Web server
| |
DDR2-533 |
DDR2-1066 |
Gain |
| PHP Calculator |
176 |
176 |
0% |
| PHPSpeed |
| Synthetic PHP |
2815 |
2801 |
0% |
| Synthetic MySQL |
2636 |
2639 |
0% |
| Synthetic Read/Write |
1884 |
1964 |
4% |
| Real World PHP |
4173 |
4358 |
4% |
| Real World PHP & MySQL |
2448 |
2534 |
4% |
| Server |
1871 |
1902 |
2% |
| Group Score |
98 |
100 |
2% |
Strange, but true: our web-server tests practically don't respond to faster memory. One of logical explanations is as follows: the amount of data they use does not exceed CPU cache size.
Data compression
| |
DDR2-533 |
DDR2-1066 |
Gain |
| 7-Zip |
0:02:46 |
0:02:31 |
10% |
| WinRAR |
0:01:29 |
0:01:23 |
7% |
| Ultimate ZIP |
0:02:19 |
0:02:19 |
0% |
| Group Score |
91 |
97 |
6% |
It's an expected result - we know that 7-Zip and WinRAR prefer fast memory. The layout of forces is also not surprising: the larger the dictionary, the higher gains. Ultimate ZIP seems to use a very small dictionary, smaller than the cache.
Encoding
| |
DDR2-533 |
DDR2-1066 |
Gain |
| FLAC |
0:01:13 |
0:01:13 |
0% |
| LAME |
0:02:05 |
0:02:05 |
0% |
| Musepack |
0:02:17 |
0:02:17 |
0% |
| Vorbis |
0:04:45 |
0:04:45 |
0% |
| Canopus ProCoder |
0:05:21 |
0:05:16 |
2% |
| DivX |
0:01:24 |
0:01:21 |
4% |
| x264 |
0:02:48 |
0:02:44 |
2% |
| XviD |
0:09:47 |
0:09:43 |
1% |
| Group Score |
92 |
93 |
1% |
We've noticed the following regularity here: fast video codecs (DivX, x264) are the most sensitive to memory performance. And audio codecs just ignore changes in the memory clock rate.
Games
| |
DDR2-533 |
DDR2-1066 |
Gain |
| Call of Duty 4 |
116 |
116 |
0% |
| Company of Heroes |
54 |
55 |
2% |
| Call of Juarez |
45 |
46 |
2% |
| Crysis |
18.52 |
19.50 |
5% |
| S.T.A.L.K.E.R. |
108 |
112 |
4% |
| Unreal Tournament 3 |
93 |
113 |
22% |
| World in Conflict |
41 |
45 |
10% |
| Group Score |
90 |
95 |
6% |
Here is the first success: we've finally seen a really significant performance gain! Alas, it happens only in two games out of seven: Unreal Tournament 3 and World in Conflict.
Non-professional photo processing
| |
DDR2-533 |
DDR2-1066 |
Gain |
| ACDSee |
0:12:12 |
0:12:08 |
1% |
| IrfanView |
0:22:44 |
0:22:35 |
1% |
| Paint.NET |
31663 |
31329 |
1% |
| xat.com Image Optimizer |
0:41:28 |
0:40:54 |
1% |
| XnView |
0:25:29 |
0:25:21 |
1% |
| Group Score |
82 |
83 |
1% |
Absolute indifference and uncommon unanimity.
Conclusions
| |
DDR2-533 |
DDR2-1066 |
Gain |
| Professional score |
87 |
90 |
3% |
| Home score |
89 |
92 |
3% |
| Overall score |
88 |
91 |
3% |
In fact, there is no need to write a lengthy conclusion. The bottom line is crystal clear. Here is a summary:
- The biggest performance gain in groups of tests: games, archivers (6%).
- The biggest performance gain in applications: Unreal Tournament 3 (22%), World in Conflict (10%), 7-Zip (10%).
- Average performance gain - 3%.
There is even no need to comment on the last fact. I remind you that memory frequencies differed twofold. Twofold difference on one hand, and 3% performance gain on the other hand - do you need any comments?
The only comment (to be more exact, a reminder of obvious things) is not to forget about the dual-channel memory controller in Phenom. That is, in the best case scenario we can configure DDR2-533 memory to provide the bandwidth of single-channel DDR2-1066 memory. It seems that such memory bandwidth is quite sufficient for Phenom at this core clock rate.
