Benoit Poulot-Cazajous sent a letter with results from LINUX. He recompiled my fibonacci source using GCC and run it on different machines. K6/255 (Linux) - 492 (754 fhz). PII/300 (Linux) - 497 (746 fhz). PPro/200 (Linux) - 753 (492 fhz). Pentium/166 (Linux) - 780 (475 fhz). Still Pentium is very fast here - with 166 MHz it is almost as fast as PPro 200 MHz (less then 3% difference). In the same time difference between PPro 200 and PII 300 is almost exactly 50%, as it should be. 32-bit code for fibonacci test generated by GCC under Linux was executed on Pentium II faster than my 16-bit code under DOS, but not fast enough to beat Pentium.

Nenad Buncic did the same thing with Pentium II 300 MHz, using Microsoft Visual C++ 5.0: 492 ticks (754 fhz) and Intel compiler: 414 ticks (897 fhz). Unfortunately, there was no information from him about results on any Pentium machine, so this data were not comparable to anything. Then after few days he sent me exe file generated by Intels compiler and (after next few days) I was at last able to check it by myself on Pentium and Pentium II. My old Pentium 133: 926 (400 fhz) and Pentium II 233: 544 (682 fhz). As you remember, there is linear dependence between fhz and MHz, so if my Pentium was clocked with 233 MHz, it should have 233/133*400 =  701 fhz. For the same clock frequency even Intels own 32-bit code for fibonacci test was slower on Pentium II than on Pentium. This is somehow the most important information - the problem lies not only in 16-bit code, as it was many times suggested by readers of this page. And I have classic Pentium, not a little bit faster MMX one.

Gardonyi Gergely used Watcom 10.0 and generated both 16- and 32-bit versions. 16 bit: PPro 200 MHz scored 730 (508 fhz), Pentium 210 MMX scored 587 (632 fhz). Do you think this is because of 16-bit code? So take a look at his 32-bit results. PPro 200: 625 (593 fhz), Pentium 210: 488 (761 fhz). 32-bit code for fibonacci test generated by Watcom 10.0 was faster on Pentium then on Pentium II.

This code I was able to check by myself. Dan Ogorchock send me exe compiled with Visual C++ 5.0. His machine (Pentium 133) scored 1098 clock ticks (338 fhz). My machine (Pentium II 233, in case you forget) scored 665 (558 fhz). What it means? If the speed is proportional to processor clock, you may expect from Pentium overclocked to 233 MHz 233/133*338 fhz. It will be 592 fhz. Pentium II has 558 fhz, so still Pentium executing 32-bit code for fibonacci test generated by Visual C++ 5.0 was faster than Pentium II (but only about 5% this time). Doesn't matter how you revolve, ass is always on the back.

This is my own work: I put my hands on Visual C++ 4.0 and I spent several hours playing with compiler and its options. In some cases Pentium II was faster, in some cases it was slower. Differences are very small and it is incredibly difficult to say which compiled version is a proper one to perform tests on different machines. That's why you may not expect nothing new in a few days - I have to try it once again. (1st February).