I think we did have some very valuable discussion sessions
here explaining quite a lot of the 68050 internals and its design.

I wonder if someone would be interesting in helping us to write this all together to create something like a review of the 68050 design.

The review should be able explain the goal and virtues of the 68050 design to the world. It would be nive if the review would be both easy to understand and interesting to read of course.

If someone would like to help us doing this - this would be real great.
We would be happy to provide all information needed for this.

If you like to do this please post here.

Cheers

btw will the developer version be out this year, any guesses?

- instruction efficiency compared to that of intel cpus. Would that be favourable for Natamis?
- number of registers
- Say it is an Amiga with full-speed DVD playback
- Say, optimistically, that there will be new programs coming for it.
- Low power consumption, no fan, therefore is green
- OS-wise:
  boots in 1.5 seconds, reboots in 1.5 seconds
  anti-aliased fonts
  mp3. wav, ogg playback
  dvd playback?

To prevent any misunderstandings.

The 68060 does not have 4 stages.
The pipeline of the 68060 looks like this:

01- Instruction Address calculate
02- Instruction Fetch
03- Decode Early
04- Instruction Buffer (1-4 slots)
05- Decode
06- EA-Calculate
07- Data-Load
08- ALU
09- Data-available
10- ALU-Writeback

Regarding the "@ 7x 20MHz?" - Don't let yourself get confused by this.
The 68060 for example is typically clocked with 50 Mhz.
This means the whole chip is clocked with this clock.
Each pipeline stage runs with this clock also.

Regarding the 68060 - the 060 is superscalar.
This means it has two integer pipes, the stages 05-08 are doubled.

Cheers

If you aren't interested, go to hell!!!!!!

the N050 has bigger caches then the 68060.
and a more "inteligent cache controller".(Advanced)
the scaleablity of the design is better.

a relative simple and clean design.

and would probably without to much trouble outrun a coldfire running legacy code.

and all this while maintaining one of the ease of use ASM ppl are familiar with using the 68K family line.

Realy does it?

Comparing the different 68k models is directly is certainly not that easy.

Oversimplified, I would group the 68K CPU like this:

68000:
- Baby CPU!
- no caches at all
- very low clock rate
- each instruction needs many clocks (4-100)
- as more bytes an instruction is long, as longer it takes

68020:
- Added many new complex address modes
- Added new instructions (32x32mul, 32/32div, ..)
- Small ICache, No DCache
- low clock rate
- each instruction takes several clocks (2-40)
- as more bytes an instruction is long, as longer it takes

68030:
- The 030 is basicly a 020 plus small Dcache
- Small ICache, Small DCache
- high clock rate
- each instruction takes several clocks (2-40)
- as more bytes an instruction is long as longer it takes

68040:
- The 040 is a radical change
- big ICache, big DCache
- high clock rate
- instructions only need few clocks (1-30)
- many instructions only take 1 clock!
- instruction length is not important - even long instructions are usually fast!

68060:
- The 060 is a radical change again
- big ICache, big DCache
- branch cache
- high clock rate
- instructions only need very few clocks (1-20)
- most instructions only take 1 clock!
- the 060 can often even execute 2 instruction in 1 clock.
- instruction length has some influence on the speed.

68050:
- big ICache, big DCache
- branch acceleration through logic
- very high clock rate
- instructions only need very few clocks (1-20)
- nearly all instructions only take 1 clock!
- instruction length is not important - even long instructions are execute in 1 clock only!

If you would ask me to "score" the CPU then I would answer the following. Giving just a single "Score" for a complex  machine like a CPU is of course always a inaccurate.

68000
- The good things about this CPU is that it belongs to the 68K family and that its cheap.
- Performance of the 68000 is very very low.
Performance Score: 1

68020
+ First 68k featuring the complete instruction set.
- Small ICache is better than nothing but its tinyness and the not existing DCache limit performance a lot.
- low clockrate limits performance. But higher clocking a 020 would make no sense anyhow as it has no DCache.
- Also negative is that complex instructions take longer
Performance Score: 4

68030
- Small ICache and Small Dcache, the small caches limit performance.
- Negative is that complex instructions take longer
Performance Score: 10

68040
+ The 040 is a real good CPU
+ big caches allow good performance
+ can handle long instructions without penalty
Performance Score: 20

68060
+ The 060 is a real good CPU
+ big caches allow good performance
+ can sometimes even do 2 instructions per clock.
Performance Score: 50

The 68060 is the best of the 68K CPUs.
Its really a good CPU and in a NATAMI it will even perform lot beter than in any old AMIGA.

In think in the NATAMI the 68060 CPU will reach a
Performance Score of around 100!
Because to the higher possible clockrate and the faster memory.

The 68050 is a good design also.
Whether the 68050 will be able to reach 68060 Natami perfromance levels - I don't know yet.
But the 68050 should certainly reach and outpass classic AMIGA 68060 performance levels.

The 68040 was really a "diva" and pretty hard to handle. First of all, it required quite some cooling by still running on +5V, thus it burned quite some power. Second, handling of the 68040 on exceptions was considerably difficult. Quite unlike the 060, it didn't follow the restart principle on exceptions, and software had to cleanup all pending writes in the 040 output push-buffer (which was a real hassle). The 060 was, again, considerably easier to handle, though motorola had quite some trouble of getting it debugged. First revisions were considerably buggy.

So long,
Thomas

68000 mulu 70 clocks
68010 mulu 40 clocks

68000 divu 140 clocks
68010 divu 108 clocks

I confirm your 68040/25 (A3640) vs 68030/25. I run A3000 with exactly these conditions. 3X speed about right for a variety of tasks.

, Michael

Thanks Gunnar for the relative info on the 680x0 family.  I only have high hopes for your 68050.

Once you have one up and flashed you hopefully will be able to create a test accelerator card for the A4000.  Then you will have no end of sites willing to look at it and compare and review it.

Unfortunately there aren't too many places willing to 'preview' Amiga stuff anymore due to a lot of other projects.

The upside of any comparison is, as you have said, that FPGAs will get cheaper and faster so whereas the MC68060 will remain forever stuck where it is; your 68050 will continue to get faster with each new generation.

A 32bit computer with a memory interface runnning at 50 MHz would be able to reach up to 200MB/sec memory throughput.
This would have been super for the time.

No no, the 68030 memory interfaces was not faster than the 68040.
But what was the case was that the AMIGA mainboards were based around a 030 memory interface. This made building a 030 CPU card very easy.

Jens Drößler wrote:

If I recall correctly my A1200 with Blizzard 1230III/50 wasn't that much slower than the 1240/25 in the usual widely available benchmarks...

Then the 040 card was broken or the testcases were it.

The 040 chip is a magnitude better design than the 030.

- The cache of the 040 was 16 times bigger than the 030 cache.
  The caches of the 030 are so tiny that they are often not really usefull.
- If you compare the integer unit of the 040@25Mhz with the 030@50Mhz then you see that at simple instruction both are equally fast but when doing long or complex instructions the 040 is twice as fast, despite the slower clockrate.

Cheers