Like most internal components, there are a wealth of snythetic benchmarks to test RAM performance. As real-life performance advantages will be slight when switching from 1800 MHz RAM to this 2133 MHz RAM, it’s definitely best to go by the numbers. I used a collection of six tests:
- SiSoftware Sandra Memory
- AIDA64 Memory Benchmark
- PerformanceTest 7.0
- LinX Linpack 0.6.4
- 7-Zip Benchmark
Here’s my test PC:
- CPU: Intel Core i7 2600k @ 4.3 GHz
- Motherboard: Asus P8Z68-V PRO
- RAM: 16 GB Geil Veloce 2133 MHz
- GPU: 2x AMD Radeon HD 6950 2 GB in Crossfire
- Storage: Corsair M4 128 GB SSD, Western Digital Caviar Black 1 TB
- Case: NZXT Phantom White
Let’s have a look at the results.
SiSoftware Memory Bandwidth
- Aggregate Memory Performance (in GB/s): 26.08
- Integer Memory Bandwidth (in GB/s): 26.1
- Float Memory Bandwidth (in GB/s): 26
Memory Bandwidth is as good as expected from something at 2133 MHz, with all scores near 26 GB/s. For comparison, we’d expect about 17 GB/s from 1333 MHz, 20 GB/s from 1600 MHz and 24 GB/s from 1866 MHz. As you can see, there are definitely diminishing returns here, but you will get significantly better performance from a 2133 MHz module.
SiSoftware Memory Latency
- In-Page Random Access Pattern (in ns): 20.3
- Full Random Access Pattern (in ns): 63.1
- Sequential Access Patern (in ns): 5.6
Memory latency isn’t as good as I expected here. Compared to Bit Tech’s results for 1333, 1600, 1866 and 2133 MHz memory we see that we have a significantly slower latency time with a full random access pattern score of 63.1 ns versus their 2133’s 58.3 ns (as this is time, slower is better). That’s about equal to their 1866 MHz memory. I’m not sure why this is occuring – I suspect that their test system is more bereft of applications than mine or that we may be running different versions of the benchmark. I’ll update this section if I solve the problem.
AIDA64 Memory Benchmark
- Read (in GB/s): 21.49
- Write (in GB/s): 19.94
- Copy (in GB/s): 23.27
- Latency (in ns): 42.1
In AIDA64 (formerly known as Everest), we can see similar levels of performance – although as the operations use much smaller datasets, the difference between 2133 MHz RAM and its competitors is less noticeable, particularly in the read benchmark – here RAM clocked at 1866 MHz gets around 19.7 GB/s, 1600 gets 17.9 GB/s, 1333 gets 15.6 GB/s.
It’s interesting to note that reading to RAM is slightly faster than writing to it, with copying being the fastest operation. Latency is excellent, at 42 ns – although it still pales to the 1 to 4 ns latency in the CPU’s three levels of cache!
- Score: 5204
PCMark07 is not quite as useful a benchmarking tool, as it doesn’t have specific results for RAM speed. A score of 5204 bests all of Anandtech’s earlier tests, which use a very similar configuration but Patriot Viper Extreme Division 2 RAM. Still, the score isn’t really telling.
- Composite average: 4077
- Allocate Small Block (in MB/s): 7136.5
- Read Cached (in MB/s): 3052.1
- Read Uncached (in MB/s): 2884.3
- Write (in MB/s): 2816.6
- Large RAM (in OPs): 12331.6
PerformanceTest is produced by PassMark, who operate a very useful website for checking up on CPU and GPU benchmark scores. Their RAM test is useful but is usually used to test overclocked components. Still, even on stock frequencies the test PC reported good results, about average for computers with the same CPU and GPU as well as 16 GB of RAM according to the program.
LinX Linpack 0.6.4
- Peak (in GFlops): 47.5399
Linpack is an excellent single-threaded number crunching benchmark. However, the differences between RAM numbers are fairly small here, and this score is primarily down to the CPU rather than the RAM.
- Total rating (in MIPS): 21480
- Compressing (in MIPS): 21172
- Decompressing (in MIPS): 21787
This popular zip utility also has a benchmarking mode, which is quite useful as an example of a real-world test. We see good ratings here, beating out Anandtech’s Patriot DDR3 2133 MHz memory which got scores of 20525 and 20771 in their tests.