Re: Intel is concentrating on ISSE2: No less than 144 new instructions...
That was pretty much my point. Consider that as long as you have code compiled and optimized for it, an Alpha processor has been much faster than anything else available for some time now - but that hasn't helped sales much. Even Office (compiled for Alpha) runs more slowly on Alpha than it does on AMD/Intel because the code hasn't been optimized for Alpha.
AMD has been concentrating on producing chips that run the existing code base more quickly, Intel used to do that, but now seems to be going for all out performance at the cost of compatibility (in the sense of performance - existing apps won't crash, they'll just crawl). In the past, the markets have favored the old-Intel, current-AMD strategy. Perhaps that will change.
We do a lot of spatial analysis using ESRI software. ESRI developed its applications on SUN SPARCs for close to 10 years, but has been emphasizing NT over Unix for about a year now. We still find, however, that a 2XXMHZ SUN is as fast or faster than a 6XXMHZ NT box (using either Intel or AMD) on most spatial operations. We know darn well that the SPARC chip is slower, and in both cases the disk subsystem is basically the same (7200rpm SCSI drives) but (at least this is our guess as to what is going on) the years of code optimization for SPARC/UNIX result in much better performance on the SUNs.
We've also seen better performance on existing LP based models from Athlons than PIIIs. Willamette sounds like a great chip, but if it needs ISSE2 to outperform a Celeron, it's likely going to face challenges. I know that until we get benchmarks, we can't know to what degree this is the case.
If Intel finds it necessary to develop new compilers and new benchmarks in order to publish benchmarks for Willamette, I'll take that to be a sign that I should buy something else to run any software that wasn't explicitly developed and optimized for Willamette.
The nature of the industry for the last 20 years has been that the sofware has lagged the hardware. As the installed base continues to grow, that probably won't change. Running existing code and running it well is very, very, important. If Willamette is twice as fast as Athlon on a few new programs, but only 2/3s as fast on everything else, it may not impress too many people.
With Intel marketing it, Willamette will do just fine, but if the first thing people do when they get their new Willamette is to load up that favorite program X and see how much faster it runs now, it would be best if X didn't run slower than before. On Thunderbird / Mustang it probably won't, on Willamette?.
(By the way, if AMD announces Edsel, I'm selling!)
Dan |
