Over 8,300 RAM models benchmarked
Notes on the Graphs
For a description of how the tests are performed see the benchmark information section. For reading and writing, the higher the results the better, but there can be several factors that can impact on the accuracy of the results in the charts. These are detailed below.
- Minimum Sample Size
- Different Operating Systems
- Background RAM Accesses
- Different Memory Settings
- Real Life Performance Comparison
- CPU dependency
- Integrated GPUs
- Naming of Memory
- Missing Memory & Misnamed Memory
- The memory aren't popular enough to have a sufficient number of samples.
- The memory or the controller doesn't support the SPD interface. Meaning the model number cannot be determined.
- Versions of PerformanceTest
Due to the sheer number of memory available, for a memory type to be included on the chart it needs to have had at least 2 submitted results. This is to help avoid few single really good or really bad results from throwing a memory type into a completely incorrect place in the chart. Due to the final score being an average of all benchmarks submitted for that memory it is still possible for one or more incorrect results to throw off the whole result when there is a small sample size. For this reason we have included information in the graph about how many memory of that type have been benchmarked. To see this information simply move the mouse over the bar of the memory you are interested in. A higher sample size means the data is more likely to be accurate.
The PerformanceTest software is designed to run on several different versions of Windows. So the results presented are a mixture of results from Windows 2000, 2003, XP, Vista, Windows 7 & Windows 8. These different OS have different levels of efficiency and as such a specific computer might perform better or worse depending on what OS is installed.
During the testing of the memory, if any other applications that may be running in the background that is also accessing the memory, it can affect the results.
We have observed that high performance memory often hasn't been configured to run at maximum speed. While the memory stick might support higher speed XMP profiles, these profiles might not automatically be selected in BIOS. Leaving the machine running a slightly slower speed than the RAM could support. This can also skew the averages.
Different applications make use of RAM differently. For some applications their performance can be limited by the speed of the RAM, but for others CPU or video card performance might be more important. So results in our charts will not necessarily match the real world performance with any specific software application or game. It is also common to experience diminishing returns in real world applications, where adding incrementally faster and faster RAM results in smaller and smaller performance gains.
It is also important to distinguish between the quantity of RAM and the speed of the RAM. Increasing the speed of the RAM can slightly improve the performance of some applications. Whereas moving from a situation of not having enough RAM (and being forced to use the paging file on the disk) to having ample RAM can dramatically improve the performance of the system. RAM speed is only important if you already have ample RAM installed.
If you are using a slow or old CPU, then it is unlikely to push the RAM to its performance limits. In these cases you can use any compatible RAM without being greatly concerned about its speed. The major bottleneck is likely to be the CPU and not the RAM. For the charts we have only included results from the fastest AMD & Intel CPUs. Including results from slow CPUs would mean we are measuring CPU performance not the performance of the RAM.
We have split up DDR3 RAM benchmarks into separate AMD and Intel charts. The reason for doing this is because the current high end Intel CPUs are much more likely to be able to use the full bandwidth of current DDR3 RAM. This means you can use pretty much any reasonable DDR3 RAM in an AMD system and not limit the systems CPU performance significantly (with the CPUs available in early 2013). There are two possible exceptions to this. One is for massively overclocked systems and the other is for AMD CPUs with integrated graphics (see below).
Even if faster RAM doesn't improve the CPUs performance much. On CPUs that have an integrated GPU in them, the GPU also shares the bus and the RAM when doing some 3D operations. The GPU can actually push the RAM harder than the CPU can in some cases.
So if you are planning on using integrated graphics and doing gaming, picking good RAM is still important. But you might see the results appear as better 3D frame rates rather than better CPU throughput.
Due to the fact that these graphs are automatically generated the names of the Memory have been taken straight from the modules itself, via the Serial Presence Detect (SPD) data. At times, the naming of memory by the manufacturers has been haphazard and because of this there will likely be some strange names in the graphs.
In some cases we have picked up names of rare or low volume memory. Just because you haven't heard of a particular memory module, doesn't mean it doesn't exist.
Some memory might not appear in the chart for the following reasons:
A memory might have an unexpected name as some manufacturers are not setting the internal memory name to match the name they use to market the memory. Or they may be rebranding an OEM memory module. In other cases too little information is included by the manufacturer, and we don't know the precise model number or size.
Only results from version 8 of PerformanceTest have been used in these charts as the memory test from earlier versions of the software does not produce comparable results.
Due to the factors above, the results provide only a general guide to which memory perform better. Memory which appear close together in the graph (with less than 10% difference in their scores) should be considered roughly equal in performance. Only larger differences should be considered statistically significant.
PerformanceTest memory suite contains a number of tests that exercise the memory sub-system of the computer (Random Access Memory- RAM). All tests use a combination of 32-bit and 64-bit data when reading or writing from or to RAM. Performance critical sections will utilize the assembler during the testing.
Measures the time it takes for a single byte of memory to be transferred to the CPU for processing. Measured in nanoseconds, lower values are better. The test will allocate a 256 MB buffer that is then filled with pointers to other locations in the buffer, essentially a looping linked list. The order of the links is randomized within certain 1024 byte ranges.
This test measures the time taken to read a large block of memory. Measured in GB/s, larger values are better. The test will uses a 256 MB block. The block is too large to be held in cache. Reads on 32-bit system uses DWORDS and QWORDS on 64-bit version of the software.
This test measures the time taken to write information into memory. Measured in GB/s, larger values are better. Similar to the read uncached test, except writing is performance instead. The test also uses a 256 MB block.