This review will deal with four RAM manufacturers:  Corsair, Kingston, Mushkin, and OCZ.  This is not a standard RAM review, in which a recently received stick of RAM is tested in a few benchmarks and results proclaimed; all of the RAM we’ll discuss has been in our labs at least six weeks, with some of it present as long as four months.  This type of pacing was deliberate, as it allowed us to use the RAM discussed herein in a wide variety of chipsets and motherboards, in synchronous, asynchronous, and overclocked modes.  Today we’re focusing on SPD (Serial Presence Detect) performance while running on a 166 MHz FSB, but follow-up articles will examine the impact of other performance situations and manual timings.

 Why SPD Timing?

 We’ve chosen to test SPD timings for our first analysis for several reasons. Memory timings are arcane to many people, not easily understood, and therefore less likely to be changed from their default settings.  Most motherboards will default to SPD, which gives us an opportunity to compare RAM performance with all sticks of RAM as timed by their manufacturers.  Secondly, because the memory we’ve tested here is set for different speeds, we’ll be able to analyze the performance impact between running in synchronous mode (166/166) and asynchronous mode (166/200, etc).  Much of the higher end RAM sold today is advertised as PC3200 or higher, but will a user actually gain any performance from running at these speeds?

 We will examine performance in overclocked synchronous mode and hand-timed settings in later articles.

 RAM We Tested:

 The following DIMM modules were tested for this review:
(Note--the following photos are stock and of 256 modules--actual module size is as listed).

Corsair XMS 3200 v1.1, 256 Meg DIMM:
Corsair XM3202v 1.1, 256 Meg DIMM: 

Kingston HyperX 3500, 512 Meg DIMM:

Mushkin PC3500, 256 Meg DIMM:

OCZ PC2700, 512 Meg DIMM:

 

Why the Disparity in RAM speeds and sizes?

 The differences in RAM speeds wasn’t planned, but as this is more of a comprehensive performance analysis than a simple review, it makes sense to examine multiple speed grades, especially when comparing synchronous and asynchronous performance.  Will PC3700 running asynchronously perform better than PC3500 running asynchronously?  Logic says it should, but we’ll find out. 

 The disparity in RAM size is simply a function of what test modules we received.  Although ideally we’d have had modules of a uniform size, we tested our selected benchmarks carefully to ensure that there was no contamination resulting from different module sizes.  Our detected performance difference in all of our tests was less than one half of one percent?well within an acceptable margin of error.    

Benchmarks

Test System:

AthlonXP 2600+ / 333 MHz FSB
40 Gig WD 400JB (8 meg cache)
ASUS A7N8X nForce2 motherboard
Chaintech GeForce4 Ti 4200

All tests were run with the A7N8X in single-channel mode.  While we'd have preferred to run in dual channel, some of our test modules were only single 512 Meg DIMM's instead of 256 meg pairs. 

 Quake 3 Normal

In Quake 3 we see the OCZ 2700 leap to the top of the pack, trailed only slightly by the Corsair XMS3200 and the Kingston HyperX back just slightly from that.  Total disparity between the top three is only about 4%.  At the bottom of pack we have the Corsair XMS 3202 and the OCZ PC3200, both a full 10% back from the OCZ 2700. 

There's an obvious performance disparity between the Corsair XMS3200 and the Corsair XMS 3202, and for good reason.  While Corsair advertised and sold its original XMS3200 RAM as certified for 3200 operating speed (and we have, indeed, run this RAM up that fast in synchronous mode), the original XMS 3200 is SPD'd to run at PC2700, or a 166 MHz bus, as is the OCZ PC2700.  In other words, the two sets of modules running synchronously lead the pack by far, with the HyperX not far behind.  After the HyperX, however, the other asynchronous models suffer significantly in asynchronous mode.

3DMark 2001

 In 3Dmark 2001 we see an identical pattern, with the OCZ 2700 again leading by a razor-thin margin.  The difference in performance isn't nearly as great between the first and the last scores here, and the PC3200 EL only trails by 3% instead of 10% as in Q3A.

 SiSoft Sandra 2003 Memory Benchmark:

 

 Scores in SiSoft Sandra are bunched so tightly that once a reasonable margin of error is factored in, it's impossible to declare a true winner.  Note that while we see asynchronous performance dropping off in games like Quake 3 and to some extent, 3DMark 2K1, the amount of bandwidth available to the system remains identical no matter what speed of RAM is being used.

 Benchmark #4:  PCMark 2002 (Memory Test Only)

 PCMark 2002's memory test definitely has a high FSB component, so our numbers aren't as clear-cut as Quake 3's, but a 6% gap between first and last scores is still evident.  Again we see the same pattern of the top three DIMM's hanging in solidly while the last group drops behind.

 Benchmark #4:  ScienceMark 2.0 Bandwidth

 ScienceMark 2.0's memory bandwidth test resembles Sandra's results, with only a 4% gap between the highest and the lowest scores.  Clearly running in asynchronous mode doesn't affect memory bandwidth nearly as much as latency?but just how much is latency affected?  Let's look at the next set of Sciencemark information:

 Benchmark #5:  ScienceMark 2.0 Latency

 

 

 

 

 

 This is the test suite that definitively makes the case against running in asynchronous mode.  While the number of cycles needed to conduct 4 byte and 16 byte operations is identical, moving to 64 bytes begins to open a gap and by 512 bytes our top performer (OCZ PC2700) is leading our bottom-feeder (OCZ 3200 EL) by over 16%. 

(0 of 4) (0 of 4) (0 of 4) (0 of 4)

---

• Pricing (0 of 4)
• OCZ, RAM Details (0 of 4)
• Overclocking, Conclusion (0 of 4)
• Benchmarks (0 of 4)