DDR3 delayed, DDR2 to go 1333

[Alpine]

The latest from Old Taipei


THIS TIME, on my regular Taipei monthly jaunt, I was in the mood for some sweet memory - so I popped by the usual suspects, high-end memory brands like Corsair, G.Skill and Geil, to see what's up. Here we go...

DDR3 was, of course, on everyone's lips. One of Corsair's top hats is holding the DDR3 talk prior to Intel's global Spring IDF in Beijing this coming Monday, and the X38 top-end chipset for Intel's Core CPUs will only use DDR3. But - well - there aren't many DDR3 dies around, and those that can be found are expensive and slower than expected.

Instead of DDR3-1333 CL6, which was anticipated as a good bargain knowing it'd run at 1.6 volts, we may have it at noticeably slower CL8. Those who want DDR3-1600 speed levels may have to, for the first time, endure CAS latency in double digits - how does CL10 sound? To me, kinda awful.

Due to the DDR3 shortage in this initial period, don't be surprised to see the first batch of X38-based boards by Intel and others to come bundled with a pair of DDR3 DIMMs - I just hope the fastest ones will be in there. There is no official comment by any of the mobo vendors on this yet.

In the meantime, DDR2 has to fill in the void left by this delay and probable performance disappointment with this year's DDR3 line.

So, by next month, expect to see the first DDR2 DIMMs reaching the DDR2-1333 level, from more than one vendor. These 2 x 1GB dual module kits should give you CL5-5-5 performance at that high speed, but with 2.5 volt input - sufficient to, over some time, fry most current memory chips except Micron and Promos (plus some Elpida) dies. Even if they work fine, they will be as hot as an Nvidia North Bridge and will need commensurate cooling.

All the usual suspects are expected to have (or at least announce) DDR2-1333 DIMMs by Computex, except that the voltages and latencies may vary slightly, depending on the PCB, cooling and the vendor's luck with the memory dies obtained (i.e. whoever karaoke'd last night in the musky streets of old Taipei with the memory wafer sorting guy).

Bear in mind that this raw MHz bandwidth might not translate into benchmark results, despite all the power and heat - I did run GEIL RAM at DDR2-1250 CL5-5-5-8 and 2.4 volts vs that same memory at DDR2-833 CL3-3-3-5 and 2.2 volts on the Nforce 680i, feeding Intel's QX6800 quad-core 3.33GHz with FSB1667. In both Sandra and PCMark06 memory, the 833 set was either slightly faster or about the same speed (plus minus one per cent) as the 1250. The in-sync perfect frequency and bandwidth matching of FSB and memory of course helps the 833 setup.

So, what's the point of spending more money, power and tweaking time (not to mention unavoidable lifespan reduction) with the 1250 memory, when there's no benefit?

And yes, do expect less expensive 2 x 2GB kits soon, as the true 1Gbit die production moves into higher gear. Also, as witnessed by recent Mushkin launch, DDR2-800++ CL4 at this high density level will be commonplace this year, possibly brought in to CL3-3-3 before year end.

This is very good news for large-memory quad-core users in need of both capacity and performance from their memory - four sticks of this give you darn fast eight gigabytes to play with...


The INQuirer