Testing
DDR-3 Setup:
- CPU: Intel Core 2 Duo E6400 (OC@2.67GHz, 1066MHz FSB)
- Motherboard: Asus P5K3-Deluxe Wi-Fi
- HDD: Samsung 160JJ SATA-II
- VGA: Nvidia Geforce 7950GT
- PSU: OCZ ModXstream 780W
DDR-2 Setup:
- CPU: Intel Core 2 Duo E6400 (OC@2.67GHz, 1066MHz FSB)
- Motherboard: Gigabyte 965P-DS4
- RAM: Crucial 2X1GB PC2-6400
- HDD: Samsung 160JJ SATA-II
- VGA: Nvidia Geforce 7950GT
- PSU: OCZ ModXstream 780W
Software used:
Sisoft Sandra 2007.5.11.35
Everest Ultimate Edition
AutoGK
In order to see how DDR-3 currently fares, we are testing it against a system running an "average" set of DDR-2 modules. The FSB speed is kept at 1066MHz in order to keep the results comparable. Voltage for DDR-3 modules is set at 1.7V and for DDR-2 modules at 1.9V. Timings are manually set in the BIOS. Intel Speedstep technology is disabled.
We also performed an attempt to overclock the SuperTalent modules beyond their specifications, but without raising the voltage above 1.7V. We managed to squeeze 1440MHz out of our 1333MHz set, with the CPU running at 2.88GHz (360MHz FSB). Of course most of the benchmark results are affected by the higher CPU speed, but we added them to the review for comparison.
Sisoft Sandra 2007 (Memory Bandwidth)
| |
Ram Bandwidth Integer (MB/s) |
Ram Bandwidth Float (MB/s) |
DDR-2 (4-4-4-12) 667MHz |
5557 |
5579 |
DDR-2 (5-5-5-15) 800MHz |
5644 |
5673 |
DDR-3 (8-8-8-16) 1333MHz |
6723 |
6740 |
DDR-3 (9-9-9-21) 1333MHz |
6662 |
6686 |
DDR-3 (9-9-9-21) 1440MHz |
6841 |
6952 |
Sisoft Sandra 2007 (Memory Latency)
| |
Memory Latency Random Access (ns) |
Speed Factor |
DDR-2 (4-4-4-12) 667MHz |
102 |
88.6 |
DDR-2 (5-5-5-15) 800MHz |
107 |
87.5 |
DDR-3 (8-8-8-16) 1333MHz |
79 |
69.5 |
DDR-3 (9-9-9-21) 1333MHz |
83 |
72.9 |
DDR-3 (9-9-9-21) 1440MHz |
84 |
74.9 |
Sisoft Sandra is a very popular synthetic benchmark which is ideal for testing the memory bandwidth and latency of any computer. Of course, as Sandra itself warns the user, it is a simple synthetic benchmark which runs a set series of tests and may not tally with real world performance. Higher bandwidth results are better and lower latency results are better.
Here we see that DDR-3 has a clear advantage over DDR-2 due to the 8-bit prefetch and raw speed. SuperTalent modules easily achieved an 18% increase over DDR-2 800MHz modules by merely running at their stock speed and timings. The latency results are also considerably better, proof that the access speed of DDR-3 RAM is much improved.
Everest Ultimate Edition
| |
Read (MB/s) |
Write (MB/s) |
Copy (MB/s) |
Latency (ns) |
DDR-2 (4-4-4-12) 667MHz |
6129 |
6045 |
4971 |
84.3 |
DDR-2 (5-5-5-15) 800MHz |
6626 |
6050 |
5135 |
77.3 |
DDR-3 (8-8-8-16) 1333MHz |
8255 |
6176 |
6696 |
59.8 |
DDR-3 (9-9-9-21) 1333MHz |
8099 |
6089 |
6593 |
63.6 |
DDR-3 (9-9-9-21) 1440MHz |
8385 |
6545 |
7047 |
62.9 |
Everest is also a very popular synthetic benchmark. Despite its similarities with Sandra, Everest is running very simple tests and provides different results which are close to the maximum possible theoretical speed of the system. Higher read, write and copy speeds are better but lower latency readings are better.
Once again the SuperTalent modules perform very well. Although the best results are taken from overclocking the whole system at 360MHz FSB, it seems that running the SuperTalent modules at 8-8-8-16 timings makes a noteworthy difference on the overall performance of the modules.
AutoGK (600MB VOB file, 75% video quality and AC3 audio encoding)
| |
Time (sec)
lower = better |
DDR-2 (4-4-4-12) 667MHz |
737 |
DDR-2 (5-5-5-15) 800MHz |
716 |
DDR-3 (8-8-8-16) 1333MHz |
663 |
DDR-3 (9-9-9-21) 1333MHz |
702 |
DDR-3 (9-9-9-21) 1440MHz |
652 |
AutoGK is a simple but popular movie encoding application. It will help us see how the difference in synthetic benchmarks affects the real world performance of a system. We are encoding a 600MB DVD VOB file to an Xvid movie down to 75% the original video quality and AC3 audio. Lower encoding times are faster.
The 18-20% speed differences we got from synthetic benchmarks will certainly not be true for every real world application since the speed of the RAM is not everything that matters in a system. Even so, the SuperTalent modules managed to outpace the DDR-2 sets even by a few seconds. It would seem that the true potential of the SuperTalent DDR-3 modules manifests when the timings are set tighter, lowering the encoding time by 39 seconds instantly without changing the CPU or bus speed at all.
