AMD Says Barcelona Will Be 40% Faster
One of the server execs at AMD recently made the following statement:
“We expect across a wide variety of workloads for Barcelona to outperform Clovertown by 40 percent.”
Note the careful and very specific language used here. Not “in general.” Not “most of the time,” or “for most apps,” or “on average.” They win by 40% in a “wide variety of workloads.” That’s not even a “wide variety of apps,” for all we know, it could be just one app!
If you have a product that can beat the competition’s across the board by 40%, or by an average 40%, you say just that. You don’t say what AMD said.
They didn’t say it because it’s extraordinarily unlikely that Barcelona could whomp Intel by that kind of margin overall in general overall computing firepower. This is still a Hammer, albeit a rather tweaked one. You’d have to have a far more radical redesign than C2D to increase overall performance that much at the same frequency (much less a bit lower), and frankly, even if you did, it would probably melt if you did.
However, first note what was said and who said it. A server guy said, “We expect across a wide variety of workloads . . . .” Sounds like he’s saying that for some server apps, Barcelona can do the floor-wiping.
Are server apps a bit different than desktop apps?
I think if you do a search of the server shootouts the past few years over at Anandtech, here’s an example, you’re going to find that even when processors are more-or-less evenly matched in raw processing power, one can handle certain tasks far better than the other. but that’s not true for other tasks.
One should also keep in mind that the server results at that time didn’t correlate too well with what was happening on the desktop; results in general corresponded much more to raw computing power.
So it’s quite believable that a Barcelona with things like HT 3.0 and better FP and just a better four-core design could handle some, maybe a lot of server tasks 40% better than Intel’s current designs.
And that’s all AMD said.
But all tasks? Not too likely, and more importantly, even less likely when the equivalent chips arrive on the desktop.
This is not to say desktop K8Ls won’t catch up quite a bit with Conroes, maybe even beat them here and there, at least clock-for-clock. In all honesty, that would be a pretty good achievement on short notice.
But beat them by 40% or more on average? No.