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: Dan3 who wrote (51249)	8/15/2001 3:45:12 PM
From: Tenchusatsu	Read Replies (2) \| Respond to of 275872

Dan, <I am suspicious of any benchmark created by authors with the chip in hand to tune the benchmarek> That would pretty much amount to ALL application benchmarks and most synthetic benchmarks as well. Benchmarking based on the greatest common denominator is useless because the code will most likely be so basic and so unoptimized that no performance-minded application would ever use such code. Tenchusatsu

To: Dan3 who wrote (51249)	8/15/2001 4:27:29 PM
From: wanna_bmw	Read Replies (3) \| Respond to of 275872

Dan, Re: "You keep tossing out the ones you lose, then emphasizing the ones created by Intel/Bapco that were tuned to cover up the failings of P4." Wrong again, Dan. First, there wasn't a single benchmark in the entire suite that was constructed by Bapco. The Bapco / MadOnion alliance makes SysMark, 3DMark, and WebMark, and neither of those were in this suite of tests. Also, in order to be consistent with what I did threw away, I had to throw out the ones where Pentium 4 did well, too. The Sandra memory bandwidth test had the Pentium 4 winning by over 2x. I might have had a higher score if I factored in all the benchmarks. The point is, which you continually try to avoid like bad cold, is that some benchmarks deserve to be removed, retired, or ignored. How many people are going to buy a brand new computer, only to install applications of 4 years ago? Any of those applications run fine on their current Pentium machine, so why upgrade, unless there are other newer applications that can be run? Winstone 99 tests these old, retired applications. Useless. How about synthetic benchmarks? They run instruction streams that are very regulated because they are testing a specific portion of the CPU. FPUMark runs gobs of FPU instructions. CPUMark runs gobs of simple integer instructions. Real life applications do neither. Any of those tests, whether they score well or poor on the Pentium 4, are completely useless. How about calculating Pi? Another completely useless benchmark. I know you don't know how processors work, but you pretend you do, so I'll tell you what makes SuperPI perform awfully on a Pentium 4 platform. PI, like Cosine, Sine, Log, and others, are approximated on a CPU by using something called a transcendental function. These go through iterative calculations that tend to occupy a pipeline for many, many cycles. On the Pentium 4, this is even worse, since the cycle latencies have been reduced in that architecture to accommodate the longer pipeline (i.e. less transistors per pipeline state means less tricks available to lower latency). On the Pentium 4, transcendental functions take almost twice as long as other x86 micro-architectures, including the Athlon and Pentium III. SuperPI tries to recursively run transcendental logarithm functions to approximate PI to X number of decimal places, and the throughput will be very low on a Pentium 4. But here's the kicker, Dan. Most applications use less than 1% of transcendental functions, even in FPU intensive calculations. SuperPI uses a far greater percentage, and thus it is another useless benchmark. wanna_bmw