In this page, the results of the tests are presented.

Several problems came up when testing the power consumption of the boards and computer systems present in the Locolab. In order to correctly measure the power consumption, the power supply needs to be the same for all equipment measured. However, the Intel Icbook comes with it's own PSU, and the Ebox 4300 also comes with it's own powersupply. We decided to use the alternative powersupplies, as the efficiency will not be dramatically different, as shown in the paragraph of the powersupply testing. The second thing we tried to keep uniform was the harddisk, which also proved to be impossible as some boards only have SATA connections, and others only have IDE connections. Since this was a specific issue with the Intel D201GLY board (which does not come with a SATA connection) we decided to do two tests with the Intel 945GCLF: One test with the SATA 2.5” disk we used in the other tests, and one test with the 3.5” IDE disk we had to use in the test with the D201GLY board. The difference in the two measurements is subtracted from the results of the D201GLY.

In the table below the power consumption test results are presented.

The outcome is not very surprising: The VIA C3 is obviously the lowest power consuming, and the Intel boards consume the most. The Intel Atom board is only slightly less power hungry than the D201GLY. This comes as a suprise, since the Atom platform is launched as being low power. The reason for the relative high power consumption of the Atom board is because the Intel 945 chipset uses a lot of power. That is also the reason that the chipset has a fan, and the CPU only has a heatsink.

The results of the ten tests we have done are presented below. The lowest scores are presented in italics, while the highest scores are presented in bold letters. Since the Intel D201GLY only supports IDE, and the rest of the testing has been done with a 2,5” SATA drive (at 5400 RPM, while the IDE disk runs at 7200 RPM) an extra benchmark test has been done on the Intel D945GCLF with the IDE disk as this board is performance-wise the closest to the D201GLY.

Some very interesting conclusion can be drawn from these results.

First of all, the difference between the tests on the Intel D945GCLF with a 2,5” SATA disk at 5400 RPM and the 7200 RPM IDE disk are negligible. This makes the systems with either disk comparable The big winner here is the Intel D201GLY. It has the best score on 7 out of the 10 tests. Second best is the Intel D945GCLF with the new Atom CPU on it. One note here however: The Atom CPU has hyperthreading capabilities, and not all the tests are hyperthreading aware. 7zip and WinRAR are optimised for hyperthreading, and that is reflected in the results. It became even more apparent when we did the Blender benchmark again. Blender allows the user to manually set the numbers of threads which are used. The result was 7:03 minutes! An obvious winner. But many desktop applications are NOT hyperthreading aware, so differences in performance are not noticed when a hyperthreading capable cpu is used in office or school environments. So, the best performing CPU would technically be the Atom CPU, but realistically, the Intel Celeron CPU is the winner (albeit with a slight margin). The worst performing CPU is the VIA Eden 500 Mhz. Of course, this is caused by its' low clockspeed. The VIA Esther CPU's perform clearly worse than the Intel Atom or Intel Celeron CPU. This makes Intel the performance winner.

In the table below are the results shown of our usability tests. These tests have subjective outcomes, and as such, the conclusions could be interpreted differently!

The OpenOffice test underlines the conclusions of the CPU benchmarking test: The Intel D201GLY wins again. The Ebox is the slowest (as it only has a 500 Mhz. Eden CPU). The Atom is second again. The HD video playback benchmarks shows some very interesting results. The clear winner is the D945GCLF board. The Intel 945 chipset clearly shows some performance gains here! The only CPU other that the Celeron and Atom capable of playing back an HD movie is the VIA 1.5 GHz CPU with the VIA CN700 chipset. However, the image is sloppy, so we wouldn't recommend for HD use. The Google Earth test shows that all hardware is capable of using the software, again with the faster machines it works smoother.

The power supply test was quite straightforward and the results are presented in the table below:

These were the powersupplies we have tested:

The fanless powersupplies have the best efficiency. The best powersupply is the Pico PSU. This is also the most expensive one.

Selecting hardware is not a straightforward task. There are many factors which have to be taken into account, such as purchase costs, running costs, performance, usability and availability Giving a straightforward conclusion is not possible, since for every user the factors listed above have another significance. In this conclusion, each of the above factors will be analysed.

The purchase costs of hardware are related to the scale of deployment. Buying 30 PC's or 3000 PC's can make a huge difference. Sometimes higher costs mean lower powerconsumption. The Pico powersupply, which costs around $60,- is a perfect example of this. When talking about motherboards, more costs can mean better support, or better warranty. It does not always imply better performance. In fact, our cheapest board (the Intel D201GLY) has the highest performance! But our most expensive board, the VIA 1000 Mhz, is fanless, which can have considerable advantages in rural classrooms. Purchase costs have to be carefully measured against the capabilities and other benefits.

The only running costs this research has investigated are the electricity costs. Maintenance costs have not been taken into account yet. This researched only looked into low power computers and hardware combinations, so even the biggest electricity consumer is nothing compared to what is on normal office desks. To give an example: The computer under my desk, a Pentium 4 3.2 Ghz uses 80 Watts when idle, as compared to around 30 Watts as highest for our measurements! The hardware has been put in three different qualifications: High consumption, medium consumption and low consumption. Low consumption means 100 % cpu usage below 15 Watts, medium consumption means 100 % cpu usage between 15 and 25 Watts, and high consumption means a 100 % cpu utilisation above 25 Watts.

Some remarks are in place here though. We need to do extra measurements!! HOWEVER, CONCLUSION IS VALID

The conclusion is here that the Intels use the most, VIA CPU's 1 Ghz and 1,5 Ghz are somewhere in the middle, and the 500 Mhz CPU's are using the least Watts. This is in line with their performance, as shown in the performance paragraph.

When selecting hardware, getting a fanless (or “open”) powersupply will enhance the efficiency of that powersupply. A 2,5” harddisk can also significantly reduce the amount of power used.

The performance results are in line with our expectations. The Intel CPU's have the best performance, the VIA 1,0 and 1,5 Ghz are somewhere in the middle, and the 500 Mhz Eden CPU is on the lower end. This is in line with the power consumption, which makes a lot of sense. What is surprising though, is the better performance of the Intel D201GLY with it's Celeron CPU: Intel's marketing machine boasts the performance of their new Atom CPU, but its' performance is less than the Celeron CPU! Of course, the hyperthreading capabilities have not been used in most of our tests, and it performs better with software optimised to take advantage of hyperthreading.

Usability is a tricky concept. All the hardware we tested was usable. The fact that even an application like Google Earth was running on all machines shows that they all could theoretically be used in classrooms. The current Cambodian curriculum consists of mainly office applications. Although it takes more than 30 seconds to open an OpenOffice presentation on the Ebox with it's 500 Mhz CPU, that is still acceptable, especially in a school environment. However, not all applications run smoothly on the slowest machines, and if you want to have a better experience, and be able to run more demanding software, it is better to go for at least a C7 Via Eden CPU. On the fastest end are the Intel CPU's.

Availability of hardware is of course a matter of scale. Currently, it is very difficult to obtain low cost hardware in computer shops in Phnom Penh. The only motherboard available is the 1,5 Ghz VIA brandless board. Intel also has an office in Phnom Penh, making it easier to obtain Intel boards. Small scale deployments will be difficult when using local shops. However, large scale deployments can be a lot easier, since the buyer can contact manufacturers directly. Since all computer manufacturers are located in Asia, getting hardware for large scale deployments will be relatively easy. Another aspect on availability is the fact that new hardware is introduced often, which makes it difficult to give a prediction on what is available in future. One example is the VIA Nano CPU, which is promising, but will not see it's first boards for the coming two or three months.

There are three different scenario's thinkable with the outcomes of this research:

• Very low power consumption, not so great performance: The Ebox 4300. It consumes only 11 Watts at 100% CPU, making it an ideal candidate. Disadvantage of this computer is its' costs: $300,-
• Medium power consumption, medium performance: Boards equipped with a VIA Eden C7 CPU (more than 1 Ghz). Power consumption is around 20 Watts with these boards. The price is slightly higher as the Intel boards
• High power consumption, high performance: The Intel boards (with Celeron or Atom CPU) fall into this category. They are cheap as well, with a price of board and CPU somewhere around $75,-

Add a 2,5” harddrive and an open powersupply to it to keep the power consumption at a minimum.

This research has been carried out to enhance the level of knowledge on low power and low cost computer hardware for education. Hardware though, is only a very small part of the overall picture. In fact, the hardware decision should be made at the very last moment, after several decisions have been taken. There must be a plan on the kind of software which is going to be used on the computers, the operating systems (and whether techniques like terminal servers are used), maintenance plans, and so forth. Only if all these prerequisites are in place, a suitable hardware solution can be implemented. To give an example: If there is no organisation available to do the software maintenance, a probable software solution would not include server based computing, as some knowledge on server based computing should be available. This means that the desktop computers will need more memory and performance than when thin clients are used.