Path: utzoo!utgpu!news-server.csri.toronto.edu!mailrus!cs.utexas.edu!swrinde!zaphod.mps.ohio-state.edu!usc!rutgers!cbmvax!daveh From: daveh@cbmvax.commodore.com (Dave Haynie) Newsgroups: comp.sys.amiga Subject: Re: Real System Comparisons Message-ID: <13678@cbmvax.commodore.com> Date: 7 Aug 90 17:12:54 GMT References: <13466@cbmvax.commodore.com> Reply-To: daveh@cbmvax (Dave Haynie) Distribution: usa Organization: Commodore, West Chester, PA Lines: 98 In article mwm@raven.pa.dec.com (Mike (Real Amigas have keyboard garages) Meyer) writes: >In article eachus@linus.mitre.org (Robert I. Eachus) writes: >1) The standard disk interface for high-end IBM PCs is ESDI, not SCSI. IBM themselves are moving to SCSI. >While I'm willing to accept those as substitutes for each other, there >are still some people who think that SCSI doesn't belong in a "real >computer." Considering that [a] SCSI is significantly faster than EDSI (5 MB/s for SCSI-1 vs. 1.5 MB/s for EDSI), and that [b] SCSI is the de-facto hard disk standard in every Amiga, Mac, NeXT, and a whole slew of Workstation class machines like Suns, I can only wonder what "real computers" are using these days. >2) EISA slots - systems with more than a couple of EISA slots tend to >be '486 based, not '386 based. You only need a system with 4 empty EISA slots to stack up to the A3000. >Allowing for those two changes, then Dave's "equivalent" systems start >showing up around $3300 or so. Neither are acceptible changes, performance-wise. Especially considering that most of the EDSI controllers are CPU-drive, not DMA with buffer/FIFO. That's not going to be an exceptionally big deal under MS-DOS, but once you go to OS/2, UNIX, or some other multitasking OS, the hit you take reading a slow, non-intelligent disk is significant. >Except that the "equivalent" Amiga to that clone would be: [...] > socket for Wietek 4167 or equivalent That's silly. The 68040 floating point performance is supposed to be in the same ballpark as the 4167. If you're dealing with a 1 MFLOPS math processor like the '486, maybe it does make sense. If you want something beyond 3.5 MFLOPS, get a 25 MFLOPS DSP chip or an i860. >And the 3000 just doesn't stack up. You can pay $10,000 for a machine like that. Probably more. I'm not looking for a server, just a desktop PC that's roughly equivalent to the A3000 for the same price. Or a different price, for that matter. The closest Mac is the IIci, which costs more, has a low performance SCSI interface, and one fewer 32 bit slot, but does support some kind of direct CPU bus slot and has a better on-board display than the A3000. >The point of all this is that (as I said before) price comparisons of >arbitrarily defined "equivalent" hardware is pretty meaningless. You >have to look at what you want to use the system _for_. Of course, if you need one specific application, you find the best system that'll run that one application. >This is only slightly less true if the application is "unreleased" on one of >the boxes and released on the other. If you need it TODAY. Of course, if you know the application WILL be on a system that's designed to handle it, it may be worth the wait, rather than getting that application now on a system that really isn't set up to run that applcation very well. >If your purpose calls for raw CPU or FPU power, and little else (i.e. >- all you do is spreadsheet recalcs), then the IBM boxes provide more >bang - either for the buck, or total (I think - anyone want to comment >on that Wietek chip?). The Weitek (pronounced "way-tek"; it's derived from the Chinese word for "micro" or something like that) chips make lots of sense in PCs. The WTL 3167 is a single chip containing a Weitek FPU core and glue logic to adapt it to the '386 bus (this used to be a circuit board, called the WTL 1167). It cranks out about 1.36 MFLOPS (handcoded Linpack, single precision), vs. about 0.3-0.4 for a '387, at 25MHz. The 3167 contains 32 single-precision registers which can be paired to all 16 double-precision registers, but it doesn't support extended precision. The numeric formats conform to the IEEE standard 754, Version 10.0. It supports data movements of constants, registers, and to/from the '386, and conversions to/from integers or between precisions. It supports the four basic arithmetic functions, multiply-accumulate, and square root. It also supports comparision functions. Being a memory mapped device, the 3167 op-codes map into a series of '386 MOV instructions, rather than following a coprocessor protocol like the '387 does. Weitek originally announced an '030 bus part, the WTL 3168, which they rated somewhat faster than the 3167, but they have since cancelled the part. There has been a 4167 announced for use with the '486 bus, though I don't have any information on that one. Since the 3167 was faster than the '486's math, I would expect the 4167 to be a bit faster than the '486 too, but the main point of a 4167 might just be compatibility, unless they've made significant tweaks to it. Motorola claims 3.5 MFLOPS on Linpack for the '040, so I'm sure there's no reason to consider a 4168 for the '040 bus. >