Practical recommendations:

Hello, captain cobalt cap> Every problem has a simple but incorrect solution. cap> You could point your finger at some kind of measuring device (for example cap> ZX_BENCH from ZX-Format#2) and calm down. This is already closer to the desired answer. cap> Axiom one: cap> ZX-Spectrum is a real-time hardware system cap> whose performance, by definition, is Z80@3.5MHz. My scheme uses two processors, let’s forget the second one, it’s a controller. The first one is the classic z80 at 20(25)MZh whose clock cycles are regulated by two ways, namely changing the clock frequency and duty cycle. The latter too you can forget, because need to do a comparison test... cap> It is important not that it is fast, but that there are no brakes. That is cap> performance should not be reduced under any circumstances cap> below that indicated. What if there is more? Much more. cap> What you need to measure is not performance, but brakes. To competently cap> measure the brakes (and determine whether they are within acceptable limits) cap> you need to know their source, and for this you need to know the hardwarecap> architecture of the machine being measured. Therefore - architecture in the studio. cap> Popular examples of brakes: video memory, rounding commands to cap> even clock, keyboard controller. Configuration Z80=20-25MZh/128 or more/HDD/FDD virtual/AY/LCD. The architecture is under construction and closer to debugging. All details in "Iron" Theme "Handheld". [http://zx.pk.ru/showthread.php?t=3115] cap> "Caching" was mentioned twice, as well as some "reference cap> cycle". If you mean the generally accepted meaning and the loop fits into the cache, cap> then the measurement result has no meaning. Caching is used in cap> machines where the processor is fast and the memory is slow. Everything in Speccy cap> is the opposite. And since it’s the other way around, then accordingly cap> the code is optimized, and just in the most critical cases cap> caching may be pointless. ..I have caching of a section of video memory, i.e. stopping the processor to work ULA is not happening! I would like to know how effectively this is used mode, like everything else, i.e. test needed. cap> Practical recommendations: cap> cap> Necessary (but not sufficient) condition for the absence of brakes: must not cap> slow down (and preferably display correctly) frame scrolls cap> multicolors and border effects, “heavy” demo. cap> cap> You can measure “practical” performance on: assembly, cap> compression, decrunching, artificial intelligence in strategic cap> games. cap> cap> Select software products to suit your taste, preferably popular ones. Until there is a test, it is difficult to talk about anything.