Path: utzoo!utgpu!news-server.csri.toronto.edu!bonnie.concordia.ca!uunet!lll-winken!iggy.GW.Vitalink.COM!widener!dsinc!bagate!cbmvax!daveh From: daveh@cbmvax.commodore.com (Dave Haynie) Newsgroups: comp.sys.amiga.advocacy Subject: Re: The 68050 - end of the 680x0? (was Re: The Amiga's Future) Message-ID: <22393@cbmvax.commodore.com> Date: 13 Jun 91 05:53:47 GMT References: <5068@orbit.cts.com> <16647@darkstar.ucsc.edu> < <1308@cbmger.UUCP> <28@ryptyde.UUCP> > <01dH!cmr@cs.psu.edu> <1991Jun10.072945.8821@neon.Stanford.EDU> <22365@cbmvax.commodore.com> <1991Jun13.003707.19785@neon.Stanford.EDU> Reply-To: daveh@cbmvax.commodore.com (Dave Haynie) Organization: Commodore, West Chester, PA Lines: 142 In article <1991Jun13.003707.19785@neon.Stanford.EDU> torrie@cs.stanford.edu (Evan Torrie) writes: >daveh@cbmvax.commodore.com (Dave Haynie) writes: >>In article <1991Jun10.072945.8821@neon.Stanford.EDU> torrie@cs.stanford.edu (Evan Torrie) writes: >>If you're building full speed systems, a 68030 is _better_ than twice as fast >>at the same clock speed as the equivalent 68020/68851 combination. > I've never seen twice the performance in CPU intensive applications. >Just as an example, the Speedometer benchmark on an original Mac II clocks >the CPU as 3.49, and my IIci as 6.71. Adjusting for clock speed, that's only >a 25% improvement. I don't have a clue about what Speedometer does to get its figures. I can pick Amiga benchmarks which claim the A3000 is anywhere from 2x to 10x the speed of an A2500/20. It all depends on what you're testing. Also, while the A3000 or Mac IIci are decent 68030 systems, consider that on both machines, the DRAM is still running less that 1/2 the maximum speed that the '030 is capable of. Burst mode helps some, but it's no substitute for 0 wait state memory. Cache is what you want, you aren't going to get DRAM going quite that fast (well, maybe a two-way interleaved 60ns memory system could give you 1 wait state...). Keep in mind that memory speed doesn't scale with CPU speed; the 68020/68851 system didn't use that much slower memory. All of this point out that it's difficult to build a full speed 68030 system. Basically, you don't for the budget me and Greg had for the A3000 and I imagine the Apple folks had for the IIci. HP did some killer, full speed 68030 systems, but they cost relatively big bucks. It's easier to build a 68040 system that goes close to full speed, since [a] the cache hits considerably more often, so external memory speed isn't quite as critical, and [b] the 68040 bus interface is just plain better -- you can run a whole clock faster with 80ns DRAM at 25MHz with the '040. >>> Why do I see this as the beginning of the end? Well, by all accounts, >>>Intel's 80586 will be as revolutionary a chip from the 486 as the 386 >>>was from the 286. >>Who's accounts. > EE Times, InfoWorld, PC Week for example. [Not exactly paragons of >journalistic integrity, but they're more often right than wrong]. I read EE Times on the can every week and didn't get that impression. They did point out that Intel is throwing most of their efforts behind 80x86s, RISCy things are now officially just a sideline. I guess Intel knows where their money is grown. The 80x86 people are a captive market, so Intel can gouge pretty heavily. Other companies make RISC chips. > Well, also because of the reports that Intel is consolidating behind the >80x86 family, and trying to bundle their two other CPU development teams >[the 860 (high-speed floating-point), and the 960 (superscalar RISC)] >in behind the 80x86 team. They're continuing to push the i860 and i960 into their established niches. The i860 as a graphics coprocessor and vector machine (it is showing up on all kinds of high end display boards), the i960 for embedded control, especially in the military and avionics markets. The i960 has nothing at all to do with the 80x86 line. They will have some kind of glue chip to make it easier to hook an i860 up to an 80x86 system. There's that second big bucks chip in the same box, as I mentioned before. > We have some hints from Motorola (i.e. the "evolutionary" upgrade). Where's the line between "evolutionary" and "revolutionary"? The 68020->68030 upgrade is generally considered evolutionary, although the speedup was 2x for fully outfitted, equivalent systems. Most people consider the i386->i486 move to be something closer to revolutionary, even though that's basically just around 2x improvement. Architecturally speaking, the i486 upgrade was much more of a revolution, but then again, the '040 was an even larger break from the '030. Anyway, looking at the Motorola track record, the '050 should be in the "evolutionary" category, which is tracked more on architectural grounds than performance upgrades: 68000,68010 16 bit 68020,68030 32 bit 68040,68050 Cool 32 bit Except for the 68000 to 68010 upgrade, each step so far has been at least a factor of two improvement. >>Also, keep in mind that Intel has much less to lose by wimping out on the >>'586 than Motorola does on the '050. > This depends on whether you consider the ACE consortium to be a serious >force. Well, no one else does :-) (EE Times was lukewarm on it, Microprocessor Report pretty unimpressed, since it doesn't really mean much of anything yet). >In fact, if you recall, one of the main thrusts of ACE is to allow Compaq to >force Intel to increase the speed of the 80x86 family, or else risk losing >the high-end market to the R4000. No, actually, it's Compaq trying to get out from behind Intel and IBM with something they can control. And it makes sense. They could fall in line behind some of the SparcStation cloners, but that puts them in a non-controlling position, just like they are in the Intel market. Actually worse, since Sun, IEEE support or not, has controlled the SPARC architecture, basic system design, and software. At least in the Intel market, you had that split three ways: Intel on the chip architecture, IBM on the system design, and Microsoft on the software. Worked good. And guess what? ACE gives you the same thing: MIPS on the chip architecture, Compaq|DEC on the system architecture, and Microsoft on the software. So, basically, I think ACE is an interesting but pretty enivitable happening (though relatively little has actually happened yet). No matter what Intel does, Compaq is likely to build ARC systems, assuming the ACE consortium can actually pull this whole thing off. >>Considering the number of processors Motorola sells to non-desktop computer >>users, like VME and other industrial applications, the Apple/NeXT/Amiga use >>of high end 680x0s, while not trivial, is the minority. > Even here though, the 88K is getting the push ahead of the 68K. Take >yesterday's Ford/Motorola announcement, for example. Sure, the 68K isn't for everything in that market. Not only that, but it's been so successful in that market that Motorola needs to make sure the 88K follows in its footsteps. But really, the 68K is alive and kicking, especially in VME. The 88K is on VME, but hasn't really gone far yet. Take a look through Computer Design sometime and check out all the '040 based VME cards out already. Ultimately, both Intel and Motorola have to "put up or shut up" with their CISC products. At present, anyway, it's easier for more Motorola customers to abandon the 68K than for Intel users to abandon the 80x86. Not that either is "easy" by any stretch of the imagination. And there's no guarantee that, when anyone abandons a 68K/80x86, that either Motorola or Intel will get the business. In fact, Intel's just about assuring that, if you give up the 80x86, you won't replace it with an Intel chip, as they're pushing their RISCs into niche markets. Motorola, I think wisely, is building the 88K to do the same things the 68K does. However, there's absolutely no assurance that a 68K user will necessarily pick up the 88K as a replacement; MIPSs and SPARCs, to name a few, are out now for both system CPUs, VME stuff, and embedded control. -- Dave Haynie Commodore-Amiga (Amiga 3000) "The Crew That Never Rests" {uunet|pyramid|rutgers}!cbmvax!daveh PLINK: hazy BIX: hazy "This is my mistake. Let me make it good." -R.E.M.