How many transistors does the MOS tech 65C816 use, or alternatively how large a diy?

[Squeak]

This topic is not related to new technology, so I chose to put it here.

It might seem silly, but I've become a bit curious about the size of the 65C816, seeing as it's a 16bit extension of the 6502, it should be less transistor hungry than dedicated 16/32bit CPUs like the 68000?
In a related topic, how does a similarly clocked 65C816 and 68000 stack up against one another?

And also, now that I'm at it, was it ever agreed upon which was the fastest; the Z80 or the 6502?

I know some people are going to say that the speed comparisons depend on the task, but I want to know about the general speed of of the processors over a wide array of typical operations. Of course a breakdown of the weaknesses and strengths would be nice, i someone feels up to it.

Feel free to let any discussion drift after/if the org. questions are answered.

Edit: Yes, I know die is not spelled "diy" (stupid auto correct in my browser).

 

[WhiningKhan]

And also, now that I'm at it, was it ever agreed upon which was the fastest; the Z80 or the 6502?

The only practical and meaningful benchmark I can think of from those days is how they ran vector games like Elite, since stuff external to CPU (such as sprite handling and hw scroll support) didn't play a role in it. My recollection is that MSX and Spectrum ports were much smoother than C64 or Apple.

 

[Grall]

6502 ran at only ~1MHz. Z80 in the Sinclair Spectrum was clocked at ~3.5MHz, so even if the CPU was substantially more inefficient (which it probably wasn't), a very CPU-dependent game would still run faster on the Sinclair machine...

There should be plenty of cycle-count tables available for the instruction sets of both microprocessors; while there's bound to be coding differences between each architecture the basics should be comparable. They're both bound to have move, basic logic and arithmatic instructions and so on.

From what I remember reading back in the day, instructions in the C65816 ran faster on a per-clock basis than the equivalent MC68000 instructions since they were hardwired in the MOS design and executed through microcode in the Motorola chip, but in reality, SNES was inferior processing-wise to the Genesis/Megadrive due to clock speed differences...

 

[Thowllly]

6502 ran at only ~1MHz. Z80 in the Sinclair Spectrum was clocked at ~3.5MHz, so even if the CPU was substantially more inefficient (which it probably wasn't), a very CPU-dependent game would still run faster on the Sinclair machine...

The Z80 usually ran at a higher clock, but the Z80 used several clock cycles for each internal clock cycle. So while the 6502 used 2 to 7 clock cycles for each instruction, the z80 used 4 to 23 clock cycles (I think) for each instruction. But the Z80 had more registers and opcodes.

From what I remember reading back in the day, instructions in the C65816 ran faster on a per-clock basis than the equivalent MC68000 instructions since they were hardwired in the MOS design and executed through microcode in the Motorola chip, but in reality, SNES was inferior processing-wise to the Genesis/Megadrive due to clock speed differences...

If you translated 65816 code to 68000 code one opcode at the time, the 65816 would run faster. 65816 instructions use 2 to 9(?) cycles, 68000 instructions use 4 to... lots of cycles. (in 2 cycle increments)

If you rewrote the code to take advantage of the 68000s larger and wider register file, the 68000 might be faster. (each memory access fetched twice as much memory, but took 4 times longer for the 68000, so you should try to keep everything in registers)

 

[Squeak]

Register spare architectures is a whole separate school or design philosophy, starting with the Burroughs B5000 in the early 60's. The basic idea being that large registers is not really necessary if the machine is well put together and programmed tightly.
It's better to put your money on addressing modes and faster mem overall.

I also hear that the 65c816 is supposed to be less than easy to program in assembly (ie. much harder than it's predecessor or any of it's 16 bit contemporarys). Is this true?

 

[Grall]

The 65C816 may have caught some extra flak due to the SNES's quirky hardware architecture (memory banks out the wazoo, CPU able to run at 3 different clock speeds depending on what hardware you want to access, and so on)...

The truly hard to program microprocessors (in assembly, that is) probably didn't appear until the pentium pro and equivalent chips appeared, with their speculative execution... Trying to hand-schedule instructions by hand when the CPU can re-order them on the fly must have been a nightmare, lol.