VIA 133A - Memory Interleaving Explored

As many of you know, I stumbled across some interesting results when testing the memory performance of my Tyan Tiger 200 motherboard a few weeks back. I postulated on the effects of DIMM size vs the number of DIMMs, but at the time I had no real means to test my theory. Well... Now I do.

Soon after my Tiger 200 review, I had the opportunity to talk with the good folks over at Corsair Microsystems. You may all know Corsair by their reputation of having some of the most overclockable memory modules on the planet (They have plenty of certified PC150+ CL2 modules), but Corsair actually started out in the server/workstation market and that is where their main focus remains today.

As I talked to "the RAM Guy" and explained my predicament, he became more and more interested in helping me test my theories. A few weeks and twenty DIMMs later, here I am.

The Question

We have all seen the benefits of using 4-way interleaving with our VIA Apollo Pro 133A based motherboards. Even the original VIA based duallie, the Tiger 133, showed great gains in memory throughput when interleaving was enabled via WPCRSET.

The question that arose during my time with the Tiger 200 was an interesting one. When I built that system, I happened to have four PC133 CL2 DIMM that were all 128mb in size. In all of my previous reviews I had only used two 128mb DIMMs. The memory results I achieved with four DIMMs installed were significantly better than those numbers run with half the memory installed.

Now obviously my first thought was "Duh! 512mb will always be faster than 256mb you idiot!", but I opened Task Manager and ran the benchmarks again. During SiSoft Sandra's memory tests, my machine (at least according to Win2k's Task Manager) never went over 235mb of memory usage... Well within the 256mb limit with only 2 DIMMs installed. Interesting.

Ever since the LX chipset days, I have been conditioned to think that having one big stick of memory is always better than having two small ones, but after doing a little more research about how the interleaving worked, I was beginning to question that mentality. Jim's review of Iwill's DVD266-R made me even more optimistic. Sure, it was a different (but similar) chipset and newer memory architecture, but his results with multiple DDR DIMMs put a serious smack-down on everyone else's tests of the same board using a single DIMM. At least in that case interleaving LOVED multiple DIMMs.

The Setup

As I mentioned earlier, Corsair Microsystems was more than generous in providing me with memory to use for this test. Actually, I was able to expand this article to cover more than I originally intended due to their generosity. Mad props to Corsair (and especially Robert) for helping out.

Now that the memory was out of the way, I needed to settle on a decent test board to use. Since I have been using MSI's new 694D Master-S for a few weeks (review is forthcoming) and have been very impressed with it, I figured it was as good a choice as any.

The hardest part of doing all of this testing was actually deciding how to enable the 4-way interleaving. The board that I had decided to use does not support the BIOS settings, so I was planning on sticking with WPCRSET. Until I found George's Memory Interleave Enabler. I don't know how anyone could make this any easier to tell you the truth. This enabler applies all of the same settings as WPCRSET, but it is installable (much like a driver file) and sets the memory registers upon boot-up. While I am almost positive that you could improve performance even more by tweaking the hell out of your machine with BIOS settings and/or WPCRSET, this is so simple and does a great job.

With the hard parts out of the way, I set out machine building. The rest of the hardware went like this (like it matters for these tests :-):