Path: utzoo!utgpu!news-server.csri.toronto.edu!bonnie.concordia.ca!uunet!comp.vuw.ac.nz!actrix!templar!jbickers From: jbickers@templar.actrix.gen.nz (John Bickers) Newsgroups: comp.sys.amiga.advocacy Subject: Re: Blitter vs. 040 (was: Computer Architecture question Message-ID: <3658.tnews@templar.actrix.gen.nz> Date: 12 May 91 03:03:44 GMT References: Organization: TAP, NZAmigaUG. Lines: 95 Quoted from by melling@cs.psu.edu (Michael D Mellinger): > I would be interested in hearing the results for both the color and > monochrome machines. The monochrome NeXT does contain 2-bit gray Then go to comp*next* and find out. > Really? Faster than 2 68030s at this sort of memory intensive > operation? > other factors that come into play though, as someone has already Yes, these "other factors" are that it's a memory intensive operation, into an area of memory that is traditionally relatively slow. If the 2 68030s have two independent places to draw in, they'll whomp the 68040 which has one. > Let's say the 030. An 040 should be > than the 68000 + blitter. No? On a machine with the kind of CHIP bus an Amiga 500 has, for example, I believe the Blitter can clear the screen faster than any CPU. If it can't do it faster, it can certainly do it in about the same time. If we make a little diagram... Blit 040 | XX <- off the scale | XX | XX | XX | XX xx -+- XX <- maximum throughput on CHIP bus XX | XX XX | XX XX | XX XX | XX XX | XX ------+------ The 040 is faster, but the Blitter is as fast as it needs to be. What makes the A3000 CPU faster than the blitter is that the maximum throughput on the CHIP bus has been increased (doubled?). And what makes a 030 CPU on previous models better for certain cases is "effective throughput". For example, the blitter may be maxing out, but perhaps 10% of the data it is moving is superfluous, and is avoided by the CPU which is driven by more customised code. > How much does coprocessing matter. Let's say that you can get 10 mips It matters a lot when you have differences like this in what you're dealing with. For example, suppose (and these figures are pure imagination on my part, it's amazing how many people aren't EEs in this thread :) a 68040 can do 100 things on the slow device in 1 second, and 1000 things on the fast device in 1 second. If you swap it between the two devices, in the .01 of a second that it takes to do a slow operation, you could have had it doing 10 fast operations. If you have two jobs getting equal time, and one takes 1000 slow operations, the other takes 10000 fast operations, and you swap every .01 of a second, then your slow job will take 20 seconds instead of the optimal 10, and your fast job will take 20 seconds instead of the optimal 10. Now if you add a coprocessor that takes 1 slow operation to trigger, and can do the rest itself... If you move the figures around (how often you swap, etc), you'll see how the relative performance balances, and you'll also see that the coprocessor system should win most of the time, until it starts taking as many operations to set it up as it does to do the slow job oneself. > and for other overhead. How well does the A3000 do animation in its > million pixel display mode? That might give us some indication of > what the NeXT can do. Why? Isn't the Amiga 3000 a better design? > Context switching just needs to be handled properly. Ray tracing for > example isn't going to require a lot of blitter time. So, in fact you It's not going to require any blitter time. The line drawing is the blitter intensive task, and you can keep it going full bore at that. > -Mike -- *** John Bickers, TAP, NZAmigaUG. jbickers@templar.actrix.gen.nz *** *** "Endless variations, make it all seem new" - Devo. ***