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 : Advanced Micro Devices - Moderated (AMD) -- Ignore unavailable to you. Want to Upgrade?

To: fyodor_ who wrote (56723)	9/29/2001 11:21:39 PM
From: pgerassi	Respond to of 275872

Dear Fyo: They could simply open up SPEC to allow for optimized libraries for any platform since given the liberties taken towards Intel, allowing anyone to use the best algorithms available today instead of the outdated ones in SPEC. No one compiles using compilers that do not work on normal code. They compile on a good all around compiler, link in the optimized libraries, test for the best algorithm and then tune further using assembly. If pressed for time or needing only one shot, just the all around compiler (no time to screw around with the code to get the compiler to compile) and linking in optimal libraries are the only things done (they are quick and easy). Since most commercial, engineering or scientific work uses one of these two development tracks, that should be the basis to compare CPUs, systems, compilers, libraries and OSes. SPEC CPU2000 doesn't even come close as no allowance is given to using the best libraries or algorithms known to do better at getting the correct result in the shortest time. Whenever someone does this, P4 loses sometimes by a lot. Just by allowing the use of ATLAS libraries that optimize the BLAS matrix routines, SPEC would score as least 2 times as high on FP. However, it is allowed for a compiler to override the code to generate the call to a library or optimized function buried somewhere in the standard libraries or compiler. It is not allowed to be overt, but you are allowed to make it on the "sly". Even the P4 would have higher scores, just not enough to be above the final scores of the Athlons at lower clock. That would be at least representative of the "real world of software currently in use". The standard is so simple, a 66MHz step in Palomino clock increases the model number by 100. By multiplying the Palomino clock by 1.5, we get that a 1.533G Palomino would have a number of 2300. Since they are numbered at 1800+, there must be an initial offset of -500MHz. Thus the equation to model number is simply 1.5 times the Palomino's clock minus 500 appending with the plus to signify the initial bias or (Athlon Model 1800+ equals a Palomino at 1.533GHz). A 1.2GHz Palomino yields a model of 1300+ and a 1.733GHz, the largest before Throughbred, yields a model of 2200+. Pete