In our first DDR2 roundup, DDR2 Roundup: Reaching for 667 and Beyond, we quickly dispelled conventional wisdom about DDR2 performance and timings just a few weeks after the launch of DDR2. While all of the eight DDR2 memories that we tested were rated at the expected 4-4-4 timings, we found that almost all of them actually performed very well at DDR2-533 at 3-3-3 timings. Even more significant, every memory in the roundup also performed well at DDR2-667, the next expected speed bump, at 4-4-4 timings. All the DDR2 also reached DDR2-686, which was the test limit of the best Intel 925X that we had available at that time. That first roundup certainly proved that the misgivings about ramping DDR2 were unfounded.

What was not different in our performance tests in that roundup was that DDR2, at 533 and 3-3-3, still performed about the same as fast DDR400 memory. As we first saw in the 925X/915 launch review using the Intel D925XCV, DDR2 had to reach higher frequencies at lower latencies if it was going to give old faithful DDR a run for the money.

While sales of the platforms using DDR2 - the 925X/915 motherboards - have been lackluster in the 6 months since launch, the developments around the new Intel technology have been coming fast. Most significant of these was the launch of the 925XE chipset, which increased the FSB from 800 to 1066, and also brought DDR2 back to a 1:1 performance ratio, since the 266 multiplier of 1066 is a direct DDR2 conversion to DDR2-533. The widely expected move to DDR2-667 just didn't happen, as we now have a new 1:1 DDR2-533 and a new "next" speed bump of the 3:4 ratio of DDR2-711.

The 1066 FSB was a very limited introduction with just one $1000+ CPU at the new 1066 speed - the 3.46EE Socket 775 CPU based on the Xenon .13 technology. However, more 1066 are on the way, although we do not expect a wholesale switch by Intel until some time nest year. Despite the limited first launch, 1066 is where we are going in Intel and Socket 775, and manufacturers quickly embraced the new 925XE chipset on Enthusiast-oriented boards like the Abit Fatality AA8, the Asus P5AD2-E, and the Gigabyte 8AENXP-D. The 925XE is clearly, and quickly, the new board of choice among Intel enthusiasts, and not just for the 3.46EE chip. The new 925XE boards offer very flexible options to extract the most from any Socket 775 chip - 800 FSB or 1066 FSB.

So, what do we feed the new 925XE beasts? Several manufacturers like Corsair, Crucial, and OCZ introduced DDR2-667 memory. Most 667 were just hand-picked 533 that performed a bit better, since every DDR2-533 that we tested already reached 667 with ease. Most offered official 3-3-3 timings at 533 and decent timings at 667, but the reach at the top was only a bit further than the best of the regular DDR2-533. No one, it appeared, was doing much more with DDR2 memory. That is, until today.

Today, we are benchmarking the first of a truly new breed of DDR2 memory. It is rated modestly at DDR2-533, but at the very unmodest timings of 3-2-2-8. This is the first DDR2 actually to claim memory latencies more in line with what we see in DDR memory. Add to this that OCZ has heard the cries for fast 1GB DIMMs because these are 1GB DIMMs rated at 3-2-2-8. What's more, we are seeing early reports that this OCZ PC2-4200EB (Extended Bandwidth) is reaching unheard of performance levels on 925XE motherboards. That's a lot to live up to, and we couldn't wait to see what this new memory could really do in our memory testbed.

tRAS and DDR2
POST A COMMENT

22 Comments

View All Comments

  • dev0lution - Thursday, December 23, 2004 - link

    1GB DC Kit for $400+ and 2GB DC Kit for $800+?? I think I'll just keep the hyper-x 4300 I got for $200 less and spend more on my other components. The performance increase isn't THAT mind-blowing but the prices sure are! Reply
  • Lord Evermore - Thursday, December 23, 2004 - link

    Another sign of declining standards: the "Xenon" .13 technology? Reply
  • GTMan - Thursday, December 23, 2004 - link

    The article makes the claim that other high end memory is achieved by "hand picking" and then says that this memory is a "new breed".

    Where is the info to back this up? What is different (in terms of technology) about this RAM? Or maybe this article is only about numbers :(
    Reply
  • Carfax - Wednesday, December 22, 2004 - link

    The reason why the memory bandwidth scores are so low for DDR2-700 and 800 is obviously because the memory is bottlenecked by the FSB.

    The P4 would need to have a FSB of 1600 to take advantage of DDR2-800..
    Reply
  • Alphafox78 - Wednesday, December 22, 2004 - link

    Funny thing is that at the 700mhz speed the memory bandwith on my A64 with PC3200 is faster... Reply
  • Anemone - Tuesday, December 21, 2004 - link

    PS and I didn't even buy the matched 2gb set, just picked up 2 1gb individual sticks from Newegg.

    And at 4+ghz and sub 50c temps on air, I don't find any reason to worry about using an AMD solution...
    Reply
  • Anemone - Tuesday, December 21, 2004 - link

    I own 2 1gb sticks of this memory and while I"ve not had occasion yet to reach 811 fsb, every other timing they tested I have managed on a P5AD2-E (925XE) board. My PS is only the Antec Neopower 480 so it can also be done with a lesser PS as well.

    Expensive, yes, but given that no way would you ever see DDR1 in 1gb sticks doing this, makes it quite worth it if you need 2gb of memory in 2 sticks.

    $.02
    Reply
  • bupkus - Tuesday, December 21, 2004 - link

    Only one question...
    When will the time come that I can/should start considering DDR2 for my future AMD gaming PC?
    Reply
  • PrinceGaz - Tuesday, December 21, 2004 - link

    I just wish there were 1GB PC3200 or faster modules available for not much more than twice the price of the 512MB ones. Instead it seems 1GB DDR modules will always be overpriced and with slow timings. The Athlon 64 is crying out for fast 1GB PC3200 modules. Reply
  • MS - Tuesday, December 21, 2004 - link

    p.s. I know how to hand pick coffee beans but with memory, I would be out of my league... Reply

Log in

Don't have an account? Sign up now