Should the A1200 have used a 28Mhz 68000 processor instead?

[idrougge]

What must have hurt the Falcon quite a lot was its memory decoding meaning that it could either have 1 MB, which is useless, or 4 MB, which was prohibitively expensive at the time. 2 MB was a just enough.
It's also odd that the STe had SIMM sockets, while the Falcon didn't.

[Photon]

Whatever caused Commodore's declining sales, it wasn't a slightly faster CPU in the A1200. Americans had already chosen PC, warts and all, by this time. They would happily pay $1300 for a computer that couldn't play games at all. So Mother could balance the checkbook once a month on it in between hourglass spins.

I think Commodore could have had a chance with the A1200 if the new graphics modes were chunky and legacy ones bitplane, coupled with getting gamedevs on board to support it, hardware docs and supporting it in OS so that it appeared and worked like existing RTG cards. AGA really was too little too late.

But as I said, Americans had already chosen PC.

[teppic]

Originally Posted by eXeler0

The interesting(?) question is why Atari chose a 16MHz 030 while using a 16-bit data bus and 24 bit adress bus. Couldn't they have saved a penny by using a 020 instead. Cheap ass Commodore aparently thought it was worth chosing a 020. (I have no idea what the prices were for 020 and 030 chips respectively at the time).

Edit: I'm talking about Atari Falcon which is contemporary with the A1200.

The Falcon matched 16bit RAM with an 030? That's bad.

Even the 68EC020 has a full 32bit data bus. It's why some of the early A500 expansions were a total waste of money, as they matched a 14MHz 68020 with 16bit slow RAM, and performance sometimes even dropped over the stock CPU.

The A1200 matched 32bit chip mem, 32bit fast mem, and a 32bit CPU, together with doubled clock rates on the CPU and the custom chips. It was architecturally a good fit. The 030 would have been even better, giving access to more than 16mb RAM, faster per clock speed, and the potential of an MMU.

[frank_b]

Originally Posted by teppic

The Falcon matched 16bit RAM with an 030? That's bad.

Even the 68EC020 has a full 32bit data bus. It's why some of the early A500 expansions were a total waste of money, as they matched a 14MHz 68020 with 16bit slow RAM, and performance sometimes even dropped over the stock CPU.

The A1200 matched 32bit chip mem, 32bit fast mem, and a 32bit CPU, together with doubled clock rates on the CPU and the custom chips. It was architecturally a good fit. The 030 would have been even better, giving access to more than 16mb RAM, faster per clock speed, and the potential of an MMU.

Except the 1200 ran at 7mhz and not 14mhz from chip ram unless running from the instruction cache. The CPU ran at half the speed. 16 bit accesses at 16 mhz vs 32 bit accesses at 7 mhz. They're both hamstrung accessing DMA-able RAM really. Both machines need fast RAM to really shine. I think Commodore went out of their way to preserve existing timing.

[TjLaZer]

Originally Posted by Amigajay

The A500 selling well into the early 90s had Commodore in a quandry, A500 sales were increasing every year so they had to decide to break this cycle and risk losing potential sales.
The A500+ was an awful machine to put out there, esp as the A600 was planned anyway to put production costs down, they should have waited for that (im guessing it was commodores way of trying to get rid of the excess A500 cases and motherboards) but no way should they have a faster cpu in it, it would have killed sales and alienated the millions of A500 owners in one swoop.

Good points, but in hindsight Commodore didn't know what would of happened if they released a faster CPU along with the ECS A500+. As we know, many frustrated users returned them due to ECS and OS2.x not being compatible with old A500 games.

But progress has to start somewhere. It would of been nice if the A500+ had a 14MHz 68000 at a minimum. But as you pointed out, if was their way to move more A500 stock and bring a small upgrade to the A500 until the newer AGA models came out.

[teppic]

Originally Posted by frank_b

Except the 1200 ran at 7mhz and not 14mhz from chip ram unless running from the instruction cache. The CPU ran at half the speed. 16 bit accesses at 16 mhz vs 32 bit accesses at 7 mhz. They're both hamstrung accessing DMA-able RAM really. Both machines need fast RAM to really shine. I think Commodore went out of their way to preserve existing timing.

Why would it run at 7MHz? The custom chips ran at 14MHz, the CPU ran at 14MHz, and they were all on a 32 bit bus with 32bit RAM. The benchmarks show a 4x increase in chip ram access speed over an A500 accessing its chip ram.

[frank_b]

Originally Posted by teppic

Why would it run at 7MHz? The custom chips ran at 14MHz, the CPU ran at 14MHz, and they were all on a 32 bit bus with 32bit RAM. The benchmarks show a 4x increase in chip ram access speed over an A500 accessing its chip ram.

Trust me it does. The CPU gets the same number of slots barring DMA contention as the 68k gets at 7 mhz. Memory is wider. It is *not* faster. Raster test it if you don't believe me! Why do you think adding fast RAM doubles a 1200's speed?

[teppic]

Originally Posted by frank_b

Trust me it does. The CPU gets the same number of slots barring DMA contention as the 68k gets at 7 mhz. Memory is wider. It is *not* faster. Raster test it if you don't believe me! Why do you think adding fast RAM doubles a 1200's speed?

As far as I know, the CPU was limited to 7MHz under OCS/ECS with chip RAM, but with AGA it was always 14MHz, with slow performance due to having to wait for access to the memory because of the chipset, something not an issue with code in fast RAM. The wider memory provides double bandwidth, which will obviously increase overall system speed.

[ImmortalA1000]

Isn't the problem that in the A1200 the memory bus isn't clock doubled like with the Amiga 1000, where the memory is effectively clock doubled to 14mhz so the 7mhz CPU and custom chipset can both see the chip RAM at 7mhz without seeing each other?

Does this also mean for games like Starglider II or Virus there isn't that much '7mhz 020' going on in the A1200.

[frank_b]

Originally Posted by teppic

As far as I know, the CPU was limited to 7MHz under OCS/ECS with chip RAM, but with AGA it was always 14MHz, with slow performance due to having to wait for access to the memory because of the chipset, something not an issue with code in fast RAM. The wider memory provides double bandwidth, which will obviously increase overall system speed.

The half speed (7 mhz vs 14 mhz) is the wait state. The ROM is likely running at 14 mhz however. Amiga reloaded will allow the 020 to hit chip RAM at full 14 mhz speed regardless of what the chipset is doing. That should boost workbench nicely.

[gururise]

The Amiga 1200 should have come with two important changes:

1. 2MB chipram + 2MB fastram
2. a HD floppy

[Jope]

Apparently the choice was 1MB Chip + HD floppy drive or 2MB Chip and a DD drive. We all know what they decided upon. :-)

[ovale]

Originally Posted by ImmortalA1000

Isn't the problem that in the A1200 the memory bus isn't clock doubled like with the Amiga 1000, where the memory is effectively clock doubled to 14mhz so the 7mhz CPU and custom chipset can both see the chip RAM at 7mhz without seeing each other?

Does this also mean for games like Starglider II or Virus there isn't that much '7mhz 020' going on in the A1200.

Are you sure? http://amiga.resource.cx/photos/photo2.pl?id=a1000&pg=3&res=hi&lang=en here I can see 150ns DRAM chips. They are enough for the 280ns as in the Amiga 500 etc...

[gururise]

Originally Posted by Jope

Apparently the choice was 1MB Chip + HD floppy drive or 2MB Chip and a DD drive. We all know what they decided upon. :-)

I wonder if Commodore would have went with a 68000 or 68010 clocked at 16.67Mhz if they could have saved enough money to add a HD floppy and 2MB Chip + 2MB Fast ram?

Doing so, the system would probably be faster than a stock A1200 due to the fast-ram and high clocked 68000 processor.

[meynaf]

Originally Posted by gururise

Doing so, the system would probably be faster than a stock A1200 due to the fast-ram and high clocked 68000 processor.

I don't think so. 68020 has instruction cache, 32-bit memory access and instructions using less clock cycles, so 14Mhz 68020 was in fact the right choice.

[roondar]

Originally Posted by frank_b

Trust me it does. The CPU gets the same number of slots barring DMA contention as the 68k gets at 7 mhz. Memory is wider. It is *not* faster. Raster test it if you don't believe me! Why do you think adding fast RAM doubles a 1200's speed?

Check! The A1200 CPU has the same 3.5Mhz speed to memory that all other Amiga's since the A1000 had. That said, Chip Ram is effectively (much) faster on an A1200 due to two factors:


1) The bus is twice as wide as OCS (as you've stated), doubling the maximum CPU bandwidth to memory from 3.5mb/s to 7mb/s
2) Properly used, AGA uses a lot less memory bandwidth for bitplane DMA compared to OCS, which means far more memory slots are available for the CPU and blitter even though there are no more slots in total

Originally Posted by meynaf

I don't think so. 68020 has instruction cache, 32-bit memory access and instructions using less clock cycles, so 14Mhz 68020 was in fact the right choice.

I agree. The 68020 was a good choice here.

The bad choice being the lack of updates to the Blitter & DMA system (outside of bitplane DMA)

[Samurai_Crow]

Originally Posted by roondar

The bad choice being the lack of updates to the Blitter & DMA system (outside of bitplane DMA)

Agreed totally!

[sandruzzo]

Originally Posted by roondar

Check! The A1200 CPU has the same 3.5Mhz speed to memory that all other Amiga's since the A1000 had. That said, Chip Ram is effectively (much) faster on an A1200 due to two factors:


1) The bus is twice as wide as OCS (as you've stated), doubling the maximum CPU bandwidth to memory from 3.5mb/s to 7mb/s
2) Properly used, AGA uses a lot less memory bandwidth for bitplane DMA compared to OCS, which means far more memory slots are available for the CPU and blitter even though there are no more slots in total


I agree. The 68020 was a good choice here.

The bad choice being the lack of updates to the Blitter & DMA system (outside of bitplane DMA)

32bit copper, with 32 bit blitter both at 14mhz would have been great! But a little bit memory bandwidth whould have been great too! Even a little 128-256k of fast mem will suffice!

[frank_b]

My pet peeves.

1) Same 7 mhz cycle access as ECS/OCS. I *bet* the 1200 was going to ship with a 68000 oroginally! I think they were going for 100% compatibility on games.
2) Bank switched palette access and separate high/low nibbles
3) 2 mb chip RAM limit.
4) losing sprites when using scrolling with higher fetch modes.

I'd love to see a DMA/CPU slot timing diagram for the 1200. My tests indicated even worse performance than I expected to chip ram with different fmode settings. Maybe something to do with my accelerator card?

[roondar]

Originally Posted by frank_b

My pet peeves.

1) Same 7 mhz cycle access as ECS/OCS. I *bet* the 1200 was going to ship with a 68000 oroginally! I think they were going for 100% compatibility on games.
2) Bank switched palette access and separate high/low nibbles
3) 2 mb chip RAM limit.
4) losing sprites when using scrolling with higher fetch modes.

Yeah, those things are kinda bad. Though the 2MB limit is not so bad considering a large percentage of users of the A1200 had a hard disk (and a large group of AGA games where hard disk installable) to spool in new data.

I'd love to see a DMA/CPU slot timing diagram for the 1200. My tests indicated even worse performance than I expected to chip ram with different fmode settings. Maybe something to do with my accelerator card?

That could be, most accelerators don't get anywhere near 100% chip ram access speeds. Faster accelerators tend to perform worse here. A base A1200 does about ~5.5 mb/sec read and ~7 mb/sec write when a 640x256x2 plane screen is shown.



It should (emphasis on should) scale from there. So a 320x256x8 screen requires 81920 bytes of bandwidth per frame using 1x fetch, and goes down to needing 20480 bytes of bandwidth per frame at 4x fetch.

Total chip memory bandwidth (@50Hz) is about 146000 bytes per frame so the percentage lost can be calculated that way. In other words, at 1x fetch the 320x256x8 screen should offer about 3 MB/sec and at 4x fetch it should offer about 6 MB/sec (write speed).

Anything below that is losses due to either other DMA or the A1200 processor bus not being effectively used by the accelerator you're using.