From: dhaynie@jersey.net (Dave Haynie) Newsgroups: comp.sys.mac.system,comp.os.ms-windows.nt.misc,comp.sys.powerpc.advocacy,comp.sys.amiga.misc,comp.sys.acorn.advocacy Subject: Re: What efficient computer system to choose? Date: Tue, 23 Sep 1997 05:18:28 GMT Organization: PIOS Computer Lines: 133 Message-ID: <34274a63.1599072584@news.jersey.net> References: <5vujnc$mrf$1@joe.rice.edu> <601qu5$1h3u@lightning.asahi-net.or.jp> NNTP-Posting-Host: swip130.jersey.net X-Newsreader: Forte Free Agent 1.1/32.230 Path: 195.95.96.10!news.unisource.be!news.att-unisource.net!newsgate.unisource.nl!surfnet.nl!howland.erols.net!news-peer.sprintlink.net!news.sprintlink.net!Sprint!newsfeed.internetmci.com!192.48.96.126!in2.uu.net!209.66.0.12!news.jersey.net!not-for-mail Xref: 195.95.96.10 comp.sys.mac.system:213626 comp.os.ms-windows.nt.misc:172290 comp.sys.powerpc.advocacy:15624 comp.sys.amiga.misc:154041 comp.sys.acorn.advocacy:11179 On Sat, 20 Sep 1997 03:55:16 BST, Mechanoid wrote: > In article <5vujnc$mrf$1@joe.rice.edu>, maliyekk@rice.edu (Anil Thomas >Maliyekkel) wrote: >> Ian Lynch (ian@ianlynch.demon.co.uk) wrote: >> : phtlb@zeus.bris.ac.uk (Tim Borgeaud) wrote: > >> : > I'd like to get an energy saving system that is as efficient as >> : > possible. > >> : The Digital StrongArm is one of the lowest power consumption chips >> : available. Its primarily designed for portable applications. This is true. DEC got involved because they really want to be more in the chip business than they have been; they have strong chip technology, and well, the computer business has been better. While the Alphas are clearly great for performance at any price applications, they had no chip for typical embedded applications. The ARM folks did, of course, but theirs was grossly underperforming compared to all the other embedded RISC spinoffs, from MIPS, SPARC, PowerPC, etc. when power wasn't quite as much of an issue (and we're not talking such a vast difference here). It was a good match. >>: In a Risc >> : PC it will allow you to do things like DTP and graphics manipulation >> : at comparable speeds to the fastest PCs and Macs It's not even a close match for a decent PowerPC or Pentium II. Of course, there's quite a bit more to it than that. The OS is one problem, and Acorn's RISCOS has plenty of shortcomings, but Mac or Windows-like bloat are not among them. And folks who think this is just a crutch to wack away at those more successful, think again. I have done tons of DTP in my time, 100 page manuals for things like Zorro III on the Amiga, other specs, etc. I recently had the displeasure of using Word for Windows, version 7, for a 150 page document for Scala, since it's their standard wordprocessor. And that, my friends, is some experience. With a minimum of graphics, a few fonts, some nested headers, running on a 150MHz Pentium under Windows NT, this wordprocessor's performance was abysmal. It was so slow in WYSIWYG mode as to be unusable. My A3000 running FinalWriter, also in WYSIWYG mode, also on a 150 page document, in practically the same resolution (1180x900 for the Amiga, 1280x1024 for the PC, and the PC's graphics engine is about 50x faster). There's no contest in CPU or memory speed here. The difference is just algorithm and OS. Of course, a good OS on a PowerPC is still going to run 2x-20x faster than a good OS on a StrongARM, or more, depending on just how much floating point matters, versus waiting for I/O and stuff (everyone waits at the same speed; a good OS lets the CPU do other stuff during the wait, Windows rarely does). >> The only problem is that the SA doesn't have a floating point unit. > That's not really much of a problem. Floating point isn't very important >for most users. Hence the ommission. Floating point isn't very important to embedded applications, hence the ommission. Make no mistake about it: RISCOS is small potatos. DEC wasn't making StrongARM for the desktop, and nothing they're doing is for anything other than making a fast chip for embedding, running off batteries, etc. > The problem is the FP emulator that RiscOS programmers rely on. A decent >FP library brings performance back to acceptable levels. Quake at 20fps on >a 200MHz StrongARM RiscPC, for instance. PC users run it at 30fps+ in 1024x768. I've done this myself (since Scala got their T1 connection, a few folks there have taken to playing net-Quake on some of the nicer PCs they keep around for showing off Scala). Floating point is critical to most modern games. It's not a big deal for wordprocessing and some other desktop applications, most multimedia-type things, pre-1992-style games, that sort. It's even more critical for 3D renedering, the kind of stuff the Amiga once ruled. > The thing that really strangles StrongARM is the 32MHz memory bus that >doesn't even support EDO RAM. :-((((( That's another thing that's no big deal when you're building embedded systems. A pure embedded application (eg, a camcorder or a TV set, not a more general purpose handheld computer) will often load up critical variables in the on-chip cache and then freeze them. Meanwhile, a faster local bus gets your DRAM and all eating more power. EDO is rather passe', modern systems use synchronous DRAM, which on 66-75MHz buses delivers about as much a boost over EDO as EDO does over FPM. Once you get to 100MHz, anything but SDRAM is a waste. In practical PowerPC systems, we've often seen ar doubling of the system performance going from 33MHz buses without L2 cache to 66Mhz buses with fast L2 cache. This is exactly what you need for desktop performance, it's not going into embedded systems. > Acorn *were* thinking of building a CHRP machine for the RiscPC II, but >with a ROMmable OS and StrongARM(s). CHRP didn't preclude other processors >or ROMmable OSes, Actually, it plays fairly well to any other big-endian processor. Obviously they'd have to write their own RTAS, which is normally in PPC machine code (compiled C), but all the Open Firmware code is CPU independent. Most of CHRP is based around common PClone standards, too. Had it kicked into high gear, it would have made PCI cards with Open Firmware drivers much more popular, again benefitting anyone supporting the CHRP firmware system, PPC or no PPC. >so Acorn could retain most of RiscOS and use a CHRP >motherboard. They couldn't use a CHRP motherboard, but they could modify a CHRP motherboard for use with a StrongARM chip and PCI bridge. That doesn't address the memory or cache controllers, but it does take care of everything on PCI. Obviously using a PowerPC bus would take care of that stuff too, but it's unlikely DEC would have done that. Plus, if you're stealing a bus, taking the x86 bus (well, one of them anyway) would get you even cheaper motherboards. > This was mostly in order to save money because Acorn was in a >state of decline and couldn't afford to design their own support logic any >more. Decline or no decline, reinventing the wheel is a bad way to spend money. > Anyway, PowerPC sux, StrongARM rulez! etc., etc... ;-) In your dreams... Dave Haynie Vice President, Technology PIOS Computer haynie@pios.de "...no RISC, no fun"