Something is screwy here with the DDR3 synthectic benchmarks.

If you look at the DDR3-1333 results in the forget the DDR3-1333 section and compare them to the DDR3-1333 in the DDR3-1600 section you will see a significant difference. DDR3-1333 get much better numbers in the testing where the reviewer is not trying to discount the product.

Great article--thanks! I have 8GB of 1.8V DDR2/800 running at 5-5-5-15 timings with a 400MHz FSB, and I chose 800 rather than 1066MHz memory mainly because I was unsure whether four double-sided modules would work stably at much above 800MHz. Nonetheless, I was still considering either reducing the timings to 4-4-4-12 or increasing the memory clock to obtain a little extra performance. However, your results have shown me that this would offer so little benefit that it's effectively a waste of time to determine where the bounds of stability lie, and now I feel a lot better for using completely standard settings and not having spent extra money for faster memory!

Nice article, even if the last half did look like an advertisement for the major memory makers.

Two comments/questions for you guys.

1) Your gaming benchmarks. If these are going to be "real-world" benchmarks they should be run at realistic resolutions (1680x1050 or 1920x1200) to show the true impact of memory speed/timings/latency in games. Even if that means a flat line with no real differences (which I expect would be the end result).

2) I'm highly curious how your FSB results would correspond to similar testing on Core i7. When you increase BLCK the bandwidth of QPI goes up, how much effect does that have in various applications? I can imagine a basic matrix for testing:

BLCK = 150
CPU = 20x150 = 3GHz
RAM = 8x150 = 1200MHz

BLCK = 200
CPU = 15x200 = 3GHz
RAM = 6x200 = 1200MHz
RAM = 8x200 = 1600MHz

BLCK = 250
CPU = 12x250 = 3GHz
RAM = 6x250 = 1500MHz
RAM = 8x250 = 2000MHz (Corsair has a kit...)

This would provide a great comparison of memory speed & QPI bandwidth. I know I would like to see this testing done!