Path: utzoo!utgpu!attcan!uunet!mcvax!enea!diab!pf From: pf@diab.se (Per Fogelstr|m) Newsgroups: comp.arch Subject: Re: Blitters and design philosophy Summary: Use what's good for your design, without breaking your neck. Message-ID: <409@ma.diab.se> Date: 3 Aug 88 10:27:21 GMT References: <5254@june.cs.washington.edu> <76700032@p.cs.uiuc.edu> <480@m3.mfci.UUCP> <401@ma.diab.se> <1988Aug1.062659.25971@utzoo.uucp> Reply-To: pf@ma.UUCP (Per Fogelstr|m) Organization: Diab Data AB, Taby, Sweden Lines: 56 In article <1988Aug1.062659.25971@utzoo.uucp> henry@utzoo.uucp (Henry Spencer) writes: >In article <401@ma.diab.se> pf@ma.UUCP (Per Fogelstr|m) writes: >>... Do that with the CPU. Even with external hardware the CPU will not >>be able to generate the addresses fast enough. (You have to generate a new >>address each 50-100ns). > >Have you, pray tell, seen the manual for the AMD 29000? I have. Please >read it before proclaiming this performance to be beyond that of a CPU. As a matter of fact i do have an Am29000 manual right here beside me. And i have read it. I suggest that you read the NS8500 series manuals. If you do you will find out that the 8500 RGP has three ALU's working in paralell, one for addresses, one for data (error computation), and one ALU performing clipping and picking operation. And then there is also a line pattern computation going on at the same time. All this gives a 2 clock cycle/10Mpix drawing rate. My graphics memory glows in the dark :-). And Henry, even if a cpu is generating adresses often, is it usable? It's like looking at the clock frequency for a flipflop without taking feedback time in consideration. Perhaps, Tim Olsen (at Amd) can clear this discussion up. I would belive that he can present some figures on how fast the 29000 are doing graphics. And i'l try to match it with the figures from my own NS8500 design. And if i'm wrong, i will of course admit that ! >(The 29000 is available today, although it's not yet cheap.) At last something important. The specialized processor (NS8500 RGP) i am using costs less than $100. This processor needs a NS8511 BPU for each bitplane plus some buffers. Total cost for these chips c:a $300 for an 8 bitplane system exept memory and video dac's. But okay, when the Am29000 is below $100 i might consider. But i still have to do something to replace the BPU's so i'm able to process all bitplanes at a time. And i hope the code for doing the things the 8500 can, comes with it. > >To sort of paraphrase a comment the Mips people have made about memory: >think twice before building specialized hardware to do something a >general-purpose CPU can do, because many people are putting enormous >resources into making the g-p CPUs better and faster, and they may well >catch up with you -- probably sooner than you think. Exploiting mass- >market products can work better than trying to compete with them. >-- What i figure out from this statement is that Henry needs to look around a litle. You don't have to build your blitter yourself. There is good silicon on the shelfs today, and if it makes your product better, what's wrong with using it ? Especially when its cheap. Im not sayin that one should break one's neck to achive something that a CPU can do better because it's not worth it. BUT i dont belive that RISC processors are the "ultimate solution" to everything. I belive that a CPU like the 29000 for ex. will do a good job on a 1024x768 screen, but what about 2600x2048 ? Even a specialized processor will have problems here to make it look really good. But mostly because we must think about new ways to access the frame buffer. So i think, its best to put things in it's context. Use whatever solves your problem best in the current situation.