> Here's what i was told...
> INTEL is saying the 1.4 Ghz Pentium-4 is 20% faster than a 1 Ghz Pentium III. That's using "their" benchmark suite
> which I would think is a little favorable at the very least.
> They would not deny it was slow on a lot of benchmarks.
You know, benchmarks almost never scale linearly. If a 1.40GHz Pentium 4 is 20% faster than a 1.00GHz PIII on a benchmark that scales half as well proportionally as the clock, then you have a situation where the two processors have the same per-clock performance (mas o menos).
"20%" tells us nothing if we have no idea what benchmark it refers to. Additionally, every benchmark will act differently between the two microarchitectures. Did you know that the supposedly much faster Athlon (well, it *is* much faster overall, but I'm trying to make a point here) can get killed by a regular K6-2 in some code?
What you need is a list of comparative benchmarks, not just one. Here you go:
Compared to a PIII-1.00, the P4-1.50 is 13% faster in 3DMark 2000, 23% faster in Quake III, and 6% faster in SysMark 2000. It is also 89% faster in a simulation of some sort called Nastran (though I'm told that results can vary wildly in the program based on its datasets and what you tell it to do). It is allegedly 36% faster in the gogo mp3 encoder. 14% faster in specint2000. 56% faster in specfp2000. There is some evidence asserting that the P4-1.50 may be even with and possibly slower than a 1.00GHz PIII in some examples of native x87 code that is not extremely memory intensive.
> Now if someone can interpolate that in terms of a T-bird and Mustang we'll have something. I don't expect Intel to
> interpolate it for us on their website.
> My guess...1.2 T-bird whups up on a 1.4 Ghz P4...
> Mustang vs P4 may not be pretty....
> Never the less, P4 has Mhz and we know that sells.
> It also has dual rambus channels and Netburst technology. <G> plus a lot more FUD...
We should wait until we know more about Mustang before proceeding. I know that Jerry (Sanders. AMD's cap'n) wasn't fooling around when he alluded to heretofore unannounced surprises in the Mustang, some of which can have a tremendous performance effect, in terms of both ramping frequency and per-clock performance.
But they're (referring, likely, to both AMD and Intel) popbably going to disappoint us by not telling us any cool info for another month. >:P
-JC
PS: Those benchmark numbers above came from various sources, and in some cases may have been from preproduction hardware. It is my opinion that they are reasonably accurate compared to Intel's own impressions, but please don't treat this as utterly confirmed fact. :) |
