The recordings above were made with a high resolution studio quality digital recording system. The microphone is 3" from the edge of the fan frame at a 45° angle, facing the intake side of the fan to avoid direct wind noise. The ambient noise during all recordings is 20 dBA or lower.

A quick and simple way to use these recordings for valid listening comparisons is to play the quietest recording on only one speaker (or a pair of headphones) and set the volume so it is just barely audible a meter away. You must turn off any special sound effects, and set equalizer / tone controls to neutral or flat. Don't touch the volume setting afterwards, and use the same one speaker when you listen to any of the other files. The end result will be reasonably close to the actual recorded sound levels.

Here is a recording of a very quiet sound that is barely audible from 1 meter away even in a super quiet room.

For full details on how to calibrate the playback level of your sound system to get the most valid listening comparison, please see the yellow text box entitled Listen to the Fans on page 3 of the article SPCR's Test / Sound Lab: A Short Tour.

POWER & THERMAL CHARACTERISTICS

Activity
AC Power Max
CPU Temp
Idle (CnQ Enabled)
49W
38°C
Full Load: CPUBurn
96W
See comments in Fan Behavior section

The AC power draw for our modest system peaked at 96W. This is nearly 10W less than obtained with a similarly configured Shuttle SN95G5, another A64-939 SFF. That system was tested with just one 512mb stick of RAM, instead of two, but the number still suggests a slightly more efficient power supply. Stressing the VGA card as well will probably add a further 10-20 watts to the total.

Our test system did not remain stable under load. After about 40 minutes of continuously running CPUBurn, we noticed that switching between windows seemed to be a little glitchy. Gradually, we began to get visible artifacts that remained on the screen. Before we could restart the computer to fix the problem, the system had frozen and we had to manually power it down.

CPU temperature was between 55°C and 65°C, well below the threshold for CPU instability. We therefore attribute the freeze to our passively cooled VGA card wedged against the side of the case. Given the already limited airflow in the case, our fanless card probably overheated.

A fanned VGA card probably would not have this problem. However, VGA fan on cards are notoriously difficult to silence, generally requiring large, double-size heatsinks that are unlikely to fit in the confines of a SFF case. Furthermore, because the EQ3901 does not come with integrated graphics, it is necessary to have a VGA card. If a fanless VGA card is not an option, this means adding a another source of noise to the case.

CONCLUSIONS

The Soltek QBic EQ3901 is not without some bright points. It looks good, has many useful features and in/out capabilities, and accommodates socket 939 A64s, the current best value / power processors by quiet enthusiast standards. Few other SFF systems offer an extra 5.25" bay to allow for hard drive suspension. The included power supply is modern, quiet, conforms to a standard form factor, and is manufactured by a company with a good reputation. The use of a standard mounting system for the heatsink allows the use of many aftermarket heatsinks.

The heat related problems that we experienced are specific to our setup, and probably represent an isolated problem related to fanless VGA cards. Silencing a PC requires balancing an acceptable level of system heat with a minimal number of noise producing components. Our choice of VGA card may have failed to achieve this balance, although the failure did occur in a high stress test environment, not during actual use.

Still, we cannot recommend the Soltek QBic EQ3901. It is far from quiet in stock form.

Making a quiet SFF system is not easy by any means; we have yet to review a sample that could not be improved in some way. The standardized parts used in the EQ3901 are easily replaced — a trait not shared by many of its competitors. So, silencing it might be fairly straightforward.

The exhaust blower could probably be removed safely; the amount of airflow that it provides through the rear exhaust vent is minimal. The CPU heatsink (or even just the fan) could be easily swapped for a quieter one; its socket 478 mounting system opens up a huge array of options in the heatsink market. Although the BIOS fan controller is probably not usable, it is easily bypassed by manually undervolting the fan. Admittedly, such a system would be restricted to low loads or slower A64s only; sustained loads run the risk of overheating. More ambitious modders might consider solving the problem of airflow by cutting an 80mm exhaust hole and adding an exhaust fan in the back of the case.

Much thanks to Soltek for providing the QBic EQ3901M sample.

* * *

Discuss this article in the Silent PC Review Forums.



Previous 1 2 3 4 5 6

Complete|Mobile Systems - Article Index
Help support this site, buy the Soltek EQ3901 SFF barebones system from one of our affiliate retailers!