Silicon Investor (SI) -- The First Internet Community

STOCKTALK

We've detected that you're using an ad content blocking browser plug-in or feature. Ads provide a critical source of revenue to the continued operation of Silicon Investor. We ask that you disable ad blocking while on Silicon Investor in the best interests of our community. If you are not using an ad blocker but are still receiving this message, make sure your browser's tracking protection is set to the 'standard' level.

Technology Stocks : Intel Corporation (INTC) -- Ignore unavailable to you. Want to Upgrade?

To: Elmer who wrote (118987)	11/22/2000 7:46:23 PM
From: pgerassi	Respond to of 186894

Dear Elmer: Why couldn't Anand get a copy of this supposedly available compiler? 3D Studio Max 2 is SSE and 3DNow optimized, and Athlon kicks both P3 and P4's butt. Matter of fact, P3 kicks P4's butt. You keep forgetting the Primordia (and its brother benchs where the source code is provided) where the P4 comes in second or worse and P3 is far behind. Originally, SPEC was to use real compilers used in a real way to compare platforms. Now, with enough money, you can play fast and loose with the rules. Intel's IFC 4.5 can't even compile correctly the Moldyn benchmark. Even they could not get it to (the compiled result they got was tested and would not even run). That is the hallmark of a lousy compiler or one that operates in a very restricted environment (SPEC). More typical in general application is the use of a base compiler to get the program to run decently and then to use profiling to see which sections need to be hand tuned in assembly language. This last step can double or even triple performance. Another way is to try three or four different algorithms that do the same job and pick the fastest and optimize that. Both methods are prohibited by SPEC in spirit (although, if one buries the hand tuning in the compiler, it is now allowed). Only the method described yields the best comparison between many different CPU implementations. Why should I write a compiler, when it is far easier to use the above method? That is why Tim Wilkin's benchmarks are better than SPECmarks. A better challenge is to define a problem and allow any method which computes the result on a randomly selected data set. All resulting source code, input data, and output results be given for peer review (will stop most dirty tricks). For code in the field now, P4 is beaten by Athlon 90% of the time at 125% of Athlon's clock rate. P4 is beaten by P3 30% of the time at 150% of the P3's clock rate. CVF6.5b is not completely optimized for Athlon, just better than IFC4.x. And at least CVF6.5b is more stable at generating P3 and P4 optimized code than the IFC 4.0, IFC 4.5, and whenever it shows up IFC 5.0 (in Intelspeak "with delivery later this month"). Pete