PC3000 Memory Comparison

The Memory, System Setup and Testing Methodology

This content was originally featured on Amdmb.com and has been converted to PC Perspective’s website. Some color changes and flaws may appear.

The Corsair XMS3000 memory is fairly new on the scene and offers a nice heat spreader on the chips to help keep them cool during high FSB overclocks.

The new Crucial PC2700 memory isn’t really intended for overclocking, but because many of the Amdmb.com readers use it, I decided to include in this review. There are no cooling solutions on the memory.

The KingMax PC2700 memory also came without any kind of cooling. I did try to get some Kingmax PC3000 memory that was announced on their website, but was unable to find it at any resellers at the time I was looking for it.

Mushkin has pretty long history with memory, as they were some of the first to offer overclocking friendly PC133 memory back in the day. Their purple heat spreader is fairly nice.

OCZ memory has been getting a lot of good press recently for having excellent overclocking results on the memory. They, too, are using heat spreaders on the chips.

Testing memory is a very delicate process. Because of variations in results from even the same chips at different testing times, lots of precautions were made to verify that the results were accurate with what the hardware was able to perform at and they were equally and honestly tested in the same manner.

The tests consisted of the latest SiSoft Sandra 2002 Memory test. For each result that is shown on these graphs, the test was run 20 times and the average of the results was found and recorded. If a result failed (i.e. the system locked up or reboot), it had to have this same problem in at least 3 out of the 20 test runs to be considered a failure. If it did not, enough runs were made to get 20 consecutive results to average and record.

And because we all know that SiSoft Sandra is not a difficult test to run on a delicately stable system, three consecutive test runs of 3D Mark 2001: SE were run to verify that each test result was taken from a stable platform. If a system setup did not complete the three-run test three times, it was considered to fail at that speed.

Using an unlocked Athlon XP processor, all attempts were made to keep the frequency of the processor identical at each stepping of the tests. However, it was impossible to keep the Athlon XP 2000+ at exactly 1.67 GHz. Some were run at 1.7 GHz and other at 1.65 GHz. This is just the vice of the dividers only coming in full numbers and halves.

The first set of tests that you will see are the five different memory types run at set FSB speeds. Each memory module was tested at 133 MHz with the same memory settings on each including CAS, Command Rate and Queue size unless otherwise noted. The benchmarks were run to test for the SiSoft results and the 3D Mark 2001: SE was run to test for stability. All the numbers were then recorded and shown in the review. This was repeated in the same manner for 166 MHz, 175 MHz, 190 MHz and 200 MHz.

The second set of tests shows the maximum FSB speed that I was able to attain with each module using the stated Epox 8K3A memory settings. Epox offers Normal, Fast, Fastest and Turbo settings, each with an increased in speed the settings such as CAS latency, Command Rate, etc. The memory voltage was maximized to provide the most power to the chips as possible and a cooling fan was placed above the chips for ample cooling. For a FSB to be considered a valid result, it had to pass both the SiSoft and three 3D Mark 2001 tests.

With that, here is the full system setup for the benchmarks:

Tests:

SiSoft Sandra 2002
3DMark 2001: SE