If you've the inclination and the bandwidth, you can download a 1,032 page PDF all about the Pentium here.
An interesting read, but it's appendix G -- Report on Transcendental Functions that we'd like to discuss today.
So if you're sitting comfortably, let us begin.
For boffins, the speed and accuracy of floating point arithmetic is often pretty important. Appendix G claims some accuracy figures using scatter plots which all fall in a narrow band on the pages.
But the trouble is that the actual results weren't as good as the not very pretty pictures on page 1017. The results in reality have points which fall not just beyond the edge of the paper but as far as Orion or Sirius.
One boffin puts it this way: "Pentium FPU transcendental function accuracy is better than that of a cheap calculator, it falls far short of what is possible with the 18 decimal digits supported. To the scientific community, the amusing claim of appendix G just reinforced the idea of buying more Sun workstations and computing transcendental functions in software to the accuracy required by the job at hand made more sense than fooling with toy Intel processors".
Even Linus Torvalds, in his Transmeta incarnation, appears to have believed Appendix G at one time, as you can read here.
But those clever folk at Watcom think there's something bogus about the scatter charts, as you can learn by reading this also somewhat lengthy Watcom C Library Reference.
Here, say the Watcom developers: "The cos function computes the cosine of x (measured in radians). A large magnitude argument may yield a result with little or no significance."
Oops! Could we describe this as somewhat Brownian? µ
Flagship will launch a day early to avoid being 'overshadowed' by Apple
EC says merged entity will 'continue to face significant competition'
Alexa, give me a reason to be cheerful about the UK economy
No, it isn't 1 April