The Max RAM Tests

For the second round of testing, I wanted to see if the actual amount of physical memory made any difference when it came to throughput. To accomplish this task, I tried three different configurations. Since I was swapping memory in and out of my machine like a madman anyway, I ran tests with four 128mb DIMMs, four 256mb DIMMs, and three 512mb DIMMs (the board's max according to manufacturer specs), which totaled 512mb, 1gb, and 1.5gb respectively.

Now this is not a test to see "how much is enough" because quite frankly, that depends completely on you and what you do with your machine. As I stated before, this is all about the throughput. Let's see if Sandra has any better news for us this time:

Corsair CM654S128LP-133 ( x4 )


Corsair CM654S256-150C2 ( x4 )


Corsair CM654S512LP-133 ( x3 )

<sigh> Is it just me or is this depressing all of you too? I was really hoping to see something. Alas... No speed gains here, but again, no losses either. Let's check CliBench again and see what's up...

CliBench Mk III SMP 0.7.10 - 512mb Tests
	128 x 4	256 x 4	512 x 3
Dhrystone 2.1	2573	2560	2573	MIPS
Whetstone	869	869	869	MFLOPS
Eight Queens Problem	3404	3413	3404	pps
Matrix Operations	73972	73948	75118	k ops
Number Crunch	119349	119349	119410	k ops
Floating Point	11902	11902	11891	k ops
Memory Throughput	151920	173622	202560	kB/sec

Okay! For the first time in this whole barrage of benchmarks, we begin to see a glimmer of hope. I am not exactly sure why CliBench would show such an improvement in throughput by adding more memory. When I dug in to the FAQ's to try and figure it out, all I could find was this: "Memory throughput: Shows the access speed to your computers RAM. Depends on the amount of RAM, the type and the bios settings."

I am assuming (and you know what they say about that :-) that it is simply the way that CliBench charges and discharges the memory during the testing differing from Sandra. By the looks of the results, I would guess that Sandra does a better job of isolating true memory throughput than does CliBench. In any case though, the numbers did at least change for once! Woohoo!

Conclusions?

First of all I would like to take a minute to thank Corsair Microsystems again. Without their generousity, I would probably never have had the resources to test my theories here. They make some high-quality products and I never experienced any problem with their memory modules throughout my testing.

Show them some love!

As I have stated before, there is nothing really earth-shattering to learn here. We have all known the benefits of 4-way interleaving on the VIA chipset based motherboards. It has been shown in many reviews on our site and others. Unfortunately my original theory was blown completely out of the water, but there are a few tidbits we can glean from all of this.

First: If you are worried that adding another stick or two of memory to your machine will slow down memory performance, don't. We proved otherwise in the first tests. Four sticks of memory or one... The numbers are the same.

Second: The actual amount of system memory doesn't really seem to change our scores much either. If you were thinking of adding more memory to your system in hopes of gaining more speed, don't. Remember though, we are only discussing memory throughput here! If your machine constantly swaps to the hard drive or the applications you run are memory greedy, by all means... Add more memory! Just don't expect the throughput performance to skyrocket when you do.

Micah "Hooz" Schmidt