Intel RoadMap with P4 Stats To Boot

Anand reader writes "In the Intel Desktop CPU & Chipset Roadmap, AnandTech details the Intel roadmap before the Pentium 4 hits the streets next week. The article includes the desktop CPU and chipset strategy. They discuss and answer the questions. Does the Pentium 4 have a chance or is it doomed from the start? What will become of the Pentium III? And will Intel ever speed up the Celeron's FSB? and more including analysis of Intel's current 2000/2001 roadmap." Also see their official P4 stats and benchmarks.

Comments

Benchmarking Timeframe

[onion2k]

For the most part AMD certainly seemed to have topped the Pentium 4 chip on those benchmark tests. This seems to be a bit weird, but..

At CPU speeds more than about 1GHz theres little to choose between the various options on clock speed alone. At these speeds the chips are limited by memory bandwidth, code optimization, and instruction sizes. Once a CPU is going faster than the maximum memory throughput of the RAM then any increase in clock speed is going to go to waste. As the article mentions, the code that was run on the P4 wasn't optimized for it at all. This is another limiting factor. If you optimize code for a 286 instruction set and then run it on an Athlon it won't go as fast as it possibly could. And thirdly, AMD have some instructions that do more than those on the P4, thus appearing to go quicker.

Until memory technology, compilers, and applications start really using the new parts of the P4 chips then there'll not be any quantum leaps forward in 'speed'. But once they do expect benchmarks like these to look very different. (Mind you, by that time AMD should have some new toy out, and the field will be level once again).

Do gcc/egcs support the new SIMD/3DNow! stuff?

[oingoboingo]

i don't really follow compiler technology, but a lot of the speed improvements in the K7 and P4 look like they will depend on compiler specific optimizations. i'm sure the Intel provided C/C++ compiler will properly support all the new SIMD instructions in the P4, as will Visual C++ probably...but does gcc/egcs?

are linux users missing out on a big chunk of the potential performance available in the newer CPUs because their compilers are more tuned to cross platform availability than to x86 specific optimization, or do the GNU compilers already do a good job of supporting 3DNow! and SIMD?

i don't know...someone please tell me.

Is it just me or is this all very boring?

[Goonie]

As a computer buyer, I couldn't care less about the details of the coming lineup from Intel. Both price/performance and absolute performance are going to continue to improve, but when I come to buy my next computer, I'll have a look at what's available *then*. Trying to pick the optimal time for a computer purchase by reading roadmaps like these is like chasing the rainbow.

As a software developer, I don't care a great deal either, as the ABI isn't going to change. Games developers might care slightly, but even they are probably more interested in what video cards are likely to be mainstream in one or two years time as.

If I was involved in the computer hardware business, particularly, say, the memory business, this might be somewhat interesting, but these articles are not written for that audience.

Similarly If I was interested in guess what Intel's and AMD's stock was going to do, I might be interested in this article - but then again there are myriad other factors likely to impinge on their stock price, and it's not written for a financial analyst either.

So could somebody explain to me who seriously reads this stuff anymore?