about z80 speed: before saying the z80 spectrum is 15-20% faster, you should also take in consideration the contended memory. ;-)
on Amstrad CPC contended memory causes a 4Mhz z80 to behave like a 3.3 One...
For example, the Z80 in the spectrum is indeed faster than the MSX one, since there's wait states introduced in most instructions in the MSX.
The frequency is slightly faster too but MSX design is more reliable. It doesn't have these bugs:
https://sinclair.wiki.zxnet.co.uk/wiki/ZX_Spectrum_16K/48K
Shared RAM also greatly limits the programs size and hardware evolution.
Yes as happened to C128. To double the cpu speed they have to completely disable the VIC-II chip and go for a VDC that worked in a VDP like way (memory not shared)
I thought that the spectrum was actually slowed down by the PLA to allow for the main memory to be used to display an image? Or I am mixing things up with another machine from that era?
May be the ULA? the PLA was in the C64
I thought that the spectrum was actually slowed down by the PLA to allow for the main memory to be used to display an image? Or I am mixing things up with another machine from that era?
The ULA basically slows down the 16K of the memory (where the display memory is defined) if it needs to reads data there and send that to the TV. If you have a 16K Spectrum, then everything slows down a bit. On the Spectrum 48K, the other 48K (16K ROM and 32K RAM) don't have that problem, only the 16K part where the display memory is stored is slowed down.
In practice, I never really see much difference where I put my sprite data. The sprite and sound routines are stored on non-contended RAM though.
has anyone measured the same asm long running loop in and out the contended area? what's the difference?
Good points about the contended area, I am not sure about that case! I was thinking of raw performance in executing CPU instructions. But you are right, all has to be taken into account. I have never actually seen any comparative benchmarks, but maybe some exist? I'd be interested in seeing results, although designing exactly what to test might be tricky
Someone in the comment thread claims that the Spectrum has a bitmap mode (and the MSX has not). Is that true? Or is the Spectrum similar to screen 1/2 with more limitations?
It’s as much a bitmap mode as screen 2 is a bitmap mode in the default layout used by MSX-BASIC where the pattern name table is filled with 3x 0-255. The Spectrum pixel layout isn’t linear either so there’s no convenience there on either side.
The Spectrum has only one "screen mode", and it's 6144 bytes ordered in some order (more details later, maybe). It's pretty much the same as the 6144 bytes that MSX-BASIC uses, but in a different order. For example, each scanline has its own consecutive set of bytes. And there are many easy ways to get the correct addresses, but you could always be very lazy and just use a lookup table.
So, to answer manuel's question, the Spectrums screen layout is very similar to the MSXs screen 2(?) mode. However, people will call the screen 2 a pattern mode, and the Spectrums mode a bitmap mode. My guess is that people are saying that because the 'bitmapped screen 2 mode' is too slow on the MSX.
But because on MSX screen 2 is a pattern mode, when the level is built from a tile map the MSX is much more bandwidth-efficient. MSX allows you to change the entire screen with just 768 byte writes. On the Spectrum you have to rewrite the entire screen bitmap and colour data (6912 bytes) because you can’t reorder the 8x8 blocks.
The situation where the MSX uses more CPU bandwidth is only applicable when the screen is used as a bitmap (due to the higher colour resolution) without using the pattern tiling capabilities.
I like your explanation. As a person who is "relatively new" to the MSX scene, I should probably ask how is it possible that a 1984 machine is released that has "similar bandwidth" as a 1982 machine that is half its price. Is it the nice keyboard? The joystick ports? The ROM slot? Microsoft BASIC? (The Spectrum had no joystick ports nor ROM slots.) But I'm more interested in the other question.
Why would someone release a machine in 1984 that has half the numbers of sprites on the same scanline compared to a similar priced machine like the C64? The C64 VDP chip is similar to the MSX VDP chip, you'd think these developers could very quickly port the C64 code to the MSX instead of the Spectrum code. (This is not strange, back in the day there were people in the industry that could do both Spectrum and C64 code.)
(Look, I like the MSX1 too, a lot. Not in a nostalgic way, but I see it as a machine with its own interesting properties.)
Obviously, all Speccy ports are coded for a bitmapped memory layout, so the pattern mode functionality is not used and you end up with those games needing more CPU power and running slower on MSX.
Yes, as you said, it's sad that the pattern mode functionality must be used to make it run faster on MSX (I think that's the charm of the MSX1, but maybe that's only me.) Many of the great Spectrum games with 2 pixel scrolling, large sprites, many sprites on the same scanline, isometric games, vector based 3d games, never really arrived on the MSX1.
Strangely, it was not until around 1989 that finally 8x8 colourful beautiful blocky character graphics finally start getting popular on the Spectrum. I wonder if the MSX1 has those games too.
Look at that! I did it! https://www.youtube.com/watch?v=fqxpjl3_bMw
Download Link, please
there will be a release 
, will be a link. Now in the beta testing process. Soon everything will be 