Comparing Results & Conclusions
Putting it all Together
Now let’s take stock of what we have just seen on three different Intel 915P motherboards running the exact same hardware. (I apologize for the table… there are too many numbers! :).
| Asus P5GDC-V |
ABIT AG8 | Chaintech V915P | ||||
| Stock | OC | Stock | OC | Stock | OC | |
| Clock Frequency (MHz) | 200 | 238 | 200 | 237 | 200 | 234 |
| Sandra CPU Dhrystone ALU (MIPS) | 10466 | 12387 | 10357 | 12283 | 10456 | 12108 |
| Sandra CPU Whetstone FPU (MFLOPS) | 4316 | 5160 | 4322 | 5135 | 4287 | 5058 |
| Sandra CPU – ISSE2 | 7502 | 8908 | 7433 | 8922 | 7421 | 8742 |
| Sandra Memory Float (MB/s) | 4424 | 5160 | 4571 | 5333 | 4523 | 5258 |
| Sandra Memory Int (MB/s) | 4422 | 5157 | 4570 | 5330 | 4521 | 5252 |
| PCMark 04 | 5316 | 6229 | 5497 | 6279 | 5300 | 6079 |
| 3DMark 05 | 2611 | 2654 | 2623 | 2693 | 2407 | 2476 |
| Half-Life 2 – Avg FPS | 60.06 | 59.65 | 59.59 | 61.23 | 48.60 | 49.21 |
| Half-Life 2 – Low FPS | 31 | 28 | 29 | 31 | 25 | 26 |
| Half-Life 2 – High FPS | 185 | 202 | 193 | 224 | 171 | 186 |
Looking over the data we can make the following final observations:
- Though there are minor differences amoung the discrete numbers, overall all boards perform perform very similarly even overclocked. The exception being the Chaintech V915P being a bit slower in 3DMark’05 and in Half-Life 2. I’m at a loss to explain these abnormal numbers.
- In our tests, overclocking does not benefit 3D gaming as seen by results in 3DMark 05 and Half-Life 2. None of the motherboards showed any substantial improvement in frame rates.
- In the tests, overclocking benefits mostly CPU and RAM performance and therefore will show better results in intensive applications like computations, media generation, and rendering applications. General desktop usage (web browsing and word processing) will not likely see much difference. Evidence of this benefit can be seen in countless other articles on overclocking published.
Final Remarks
This was an interesting exercise in overclocking. With the exception of the ABIT AG8, it would seem that the board with better overclocking at 14x CPU with asynchronous multiplier (see the individual reviews of these motherboards for 14x overclocking results) implies better overclocking at 18x with synchronous settings. However, without more 915P boards to test using the same hardware, it’s impossible to state this as a fact.
The ABIT AG8, though our best overclocker at 14x, failed to beat the Asus P5GDC-V due to an unusual overclocking bug. This is a pity since we would have very much liked to have seen how far the AG8 can go. I hope ABIT can look into this issue and perhaps release a fix for me to attempt this experiment again.
Lastly, and more importantly, though overclocking is a fun enthusiast activity, it is a bit touchy for gaming purposes. Since so many games are programmed differently, some games will benefit more from overclocking than others (i.e. some games are more CPU intensive than GPU intensive). However, it is a bit surprising to see the results in Half-Life 2 flat and show absolutely no real improvement. It’s clear that if you really want to get more performance in games, forget the system clock and focus on tweaking your graphics card instead.
However, if you’re crunching numbers, proteins, SETI work units, or rendering models, overclocking will likely give you a nice boost in productivity.
What Now?
The experiment you just read is far from complete or thorough. There are many places in which the process can be improved and expanded (i.e. add more games to test). If you are attempting similar tests, or have improved upon the experiment, please let me know your results! I am very much interested in hearing from you. Thanks for reading!
