|
A WHILE AGO, we told you that AMD would not be adopting DDR2 until early next year, or about the time of the first speed bump in the DDR2 roadmap. AMD plans for DDR2 starting with DDR2-667, but Intel is about to introduce its chipsets soon. In a few months, we will know who picked the right path, but from the data, I think it is AMD.
Let me just start off by saying that DDR2 will initially have the "new and shiny" price premium attached to it. Until volume ramps up, it will not get cheaper, or at least in the ballpark of DDR. This is a no win situation for Intel, mainly because until it dives in and puts out large quantities of DDR2 supporting motherboards, there is no incentive for the Dramurai to ramp. Until they ramp, there is no market pressure to lower prices. It can suck to be the market leader, but I think Intel will somehow grin and bear it.
So at first, DDR2 will be expensive, with initial speeds of 400 and 533, the 533 variant will be more expensive. How much more? The numbers we saw have prices for the second half of '04 around $100 for a 512MB stick of DDR-400. A quick check shows this is about halfway between Crucial and the generic sticks on Pricewatch, so a little drop will put us about where these numbers are for the high end.
The same sheet puts DDR2-400 at $200 in the same time frame, and 533 at $250. This is a pretty hefty premium for memory, especially if you want the near mandatory 1GB in a new high end system. The same presentation has the numbers dropping to $85/120/150 in early ?05 and $75/80/110 in late '05.
So for at least the first year, you are going to pay a substantial premium for the new stuff, and it will stay that way until the arrival of DDR2-667. New stuff always drives the price down.
Price premiums are OK when the performance is there, and you can always count on the lunatic fringe of performance nuts to buy the initial samples. Sadly, in this case, the initial DDR2-400 will bring you less performance. Once again, that mysterious presentation we can't tell you about tells us that a Pentium 4 and an Athlon 64 will lose about two and three per cent respectively of overall system performance with the new ram. DDR2-533 will bring a whopping one per cent increase for both systems, well worth the $300 premium for two sticks.
It looks like with boosts like this, they aren?t going to sell all that well initially. The lunatic fringe is a small market. The abjectly stupid lunatic fringe is smaller still. Most people with a clue will avoid DDR2 for the first few months, but since I live in America, it will probably sell really well.
With the more costly arrival of DDR2-667, we get the first actual speed boost. Both platforms gain almost five per cent, with the A64 edging ahead by a few tenths of a percent. If you are underwhelmed by all of this, you probably have a firm grasp of the situation. Hundreds of dollars for a speed boost that $50 spent on the next higher CPU grade would eclipse.
Why the lack of performance? Two reasons, Latency and FSB. Latency is the easier one. DDR2 is simply a higher latency than DDR. While DDR has CAS latencies that started out at 2.5, DDR2 is at a minimum of three, initially. DDR with CAS latencies of two are readily available from almost every vendor out there, but from what I hear, it will be a long time until DDR2 hits that, if ever. Latency can kill performance, and it shows.
The FSB is also a problem. On an A64, if you run the memory out of synch with the CPU multiplier (there is no external FSB on these chips), it does a weird thing. Early Opterons were famous for having superlinear speed increases when you went from a speed where DDR-333 was not an even multiple of the CPU clock to one where it was. The CPU would ratchet down the memory speed to a more tolerable number. This was bad, and could happen again with 533 and 667.
Intel takes a bit bigger hit, and I would assume Prescott will fare worse than Northwood due to the longer pipeline. Either way, Intel has been stuck on an 800FSB for a long time, and help is not on the way in the near future. By the time you see an Intel 1066MHz FSB, DDR2-667 will be out, or at least be on the very near horizon.
This leaves Intel in the uncomfortable position of running the FSB in sync with memory at DDR2-400, taking a latency hit for no benefit over DDR-400, or running asynchronously. DDR2-400 looks to be a non-starter, but an expensive one. DDR2-533 looks to not suck as much, but it is even more expensive. DDR2-667 is where the fun starts, but even if Intel is on a 1066FSB by then, it will still be out of sync.
Overall, DDR2 looks to be an expensive no win situation for both companies. AMD, by delaying almost a year, avoids the bulk of the nastiness. Intel set a path, and looks to be following it. It is probably in for a bit more pain, but nothing Intel can't handle.
I think we need a new approach
Source:
The INQ!
|
Hybrid Mode
