AMD has claimed in a press release that its Barcelona will be a lot better than Intel’s stuff.
How much so? Here it is, straight from the horse’s mouth (emphasis ours):
The new Barcelona projections are based on the latest SPECcpu2006 benchmarks and show that AMD expects to have up to a 50 percent advantage in floating point performance and 20 percent in integer performance over the competition’s highest-performing quad-core processor at the same frequency.
This sentence really doesn’t make a whole lot of sense. You don’t “project” a benchmark when you have a finished product, you run one. You don’t “expect” a result, you have one. Yes, someone familiar with the spec benchmarks who has run X processor at a slower speed can probably estimate a benchmark of the same processor running faster with relative ease, but that means you don’t have that faster processor, doesn’t it?
To add to the mystery, even though the press release says there is Barcelona benchmarking available on a linked AMD webpage, there isn’t.
Well, estimates made under, well, pressure, can be . . . a little generous.
“At the same frequency” is another big red flag. Barcelona’s estimated speed is 2.5GHz; Intel’s is 3GHz. At the very least, any claims are clock for clock, which means a lot less when the real products aren’t running “clock for clock.”
Either AMD can’t make Barcelonas in any sort of volume yet, or the “projections” are based on an estimate of the results of some Extreme Tweaking that hasn’t happened yet.
There may be room for some more shenanigans, too. Is the Intel performance an “estimate” too? I don’t even know if you can even run a Xeon quadcore at a half-multiplier, which would be required to run an actual test.
Some of AMD’s current product spec benchmarking leaves a little to be desired, too. They claim a spec lead over Intel products based on a product tested in the first quarter, but if you look at the same product tested in the second quarter, rather than winning the close ones, it’s losing them.
Not that Intel is completely innocent of miscreancy, either. If you look at some of the spec scores generated by Dell and IBM for Intel products, you’ll find that a few of them . . . just weren’t run. Coincidentally, they seemed to be precisely the ones where it looked like the Opteron would edge the Xeon. 🙂
All in all, if you strip out the games (and outside of the clock difference, we are talking about just a few percentage points here and there), that 20% lead in integer probably gets stripped down to nothing or a few percentage points, but it would take suicidally colossal gameplaying of the benchmark for that 50% FP lead to be complete nonsense. 25% is probably more like it, but that’s still pretty large, and even if you assume less of a lead in real-time apps and games, you’d have to have deep-blue tinted sunglasses to think an FP lead wouldn’t remain substantial.
However . . .
If the competition in the real world was going to be a K10 vs. Conroe fight the second half of the year, things would look rather chipper for AMD. If you read some of the claims made at this (admittedly rather pro-AMD) website to be more-or-less true (and I think they are), the K10s should be very competitive against Conroes, maybe even a bit better.
However, that’s not going to happen because . . . AMD can’t make them yet. The place mentioned above also says that volume production of K10s won’t happen until early 2008.
This means it’s not K10s vs. Conroes, it’s K10s vs. Penryns.
Fortunately for AMD, Intel can’t make Penryns in volume anytime soon yet, either.
What will happen the second half of the year is that AMD will put out a relative handful of K10s, and Intel will put out a relative handful of Penryns in the server and extreme high-end desktop market.
This will be the occasion for a lot of screaming and yelling, but it will not basically change the current balance for the other 95% of the market.
The terms of the conflict will shift a bit; AMD will have more 65nm processors (and bring them to mobile); Intel will have cheap C2Ds.
Intel might be able to get a bit of an advantage if they can get more than a few Penryns out 4Q, but they probably can’t.
So the real battle will not be new vs. new, but old vs. old.
Nor will a handful of K10s transform AMD’s financial situation in the sense of profit-and-loss. It will improve the situation a bit, but good profits from 5% of the market can’t make up for losses from the rest of the business.
The real financial benefit of competitive K10s will be to make it more likely AMD will get extra money from someone to tide them through the rest of this year.
And frankly, from AMD’s perspective, if they get that, they will be more than happy with the outcome.