There’s a review of various PIV boards, including the DDR contenders, here.
While this sort of test really needs to be done with Northwoods, it does demonstrate a few valuable points:
Update: Sander Sassen sent me an email respectfully disagreeing with point one by noting that soon-to-come faster RDRAM provides significant benefit to even current systems. You should read it.
You may say, “DOH” to the latter, but it’s a bad “DOH” to make.
CPUs are not capable of fully utilizing infinite bandwidth. Exceed the bandwidth the CPU can handle, and you get little or no real performance improvement.
The best example of that is the PIII/Celeron. DDR does virtually nothing for these chips, and that is the reason why. Those CPUs just can’t effectively use 2.1GHz of bandwidth.
On the other hand, even the initial Willies were designed to handle 3.2GHz of bandwidth, and next year, we’ll see Northwoods designed to handle 4.3GHz.
While the importance of memory bandwidth has been grossly exaggerated, especially for office-type apps, it does help games and other streaming media to a significant (though still hardly huge) degree.
One of the biggest unanswered questions about the current Palomino and future Throughbred is: How much bandwidth can it effectively use?
Right now, we’re handicapped because true DDR333 memory is not readily available yet, but that won’t last long.
AMD’s reluctance to embrace the DDR333 standard in its roadmaps may be an indicator that their current and near-future-term CPUs won’t really benefit from DDR333.
While that may not be too big a hurdle for Palominos, that is likely to become a competitive disadvantage later in 2002 when Thoroughbreds are facing 533Mhz Northwoods.
Not much of one, maybe a 3-5% advantage at most, but in a likely close race, that could tilt the advantage to Intel.
Could. Not will. There’s a bunch of unknowns at the moment which may prove to be decisive.
We don’t know about relative overclockabilty, for instance. On paper, an overclocked Northwood should be able to wipe the floor with an overclocked Palomino, but initial hints and clues suggest a “Let’s see it first.”
Even if it does, it hardly matters because Northwoods won’t become price-competitive until late spring if even then. For at least this market segment, the competitive matchup will be Thoroughbred/Northwood, not Palomino/Northwood, and that starts about the same time as summer.
I’m getting the growing impression that Thoroughbred itself is yet another six-month wonder for AMD; a stop-gap until Barton, which itself looks like the Duron for ClawHammer.
It almost seems like AMD needs Barton to get the Athlon design running at very high speeds; their roadmaps point in that direction. I grant you, this is close to reading tea leaves, but don’t be shocked if Thoroughbred isn’t a great overclocker.
This approach may work well enough for AMD in the overall broad market, but it doesn’t look so hot from the overclocker’s perspective in 2002.
Six months from now, we may well be looking at Northwoods capable of 3GHz+ with mimimal overclocking skill facing Thorougbreds of uncertain overclockability and ability to use bandwidth.
At that point, I think we’ll see a shift back to Intel by at least some, and Waiting For Hammer for many others.
Too early to tell, but not too early to figure out what to watch for.