Performance comparison test

Hello, All For the handheld ZX-Spectrum project, it was required write a comparison test of processor performance for comparison with clones and the original, if available. Are there processor performance testing programs for the ZX-Spectrum? :v2_cheer:

Hello, TYAX TYA> The ZX-Spectrum handheld project required TYA> write a comparative test of processor performance TYA> for comparison with clones and the original, if found. TYA> There are processor performance testing programs for TYA> ZX-Spectrum? :v2_cheer; Cheto; It’s not clear what you mean by performance test? Well, write it down more details!

Hello spensor To compare the “performance” of different Spectrums, I used the test, built into the FUT (Format Utility) utility.

Hello, moroz1999 mor> to the same magazine where the citadel demo was Citadel was in ZX-Power#2, but I don’t remember about the test

Hello, TYAX there was something in the appendix to the same magazine where the demo version of citadel was. sorry, I don't remember more details.

Hello, CityAceE Cit> To compare the “performance” of different Spectrums, I used Cit> test built into the FUT (Format Utility) utility. Can anyone explain to me what is meant by productivity? - processor operating frequency in turbo/non-turbo (is a parameter productivity); - number of cycles in turbo/non-turbo (not a parameter productivity); - strategy for setting WAIT when accessing memory in turbo (is performance parameter); - work with subsystems - disk drive for example (is a parameter productivity); - determination of the type of processor used Z80, Z84, T34VM1, 1858VM1-3 (not is a performance parameter).

Hello, CHRV probably the person means measuring the performance of the entire computer as a whole in "cockroaches" (as I call them) are in conventional units :), and you will be very happy His brainchild will eat 3 insignificant U.E. more

Hello, TYAX TYA> Almost like that, but I need to compare different modes in the product. Turbo, TYA> i.e. full load at 20-25 MHz and look objectively at the methods TYA> allowing you to achieve productivity gains through TYA> caching, etc. Well, naturally, compare with other clones. TYA> Naturally, I will be happy if productivity increases from TYA> changes introduced to the scheme (for some from 10-20%). Again, everything is somehow blurred and uncertain for you. Write what exactly you want to test, I even told you - listing some parameters. When you write, people will probably tell you absolutely specific programs. Productivity is a combination of various parameters, and not a certain quantity by itself.

Hello, CHRV CHR> Again, everything is somehow blurry and vague for you. CHR> Write what exactly you want to test, I even told you - CHR> listing some parameters. When you write, people will probably CHR> will suggest very specific programs. CHR> CHR> Performance is a combination of various parameters, not CHR> is a certain value in itself. Now I need to measure mips or simply the number of operations per second with "reference" code for a loop within a loop. Run this on your device and launch different modes to evaluate in reality the effectiveness of the applied caching methods, etc. In theory, you need to call the interrupt procedure. How many are going to take to the corner screen or to the attributes area. And in theory, you can roughly estimate the clock processor frequency or more precisely how much the processor is used as a percentage at nth clock and even more precisely what it corresponds to. It’s difficult for me to assess what other variations there are. That was the question.

Hello, TYAX not complicated but obscure... Let's say 3dmark understands what it does and why, but what you want is unclear... almost all clones and their features are described... i.e. the original is 68000 clock cycles in the inte Pentagon 71000 something if you need a measure of beats, then write it for no reason, but if something other - describe in more detail

Hello, TYAX TYA> What if there is more? Much more. This is more bad than good. TYA> ..I have caching of a section of video memory, i.e. stopping the processor TYA> for ULA work does not happen! How does braking happen? Always? Or only in case of a collision in certain areas of addresses, then which ones? For example, on 128 machines, four out of eight 16K pages are slow. TYA> Until there is a test, it is difficult to talk about anything. There are no "standard" tests. Each ironworker must write tests himself to check specific features of your hardware solution. Some tests were mentioned here [http://zx.pk.ru/showthread.php?t=1831].

Hello, TYAX Every problem has a simple but wrong solution. You could point your finger at some measuring device (for example ZX_BENCH from ZX-Format#2) and calm down. Axiom one: ZX-Spectrum is a real-time hardware system, performance which by definition is equal to Z80@3.5MHz. It is important not that it be fast, but that there are no brakes. That is performance should under no circumstances be reduced below designated. What you need to measure is not performance, but brakes. To correctly measure the brakes (and determine whether they are within acceptable limits) you need to know their source, and To do this, you need to know the hardware architecture of the machine being measured. Therefore - architecture to the studio. Popular examples of brakes: video memory, rounding commands up to an even beat, keyboard controller. "Caching" was mentioned twice, as well as some kind of "reference loop". If means the generally accepted value and the cycle fits into the cache, then the result measurement makes no sense. Caching is used in machines where the processor is - fast, but memory is slow. In Speccy it's the other way around. And since it’s the other way around, then the code is optimized accordingly, and precisely in the most critical In some cases, caching may not make sense. Practical recommendations: A necessary (but not sufficient) condition for the absence of brakes: they must not brake(and preferably displayed correctly) frame scrolls, multicolors and curb effects, "heavy" demo. You can measure “practical” performance on: assembly, compression, decrunching, artificial intelligence in strategy games. Choose software products to suit your taste, preferably popular ones.

Hello, TYAX TYA> Yes, on the topic of what’s just above... I’m a hardware guy who finds it easier TYA> it is much easier to connect a clock generator through a logic element TYA> "and" before or after the arrival of INT and measure the number of clock cycles TYA> frequency meter.. you can measure by clock cycles, you can measure by cycles.. How can you measure how much free time a processor has using a frequency meter? (or pulse counter)? It’s just that for these guys there are 3 floors in the operating rooms designs. If there is a halt, then in any case you need to write something of your own but it's easier with a calculator.