Most of the initial 1GHz-class DDR2 modules seem to be based on Micron's D9DQT chips, so the performance achievable would be quite close between - main differentiators being PCB quality, voltage settings and heat dissipation efficiency. That's where the differences happen at the very top end, pushing the clock beyond GHz.
Here we have the first such models from Corsair and Crucial in this league - we ran both DIMM pairs on Pentium4 EE 3.73 GHz 1066 FSB CPU, overclocked to 4.2 GHz on a quickly borrowed Asus 955X-based P5WD2 board at FSB1200 using Corsair COOL water cooling system. The memory was first ran at default dual-channel 600 MHz (i.e. DDR2-533 overclocked) in sync with the CPU FSB, and then at 900 MHz (overclocked DDR2-800 setting). Lastly, we used another Pentium 4 CPU, the 3.4 GHz EE 800 MHz FSB (old Northwood version) and pushed the FSB to 1000 MHz (clock 4.25 GHz) using the same water cooling system, to test the RAM at its advertised DDR2-1000 speed.
I ran both DIMMs at 2.1 volts and 2.2 volts to see if any difference in obtainable top clock. How were the latencies then? Well, at DDR2-600, both RAMs had no problem operating at 3-2-2-8 setting at 2.1 volts, while at DDR2-900, there was slight difference: Corsair managed 4-4-4-9 after being pushed to 2.2 volts, while Crucial had to run at 4-4-4-11 at this speed. At the 1000 MHz at 2.2 volts, Corsair ran at 5-4-4-9 claimed speed, while Crucial ran at 5-5-5-15 - still good settings for both memories, but obviously Corsair is somewhat ahead.
For the Sandra 2005 benchmark, I thought of comparing the best Pentium 4 results in this case (i.e. Corsair memory, 1200 MHz FSB / dual DDR2-900 and 1000 MHz FSB/ dual DDR2-1000) with the best I could get from a 2.4 GHz Athlon64 X2 4800+ - another Asus board lent to me, A8N SLI, with updated BIOS and 300 MHz CPU clock generator and one of the fastest DDR memories around, OCZ Platinum PC4800 EL Edition, certified to run at DDR1-600 with 2.5-4-4-10 timings. Just like Corsair PC2-8000 part, the OCZ memory had dual SPD settings, one low-latency 2-2-2-5 for DDR1-400, and another high-bandwidth for DDR1-600.
And here are the results:
CPU P4EE-4200 P4EE-4200 P4-4250 AthX2-2400
FSB MHz 1200 1200 1000 nil
memory DDR2-600 DDR2-900 DDR2-1000 DDR1-600
latency 3-2-2-8 4-4-4-9 5-4-4-9 2.5-4-4-9
Sandra
Mem Int 6149 7015 6064 6781
Mem FP 6053 7009 6067 6806
Well, as you can see, on the Intel platform, 50% higher memory bandwidth wins by far if everything else's kept same, even at the expense of substantially higher latency. Too bad we couldn't push the FSB much greater than 1200 MHz, or else, that DDR2-1000 memory could have had a better match in a 1320 MHz FSB CPU. The interesting this, Athlon 64 with its DDR1-600 dual channel memory was definitely faster than the overall lower-latency DDR2-600 on the 1200 FSB Pentium, and close to the winning Dual DDR2-900 setting on the Pentium FSB 1200.
So, in summary, both DDR2-1000 DIMMs lived up to the expectations, the Corsair with a slight advantage over Crucial. If you want to run them, well it may make better sense to use them with system with faster FSB, even if the memory clock is lower, to get the most out of them - at least until stable DDR2-1066 series appear to run in sync with 1066 MHz FSB systems. And yes, the performance difference vs DDR is still not there to the sufficient extent - let's see how the next-generation M2 socket Athlon64's perform with these DDR2 DIMMs!
Finally, a buzz to Intel: please up the memory voltage range on your 955X boards to at least 2.2 volts and of course 33% overclocking to enable FSB 800 -> 1066 move, then I wouldn't need to run around borrowing those Asus boards, which otherwise offer barely more than the Intel's 955XBK. Anyone at Intel reading this? ยต