Was anyone else disappointed with the A1200?

[roondar]

Originally Posted by activist

sure. Go to 19'40" here:

[ Show youtube player ]

the question was "Is there anything you would have done differently to Amiga before it was released".

That is very interesting and he did indeed say that with the benefit of hindsight it probably would've been better to do it that way. So I admit it, I was wrong.

And thanks for the video, I had never seen it.

That said, I do want to point out that he also very specifically pointed out that at the time of the Amiga's actual design, planar graphics were the better choice and that there were many things that he'd have done differently with the wisdom of 'today'. Maybe things would've been different had the man still worked at Commodore in 1989. Sadly, we'll never know.

Was just quoting from Dave Haynie public conference December 1993 in this old Amiga report edition:
http://www.amigareport.com/ar210/p1-3.html

You could probably easily explain why the 600 MB/sec figure is mistaken /theoretical if bothered

I already did that. The document I linked, written by Dave Haynie, shows it to be much less. Not just by virtue of the number I quoted earlier, but also by looking at things like the quoted Blitter speeds.

There is another reason for it to be extremely unlikely to be so high and that is VRAM speeds at the time. See, memory bandwidth is limited primarily by the speed of memory chips you put in (though you can play around with the bus width to improve speeds). In 1994, high end PC's and (later in 1995) even consoles such as the Playstation, 3D0 and Saturn didn't do much better than 100-200MB/sec on the internal video bus (and usually less on the external bus to the CPU).

Expecting Commodore to do a system that was up to 6x faster while using VRAM that was a couple of years older and designing on a shoestring budget? That's just not realistic.

Originally Posted by grond

Which is why you can safely conclude that the Amiga Doom was optimised quite a lot more in the several years from the publication of PC Doom and Mac Doom to Amiga Doom. This makes comparisons across different computer or even CPU architectures even less meaningful. As I said, if you really want to know, you can take the Amiga Doom sources, replace the c2p with a simple mem copy and run the timedemo. You will only get to see garbled stuff on the screen but the result (printed in the CLI) will be 100% accurate for how a chunky-AGA-Amiga would have performed. I have written my share of c2p routines and code using it, my estimates are good enough for me.

Like I said, like for like would require a setup I don't have. And sorry, but I'm not going to recompile DOOM just for this thread.

That said, you yourself pointed out that DOOM was not so complicated and really rather efficient at what it did when first released. You also said that DOOM didn't care as much about a fast processor and instead really needed memory bandwidth. Expecting a massive improvement in optimisation to the core game IMHO does not fit with these earlier statements.

Come on, use some common sense! An 8bit c2p routines needs four passes of two moves, four ands, two shifts and two ors (and some more). One pass operating on two registers worth of pixel data thus already takes 24 clock cycles. All four passes hence 96 cycles plus loading and storing (which are the same for our hypothetical chunky-AGA and thus do not count). You cannot hide that many clock cycles on an 030 in chipmem access times for the planar data you are producing because you are running at only 14 times the chipmem cycle speed.

On the 060 (and possibly on the 040) you can easily do c2p in fastmem/cache in very little time and copy the data to chipmem while doing other useful stuff in fastmem. You can also do the c2p from fast to chip at copy speed. But that really doesn't prove that planar graphics aren't a disadvatage but rather how little sense it makes to combine an abysmally slow graphics bus with a fast CPU. In a well balanced moderately priced setup planar graphics were a lot slower for a Doom type game.

Everything you post is technically true. And understand, I am not saying c2p should have zero impact at all. By now, I actually accept that the difference between planar and chunky come to about a CPU grade as you said earlier.

However, despite all of that, the videos I've seen show that such a small speed difference (a CPU grade would be anything from 10-20% clock speed difference) is hardly noticeable in practice. When I say "small enough it might as well not exist", I don't mean there is zero difference - just that it's not enough to matter.

This actually fits with your position that there is about a CPU grade of difference between the two - that is simply not a lot. The perceived difference in speed between a CPU and a CPU that is 15% faster is very minimal. The same goes for frame rates: sure, a 40MHz 68030 might end up being 15% slower than a 40MHz 386DX when rendering Doom, but that's not going to be all that notable (this would a difference on the order of 2-3FPS).

That is what I mean.

So yes, all things being equal, planar at the same CPU speed as chunky is slower for DOOM like games. But if you look at real world results, with optimised c2p you'll barely notice this difference - unless you specifically go looking for it. I actually tried Doom through GOG on my PC the other day, it ran at maximum speed: 35FPS. When playing, I couldn't tell the difference between it and the 68060 Amiga versions I saw on YouTube. Even though the 68060 versions ran (roughly) around 25FPS (a full 10FPS difference). Only if I actually put the two side by side does it become notable. And even then there's not much in it.

Yes, as little as somebody else's dislike for Doom-type games. But I was also expressing how the majority of people felt at the time when the topic of planar vs. chunky graphics was important.

I'm not saying he has a better point. I'm saying both of you don't have a point when you start claiming stuff about what games you like or dislike.

As for 'the majority of people'... This is a thread about a machine launched before games that genuinely benefited from chunky graphics became popular. I also don't remember the whole 'Amiga can't do Doom' thing to become 'big' until after 1994, so that makes me doubt this claim as well.

OK, then I retract my statement that 8 bit planar graphics mode have absolutely no advantage over 8 bit chunky graphics modes and correct myself: 8 bit planar graphics modes have no relevant advantages and very big relevant disadvantages.

Which is still just as false, but I'm going to stop this discussion here.

This isn't going anywhere and frankly, I have better things to do with my time than to discuss this subject if this is how you're going to react.

I guess they finally saw how little effort it was to support both chunky and planar modes and thus decided to do both instead of repeating the mistake of supporting only one of the two (the wrong one). This only underlines my point how simple it would have been to add chunky to AGA.

Which is not really relevant. My point was that they kept doing new planar stuff. You don't do that if you think it's useless. Had they only kept the 8-bit planar modes and not added any new things, that would've been different. But they didn't.

I have to admit I did not know how much bigger Apple apparently was and, since I'm not very familiar with any of their products, it seems hard to explain why they did so much better. We probably need another discussion about what Commodore did wrong while competing with Apple...

Not here please, this thread is big enough as it is

Yes, but today there is hardly anything wrong with PC hardware. I have a hard time admitting it but, while I still don't personally use Windows at home and have no need nor want to do so in the future, it is also OK to use in its modern form. Computing devices are now almost exclusively defined by the software they run. And since there are so many solutions available, it boils down to a matter of taste and preferences. Pepsi or Coke? A surviving Amiga could have made the path more convenient even if it eventually had led to more or less the same place.

You cut out the part where I point out that the Mac was already essentially a PC in 1996 (the only remaining difference was the CPU). I did not just mean 'today'. This started many, many years ago.

[chb]

Originally Posted by roondar

I already did that. The document I linked, written by Dave Haynie, shows it to be much less. Not just by virtue of the number I quoted earlier, but also by looking at things like the quoted Blitter speeds.

There is another reason for it to be extremely unlikely to be so high and that is VRAM speeds at the time. See, memory bandwidth is limited primarily by the speed of memory chips you put in (though you can play around with the bus width to improve speeds). In 1994, high end PC's and (later in 1995) even consoles such as the Playstation, 3D0 and Saturn didn't do much better than 100-200MB/sec on the internal video bus (and usually less on the external bus to the CPU).

Expecting Commodore to do a system that was up to 6x faster while using VRAM that was a couple of years older and designing on a shoestring budget? That's just not realistic.

I also do not see how the 400-600 MB/s figure is reached, however the linked document shows that it's not that far fetched: The dual VRAM system can do 1280x1024 in "chunky" mode (which is 16 bit per pixel here) in non-interlace, where the frequency is not specified. If we assume 60 Hz, we get about 160 MB/s. If we take the given pixel rate value of 110 MHz, we get 220 MB/s (probably a rather theoretical value). Add to that about 30 MB/s for blitter or CPU in chipmem (with VRAM display refresh has almost no effect on the bandwidth for other devices on the bus), plus fastmem bandwidth (64 MB/s for a 68040/40 in burst mode) and you're close. If you add the 68040's cache bandwidth, you may even arrive at the number quoted...

The consoles did not need high resolutions at that time, so the total memory bandwidth is probably lower, but the more important GPU-VRAM bandwidth is higher. And the dual VRAM system would have been much more costly than the consoles.

[roondar]

Originally Posted by chb

I also do not see how the 400-600 MB/s figure is reached, however the linked document shows that it's not that far fetched: The dual VRAM system can do 1280x1024 in "chunky" mode (which is 16 bit per pixel here) in non-interlace, where the frequency is not specified. If we assume 60 Hz, we get about 160 MB/s. If we take the given pixel rate value of 110 MHz, we get 220 MB/s (probably a rather theoretical value). Add to that about 30 MB/s for blitter or CPU in chipmem (with VRAM display refresh has almost no effect on the bandwidth for other devices on the bus), plus fastmem bandwidth (64 MB/s for a 68040/40 in burst mode) and you're close. If you add the 68040's cache bandwidth, you may even arrive at the number quoted...

So, if we give AAA all the advantages in the world we're still only slightly over half way to 400MB/sec

By the way, the document says that the pixel clock and data fetch rate are decoupled: you can have a higher (or lower) clock than the speed of data fetching if desired. I have no clue why they did this, but there it is. Still, it's better than I thought. Does make the following at the start rather strange, though:

Originally Posted by AAA docs

The AAA chip bus is an improvement over the ECS and even AA chip bus in terms of memory speed. The AAA chips run a four cycle burst to Chip RAM, which in raw performance is 4.56 times faster than ECS memory access or 1.14 time faster than AA’s two-cycle burst. However, the real key to AAA’s memory architecture is its support for VRAM. With VRAM, display fetches have practically no effect on the normal parallel chip RAM bus, freeing it for use by Blitter, Copper, and CPU

Moving on...

The consoles did not need high resolutions at that time, so the total memory bandwidth is probably lower, but the more important GPU-VRAM bandwidth is higher. And the dual VRAM system would have been much more costly than the consoles.

Nevertheless, 600MB/sec is an awful lot for a 1989-1993 design and your calculations showed it to be much less than that, albeit indeed more than I had figured.

[NorthWay]

Wasn't AAA designed for multiple memory layouts? From simple DRAM to multiple VRAM? Something like (at least) 4x difference in bandwidth?

[Mick]

With regards to Apple surviving both of my schools (junior/infant and comprehensive) had Apple Macintosh machines, if Commodore could have gotten into the education sector charging insane amounts of money to people with more money than sense then perhaps they would have had the funding to really push R&D.

[malko]

Originally Posted by activist

[...] Hypothetical conjecture, pointless going there again

I only share a piece of the puzzle and I do not pretend to show the whole picture but for sure, it's no "hypothetical conjecture". Simply the reality of the Amiga market in my country at that time period. The Amiga was nowhere professionally speaking, at least officially (not even in schools - but Mac & PC were). For a computer so "ahead" of its time, it always seemed strange to me...
Of course it may have been different in your country.

[roondar]

Originally Posted by NorthWay

Wasn't AAA designed for multiple memory layouts? From simple DRAM to multiple VRAM? Something like (at least) 4x difference in bandwidth?

It sure was: there were four versions planned (on paper at least). Ranging from a "single" version with DRAM for chip memory up to a "double" version with VRAM for chip memory. The numbers we've been discussing apply to the fastest of the four, the slowest could not actually display 1280x1024.

Looking back on it now and seeing how many chips they needed for the high end version I do kind of understand why it was scrapped. But the 16 year old in me still wants to see a working version, even though it'll never happen

[DofD]

Originally Posted by malko

I only share a piece of the puzzle and I do not pretend to show the whole picture but for sure, it's no "hypothetical conjecture". Simply the reality of the Amiga market in my country at that time period. The Amiga was nowhere professionally speaking, at least officially (not even in schools - but Mac & PC were). For a computer so "ahead" of its time, it always seemed strange to me...
Of course it may have been different in your country.

When I left high school in '86, my school had 3 Amigas for their art and design classes. My A1200 and one of my A3000's came from a university where they were used for audio and video work.

[DofD]

Originally Posted by roondar

It sure was: there were four versions planned (on paper at least). Ranging from a "single" version with DRAM for chip memory up to a "double" version with VRAM for chip memory. The numbers we've been discussing apply to the fastest of the four, the slowest could not actually display 1280x1024.

Looking back on it now and seeing how many chips they needed for the high end version I do kind of understand why it was scrapped. But the 16 year old in me still wants to see a working version, even though it'll never happen

Same here, it would be fun to have seen what a dual AAA system would have been capable of. With today's tech, most of the chips could probably be implemented on one or two pieces of silicon. It's too bad that as far as I know the chip designs have been lost.

[swinkamor12]

These videos with PS-2 386SX running DOOM is propaganda bullshit.
PS-2 386SX was sold by IBM in 1989 three years before a1200 realease.
386SX from 1992 with SVGA card from 1992 runs better.
Affordble SVGA card from 1992 has few times more bandwith than PS-2.

It was easy to add chunky pixels to AGA but Commodore failde to do it.

c2p? interleaved with other calculations?
Be serious, it is too much work and thats why there are no good 3D games on AGA.

AGA video quality was good enough in 90's and is still good enough.

Problem with AGA is this shit was too slow in 1992 and still is too slow.
AGA should have chunky pixels.

Communism style propaganda of success of some Amiga classic fanatics not change the fact that Commodore banckrupt because AGA has not chunky pixels.

[vulture]

@swinkamor12

Ah! So, there's a conspiracy going on from the AGA illuminaty! Now I see!

[roondar]

Originally Posted by swinkamor12

These videos with PS-2 386SX running DOOM is propaganda bullshit.
PS-2 386SX was sold by IBM in 1989 three years before a1200 realease.
386SX from 1992 with SVGA card from 1992 runs better.
Affordble SVGA card from 1992 has few times more bandwith than PS-2.

I also showed a video of Doom on a 386SX@40MHz and several videos of it on 386DX's, one of which had a VLB SVGA card. Not one of them ran better than the A1200 video Vulture posted.

Suppose I could go on spamming ever more YouTube videos of Doom on the 386 on here. Videos that everyone knows were secretly made by the New World Order to promote the return of the Amiga and the path to world domination. They do give me a lot of money for that stuff, so that's nice.

Oh. S**t. Wasn't supposed to say that...
Nothing to see here, move along!

[Dunny]

Originally Posted by swinkamor12

Be serious, it is too much work and thats why there are no good 3D games on AGA.

I enjoyed AB3D. It was an excellent game.
I enjoyed Breathless, it was great fun to play.
I enjoyed Fears, though not as much - and it certainly wasn't the graphics that made it a bad game.

So there's three.

[roondar]

Breathless was pretty nice, sadly never played AB3D.

I know it's not as popular and runs terribly on just about everything, but I spend days playing AB3D-II. Didn't even care about the low speed. Some people also rather liked Gloom and Genetic Species.

[daxb]

@Dunny:
What about Genetic Species?: http://hol.abime.net/2716

[chb]

Originally Posted by roondar

So, if we give AAA all the advantages in the world we're still only slightly over half way to 400MB/sec

By the way, the document says that the pixel clock and data fetch rate are decoupled: you can have a higher (or lower) clock than the speed of data fetching if desired. I have no clue why they did this, but there it is. Still, it's better than I thought. Does make the following at the start rather strange, though:

It's the way VRAM works: At a single read access, a whole row (typically 1024 bits) is copied into an internal buffer, which is then output on the second port with an independent clock, without affecting the normal dram bus. So AAA would have a DRAM bus for CPU and blitter with a quite slow 30 MB/s (1.14 times AGA), but that bus is not used for video refresh, which can use much higher bandwidth for serial data (let's estimate between 160 and 220 MB/s for AAA).

Originally Posted by roondar

Nevertheless, 600MB/sec is an awful lot for a 1989-1993 design and your calculations showed it to be much less than that, albeit indeed more than I had figured.

Not sure if much less... I forgot that Dave Haynie might have included a DSP with its internal RAM in his calculations, and then his figure is quite credible, even without counting 68040 cache bandwidth . IMHO the total bandwidth is not that absurd for a VRAM based system - it's just that it mainly helps in displaying high resolutions/high color depths, so it's very useful for classic GUI and maybe video applications. The bandwidth to the chipram is the bottleneck here, 30 MB/s is dog slow compared to PS1 or Jaguar which have 3-4x that speed (the Jaguar has to share that with nearly everything else) and is not improved much by using VRAM (just display dma does not eat further into it).

I think AAA would have been rather outdated already in 1994, nice specs but an expensive architecture from the 1980's not well suited for 3D or blitter-based 2D.

[roondar]

Originally Posted by chb

It's the way VRAM works: At a single read access, a whole row (typically 1024 bits) is copied into an internal buffer, which is then output on the second port with an independent clock, without affecting the normal dram bus. So AAA would have a DRAM bus for CPU and blitter with a quite slow 30 MB/s (1.14 times AGA), but that bus is not used for video refresh, which can use much higher bandwidth for serial data (let's estimate between 160 and 220 MB/s for AAA).

In AAA, VRAM is used as dual ported RAM. There is no separate DRAM for the Blitter and CPU, they also use (the same) VRAM.

Not sure if much less... I forgot that Dave Haynie might have included a DSP with its internal RAM in his calculations, and then his figure is quite credible, even without counting 68040 cache bandwidth . IMHO the total bandwidth is not that absurd for a VRAM based system - it's just that it mainly helps in displaying high resolutions/high color depths, so it's very useful for classic GUI and maybe video applications. The bandwidth to the chipram is the bottleneck here, 30 MB/s is dog slow compared to PS1 or Jaguar which have 3-4x that speed (the Jaguar has to share that with nearly everything else) and is not improved much by using VRAM (just display dma does not eat further into it).

I think AAA would have been rather outdated already in 1994, nice specs but an expensive architecture from the 1980's not well suited for 3D or blitter-based 2D.

Your calculations come to slightly over 274MB/sec (assuming 180MB/sec - which seems reasonable considering the 1280x1024 mode was limited to 72Hz). That's nowhere near 600MB/sec.

As for including DSP bandwidth, that is rather strange as AAA does not have one. It does have Mary, but that runs of the same bus as the Blitter, CPU and video.

Again: there were no other systems in 1993 that had bandwidth even remotely close to 600MB/sec. This is especially worth noting as Commodore was limited by the same RAM speeds as everyone else and didn't exactly throw money at AAA.

Anyway, perhaps any further AAA discussion should move elsewhere, it's not really on topic for the A1200.

[NorthWay]

Originally Posted by roondar

But the 16 year old in me still wants to see a working version, even though it'll never happen

Heck yea, I want AAA even as my grumpy old bastard self.
I even tinkered a bit with the Verilog of the Minimig to try to add in a Blipper as AAA does.

[chb]

Originally Posted by roondar

In AAA, VRAM is used as dual ported RAM. There is no separate DRAM for the Blitter and CPU, they also use (the same) VRAM.

Sorry, I think my statement was slightly confusing. VRAM has two ports: The DRAM port (that is addressed like standard DRAM) and a read-only video port. So for CPU and Blitter operations, VRAM and DRAM looks identical, as they only use the DRAM port. For display generation, the graphic chip has to put the row address on the DRAM bus at the right moment and read the data from the video port in sync with the video beam (during the latter one the DRAM bus is free).

Originally Posted by roondar

As for including DSP bandwidth, that is rather strange as AAA does not have one. It does have Mary, but that runs of the same bus as the Blitter, CPU and video.

A DSP was not part of AAA, true, but that does not mean it would not have been a part of an AAA Amiga. The DSP in the A3000+ also was not considered to be part of AGA. You're right, that's pure speculation, but again, we do not know what system Haynie was thinking about when making his quote.

Originally Posted by roondar

Again: there were no other systems in 1993 that had bandwidth even remotely close to 600MB/sec. This is especially worth noting as Commodore was limited by the same RAM speeds as everyone else and didn't exactly throw money at AAA.

Mac Quadra AV 840 from 1993:
https://everymac.com/systems/apple/m...dra_840av.html
About the same VRAM bandwidth as AAA, probably same main memory bandwidth, 66 Mhz DSP (same DSP3210 type as in the A3000+). If you count the bandwidth to the 8k of internal DSP RAM (that's of course debatable, but Haynie might have done so), and assume it can do a longword every clock cycle (very typical for a DSP), you'll arrive at ~500 MB/s total bandwidth. Not counting yet some blitter-like chip that works in VRAM while the CPU is busy (no idea if those Macs had something like that, but probably yes). Again, the distribution of this total bandwidth is very different from the consoles.

I think those Macs are a quite good comparison, as an high-end AAA Amiga probably would have been positioned roughly in the same class as the Quadra AVs. So nothing outlandish about such a bandwidth.

Originally Posted by roondar

Anyway, perhaps any further AAA discussion should move elsewhere, it's not really on topic for the A1200.

Yep, but a lot in this thread isn't . After all, I think we're all just discussing some old obsolete systems for our pleasure... But if any mod would move that branch to an own thread, I would not protest.

[Foebane]

I haven't even been able to follow this thread, too many walls of text and loads of technical jargon that make my eyes glaze over.