View Single Post
Old 16 July 2019, 13:56   #488
Registered User

Join Date: Jul 2015
Location: The Netherlands
Posts: 1,364
Originally Posted by grond View Post
For 030-class CPUs you'll lose something like 20 to 30% to c2p at low framerates. The loss will be higher for increasing framerates. E.g. if you run Quake at 1fps the added effort for c2p is negligible. If you want to run Wolfenstein at 40fps, well, you can't.
Surely higher frame rates do not make c2p relatively more expensive? They should make drawing graphics more expensive at about the same rate (irrespective of chunky or planar). The limits of c2p on the Amiga have a lot to do with chip memory bandwidth and not with frame rate per se: there is only a maximum of about 7mb/sec available to the CPU (in a best case scenario) no matter how much or little GFX you wish to draw.
Having to deal with c2p is not very attractive to a coder. It has been a sport for Amigans to find fast c2p routines and, as already stated, it took several years to arrive at the solutions we have had since the late 1990s. None of them make the planar disadvantage disappear. It is a nuisance.
I never said it wasn't a nuisance, I said you could still run Doom on an Amiga with a fast CPU with reasonable results even with c2p. That's not actually the same thing as there not being at a disadvantage. The point here wasn't that it was no problem, but that it also had to do with other things than just the AGA chipset - merely changing AGA in the A1200 into a fast chunky display would not have given the vanilla A1200 Doom.
That's why Commodore should have offered 020 and 030 CPU cards right from the start just like they had the 030 and 040 for the A4000. And looking at how Ambermoon runs OKish (for an RPG) with floor and ceiling textures on the unexpanded A1200, I think a chunky mode would have meant a considerably larger 3D window.
Sure. I never said there would be no advantage to chunky. I was only talking about Doom on a base A1200.
I tend to disagree. One 320x200 frame are 64KB. At 30 fps you would spend about 30% of the total available CPU time just for copying the data to chipmem. Even a fast 030 cannot do 8bit c2p at copy speed, so the percentage becomes even larger. If, on the other hand, chipmem had been as fast as normal RAM of the time minus the screen DMA, you would simply have drawn the buffer in chipmem, updated the DMA pointers and be done with the frame.
A genuine question: if what you say about the game not actually needing CPU grunt but rather memory bandwidth is true, then why is spending 50 or so % of the available CPU time drawing pixels a problem? You clearly have memory bandwidth to spare at that point?
Furthermore, Doom really doesn't need much CPU grunt, it needs memory bandwidth above all. All the renderer does is to collect pixels from textures and put them in their place. Since the binary space partition tree algorithm is so good, it hardly ever needs to draw a pixel twice. But it needs to set each and every pixel once per frame. The added stuff is a DIV per pixel column on the walls and ceilings. That's not really computing intensive.
I'm not so sure I agree. Certainly, give Doom more memory bandwidth and it runs a lot better, but even then... Well, take a look at these four videos of Doom running first on a 386 & 486 with the same ISA VGA card and then on a 386 and 486 with a VLB graphics card (which has lots of bandwidth). Edit: for fun, I've added Doom on the A1200 with a 68040@50MHz.

See here for the ISA 386: [ Show youtube player ]
And here for the ISA 486: [ Show youtube player ]
Here for the 386 VLB: [ Show youtube player ]
And here for the 486 VLB: [ Show youtube player ]
Edit: Lastly, here is the A1200 with a 68040: [ Show youtube player ]

Now, it's clear that neither ISA card runs the game as well as the 486+VLB does, but there's still a pretty visible difference between the two ISA versions - the 486 clearly does better. More to the point, the 386 with the VLB card is visibly slower than the 486 using ISA.

Clearly, running the VLB 486 version is by far the best - so bandwidth does help. But without a fast CPU to back it up, Doom just does not run that well even if it has bandwidth to spare. The 386 VLB version ran very poorly, even though it had a ton of video bandwidth. Only when the fast memory bandwidth was coupled with a fast CPU did we see frame rate go up to much higher levels.

And lastly, the A1200 with the fast 68040 does a reasonably good job as well: it's clearly faster than either 386 even with the c2p penalty and it seems to be faster than the 486+ISA as well. Edit: I know the A1200 does not run full screen, but it seems to be 320x200, which is identical to the PC version.
Doom can run well enough on an 030.
I've never seen that happen myself. Just take a look at Doom on a 68030 Mac (which has chunky graphics and fast processor access to video memory*): [ Show youtube player ]

Maybe the Mac version is horrible, but I've never seen Doom run well on a 68030. That said, if you have a better example I'll be sure to take look
*) Compared to AGA at any rate: the 68030 in the Mac IIvx can push 4 bytes to VRAM every 8 cycles = 16MB/sec write speed.
That's not how I remember it. Originally the AGA features were already planned for the A3000 but Commodore pushed the A3000 out of the door before it was finished giving us "ECS". Then nothing happened for another few years...
Interesting, I remember it very differently. I don't remember anything about AGA features on the A3000 at all - only the A3000+ that became the A4000 instead. Now, I'm more than willing to accept I'm wrong, but I would like some more info to be sure.

Do you know anywhere I can find this stuff?
The A600 was a total desaster and actually resulted in heavy losses. It came in 1991. When it was demonstrated inside Commodore, the staff went "wtf, even more of the same old, same old?"

Commodore even made it for another three years after that. AGA+chunky should have come with the A3000 in 1989, then things might have gone differently. Commodore wasted all that time and development budget on the stupid 8bit computers and then couldn't compete when there were better machines on the market. Those machines didn't come out of nowhere, they had been in development for some years, too...
A small correction: the A3000 was first sold in 1990 and the A600 went on sale march 1992 (only a couple of months prior to the A1200 ). I do remember stories about the staff being disappointed with the A600 but as I understand that had more to do with the result than the original plan: the original plan was the A300, a cut down A500 on the cheap.

But the A600 was more expensive than the A500, while not offering any real benefits over it to those who didn't want an (expensive) hard disk, so it failed. Some of the first A600 motherboards actually say A300 on them, which is kind of funny I guess.
Yes. The question is how much too long they had waited. They didn't understand that the evolving computer business absolutely required them to start developing the next big thing the very moment they had their newest product out the door.
To be fair, the generation of computers they had made their money on all had unusually long cycles. Take for example the NES: a 1982 design still on sale in 1992. The upgrade to that was released in 1990, eight years after release. Or the 286, a 1982 design still put in new products (that managed to sell) in 1990.

This is not to say you don't have a point, merely that it was not just Commodore that wasn't putting new stuff on the market quickly and that the market mostly seemed to accept older stuff. This did rapidly change around the early 1990's and they should've seen that coming with the specs of some of the newer stuff on the market.

So I'm not giving them a free pass or anything.
To me the Archimedes always was a mystery machine that was supposed to be vastly superior to everything else but that nobody had ever seen. It was a unicorn.
It was indeed very rare outside of the UK, but the specs are simple to find as are plenty of examples of the machine in action that show that it indeed is rather interesting and fairly powerful for the time. Note that I never owned one. I got interested in them after hearing the Amiga developer of Pacmania comment in an interview that the Archimedes version of that game was best.

And then I started looking it up and checking YouTube videos and it is indeed an impressive machine.
Yes, because they were the biggest of the remaining bunch and Microsoft needed a living competitor to avoid monopoly related troubles. Commodore could have been the biggest remainder. They once were the No. 2 in the computer business behind IBM.
They were once No. 1, though that did not last very long. It is a giant shame what happened to Commodore and mostly self inflicted.

A good cautionary tale of what happens when you let the owners of a company focus purely on personal and shareholder gain and not consider the rest much. Sounds sadly familiar to how Apple is being run today.
I remember reading a discussion about this. Thomas Richter knows for sure. I seem to remember that he said something along the lines that compiling for 020 would only give a few percent more speed but would increase the maintenance and testing effort.

EDIT: I have just seen what Toni Wilen has written. I guess we can conclude that at least parts of the available 3.0+ kickstart versions are compiled for plain 68000.
I certainly accept that. It was merely me wondering as the kickstart refused to start and now I know why.

Last edited by roondar; 16 July 2019 at 14:03. Reason: Added an A1200 with 68040 to the comparison video list
roondar is offline  
Page generated in 0.04137 seconds with 11 queries