callen@ada-uts (11/12/87)
Is anyone aware of any Unix systems based upon the Fairchild Clipper
ship set? I got the data sheets on this set (well, board) maybe a year
agao and it looked like a nice clean architecture. I am curious how
its performance might stack up to, say, the AMD29000 or MIPS machines.
-- Jerry Callen ...{ima,ihnp4,harvard}!inmet!ada-uts!callenbcase@apple.UUCP (Brian Case) (11/16/87)
In article <56700004@ada-uts> callen@ada-uts writes: >Is anyone aware of any Unix systems based upon the Fairchild Clipper >ship set? I got the data sheets on this set (well, board) maybe a year >agao and it looked like a nice clean architecture. I am curious how >its performance might stack up to, say, the AMD29000 or MIPS machines. How fortuitous. I was just comtemplating generating a posting (read: flame) about an Electronics article about the Clipper. So now I get to bundle info with flame. The intergraph C32 personal computer/workstation is based on the Clipper. I hear it isn't so bad, but the performance isn't much greater than a 25 MHz 68020 based machine (I heard this from someone developing compilers for it. SPEA, a West German company, is apparently making IBM PC plug-in cards based on the Clipper. The CEO of SPEA claims (according to Electronics): "A 33-MHz Clipper-100, Seng says, runs six times faster than a 20-MHz Intel 80386." [I'm sorry, I just can't believe this.] Now the flame. Read in last week's Electronics Magazine an article about the new clipper which will run at an amazing (I say that sincerely) 50 MHz clock rate. At 50 MHz, I personally would expect much more than what they can deliver, but let's face it, they win hands down in the catagory of exploiting old technology (the 33 MHz version was implemented in 2.0 micron CMOS, the 50 MHz is probably in 1.5 or so). My beef is with the bar-chart showing what speed memory is needed to achieve claimed performance. They try to show that the Clipper has the upper hand because it requires only regular DRAM -- no cache. However, they go a little to far when they claim the the Am29000 requires 3 ns, yes three nano seconds, static RAM. I know that for single-bus-cycle memory accesses, fast access times are required, but the actual access time is more like 14 ns (admittedly, this is still quite fast). BUT, even the performance claims with caches were only talking about 2 cycle caches, which leaves quite a comfortable margin for implementation. And VDRAMs are hardly required in 3 ns speed grades. What burns me here is not so much the specific attack on my own work, but the fact that these STUPID INDUSTRY MAGAZINES WILL PRINT, HAPPILY, SUCH UNMITTIGATED BULL PUCKY!! THEY EVEN SEEM TO TAKE DELIGHT IN THIS STUPIDITY. I have seen it happen again and again. I mean, get this quote from the second paragraph of the article: "However, the latest 32-bit contenders are closing in on Clipper." Er, I thought it was the other way around, that is, the latest Clipper might be closing in on the other guys. I have seen these damnable publications make errors on our stuff too, both making it sound better than it is and worse than it is. Once you know what is really going on, you see how bad these jerks really are. Often, their words are no better than noise! Isn't there something we can do?!?! Should we all write letters to these editors? Should someone start a publication that actually has some technical accuracy? I am getting so fed up with this situation that I really believe there is room for a new trade magazine. I still remember the time I tried to get some RISC references printed along with an article in Computer Design (one of the WORST magazines): "I'm sorry, references are usually reserved for scholaraly journals. We're not like that." Well, that's for damned sure. We wouldn't want to actually educate or otherwise help our engineers, now would we. I know John Mashey has taken the initiative a few times. John, what do you have to say abou this situation? One possible comment is that this discussion doesn't belong in this newsgroup, so suggestions for other homes will be gladly accepted.
mash@mips.UUCP (John Mashey) (11/16/87)
In article <56700004@ada-uts> callen@ada-uts writes: > >Is anyone aware of any Unix systems based upon the Fairchild Clipper >ship set? I got the data sheets on this set (well, board) maybe a year >agao and it looked like a nice clean architecture. I am curious how >its performance might stack up to, say, the AMD29000 or MIPS machines. we don't have many benchmarks in common, but (excerpts from MIPS Perf Brief): Dhrystone: Dhyrs Rel System 1757 1.0 11/780, VAX/VMS 8309 4.7 30MHz Clipper, InterGraph InterPro 32C, GreenHills 13000 7.4 MIPS M/500, 8MHz, -O3 23700 13.5 MIPS M/1000, 15MHz LINPACK: FORTRAN DP & SP MegaFlops & Rel: FORT FORT System DP Rel. SP Rel. .14 1 .25 1 11/780, VMS .29 2.0 .45 1.8 30MHz Clipper, as above (Opus slightly faster) .60 4.3 .93 3.7 M/500 1.2 8.6 2.3 9.2 M/1000 Doduc (Monte Carlo simulation, heavy FP, relative performance): 1.0 11/780, VMS 1.3 30MHz CLipper, as above 3.8 M/500 8.8 M/1000 Whetstones, DP & SP, relative perf: DP SP System Kwhets Rel Kwhets Rel 830 1.0 1250 1.0 11/780, VMS 1740 2.1 2980 2.4 30MHz Clipper, as above 4220 5.1 5430 4.3 M/500 7960 9.6 10280 8.2 M/1000 Of course, some of these numbers may be old, and you should always take what one vendor says about another with a grain of salt. We did have some Byte benchmark nubmers also, but I know they're old, and the Clipper kernel numbers were so low (about 1.5X VAX performance) that I have to believe the kernel was unoptimized, or something). Also, I assume the newer parts run 1.1X faster (33MHz). **OPINION** From the limited data that we have, and comparing it with other machines (like Sun-3/200s, VAX 8600s, if we were selling it, and had to put 1 single number on it, we'd call it a 3 or 3.5 vax-mips system (the integer performance may be 3-4X overall, but the FP is 2-3, and the kernel performance is probably lower (due to small caches)). If somebody has recent benchmarks, we'd love to see them. -- -john mashey DISCLAIMER: <generic disclaimer, I speak for me only, etc> UUCP: {ames,decwrl,prls,pyramid}!mips!mash OR mash@mips.com DDD: 408-991-0253 or 408-720-1700, x253 USPS: MIPS Computer Systems, 930 E. Arques, Sunnyvale, CA 94086
mash@mips.UUCP (11/17/87)
In article <6745@apple.UUCP> bcase@apple.UUCP (Brian Case) writes: ...complaints on awful magazine practice.... >Isn't there something we can do?!?! Should we all write letters to these >editors? Should someone start a publication that actually has some >technical accuracy? I am getting so fed up with this situation that I >really believe there is room for a new trade magazine..... >I know John Mashey has taken the initiative a few times. John, what do >you have to say abou this situation? .... 0) At the rate of speed this business moves, writers and editors are hard-pressed to keep up, even when they try very hard. I have some sympathy for them, as they are beset by tons of information and argumentative folks. [Quick: explain why RISCs are good, but every RISC vendor has a different view of why theirs is better, and why Moto thinks both RISCs and CISCs are good! Sort that out in 2 columns..] This is why you see "vendor claims ....N mips"...: how does a harried writer know one way or the other? 1) Some try hard, some don't. Some will print almost anything, even gushing about performance leaps on products that aren't yet built, and without even a SINGLE actual benchmark number [the cited Electronics article was like this; however, I've seen other articles there that were the same, so it's not unusual.] Since many of the trade rags are controlled circulation, you can't usefully threaten to cancel your subscription! 2) About all you can do are the following: a) Write articles that are reasoned, and send them to magazines. Sometimes they might get accepted and run. b) Write non-flaming, well-reasoned letters-to-the-editor citing inaccuracies, or noting the lack-of-substance in an article. 3) I've written a lot of letters-to-editors. About 1 in 10 gets printed. However, I often get reasonable phone calls from writers / editors looking for more information, or thanking me for the letter, so I can't complain. 4) Regarding performance claims, perhaps letters can encourage magazines to at least follow a few simple rules and answer a few questions: a) Give me some real benchmark numbers, not just mips & mflops. Especially, don't tell me that it's 20X a VAX (on some Dhrystone), give me the Dhrystone numbers (usual caveats: substitute other benchmarks; opinions on Dhrystone well-known). b) Is this on a real system? Can we buy this system now? How much does it cost? c) If it's not on a real system, is it a careful simulation, or a guess? If it's a simulation, give me the parameters, like what kind of memory system, etc. d) If it's not available now, when will it be available? e) Always be skeptical about what one vendor says about another, especially with little or no backup data. 5) In general, it is hopeless to improve some of the rags, which are little above the National Enquirer. Some of the magazines try very hard, even to having their own benchmark suites which they want to watch running on a real machine. -- -john mashey DISCLAIMER: <generic disclaimer, I speak for me only, etc> UUCP: {ames,decwrl,prls,pyramid}!mips!mash OR mash@mips.com DDD: 408-991-0253 or 408-720-1700, x253 USPS: MIPS Computer Systems, 930 E. Arques, Sunnyvale, CA 94086
doug@edge.UUCP (11/19/87)
>0) At the rate of speed this business moves, writers and editors >are hard-pressed to keep up, even when they try very hard. More correctly, that should read, "At the rate of speed this business produces new product announcements and other marketing hype..." >4) Regarding performance claims, perhaps letters can encourage magazines >to at least follow a few simple rules and answer a few questions: Not likely. With only a few exceptions, magazines are simply a promotional tool. (This is true of most fields, not just computers). It is usually against editorial policy to publish any unfavorable information about any company/product which is, or could become, a subscriber. Any magazine which is provided "free to qualified subscribers" is going to be a "puff piece publisher". Don't expect otherwise. -- Doug Pardee -- Edge Computer Corp., Scottsdale, AZ -- ihnp4!oliveb!edge!doug
einstein@bucsb.UUCP (11/23/87)
>Doug Pardee -- Edge Computer Corp., Scottsdale, AZ -- ihnp4!oliveb!edge!dou >Not likely. With only a few exceptions, magazines are simply a promotional >tool. (This is true of most fields, not just computers). It is usually >against editorial policy to publish any unfavorable information about any >company/product which is, or could become, a subscriber. > >Any magazine which is provided "free to qualified subscribers" is going to >be a "puff piece publisher". Don't expect otherwise. Doug you almost get a cigar with this one, but I disagree... The basis of controlled - circulation (meaning give it away if they've got anything to do with it) is based on one of the following: (1) Its an enhanced mailing list; with the advertisers sharing costs some magazines have their editorial written by their advertisers. (2) Its an easy way of getting wide coverage for your advertisers. (3) Paid - Subscription usually gets much smaller circulation and costs a lot more to get !!! Puff-piece... yeah... why not "you get what you pay for" on the other hand I don't think that its so much "editorial policy" as the bigwigs (read: advertising sellers) which keep the company in bread and butter don't want you cutting down their clients on page 1... Look what happened with Digital Review. DR published some "leaked information" and was banned from: (1) DEC press conf. (2) all other outlets of DEC - controlled information (3) No DEC advertising. What are you going to do when the companies you're writing about are your sole source of income? Besides there are no magazines that survive based on subscription income. Only newsletters get away with that and the really profitable ones charge over $100/year. It is possible to become a successful "controlled -circ " its just that nobody is going to help you out. david -- ============================================================================== David K. Fickes Center for Einstein Studies/Einstein Papers Project UUCP: ...harvard!bu-cs!bucsb!einstein Boston University BITNET: oth932@bostonu 745 Commonwealth Avenue PHONE: (617) 353-9249 (617) 277-9741 Boston, MA 02215
mash@mips.UUCP (John Mashey) (11/24/87)
In article <1276@bucsb.UUCP> einstein@bucsb.UUCP (David K. Fickes) writes: ... >don't want you cutting down their clients on page 1... Look what >happened with Digital Review. DR published some "leaked information" >and was banned from: (1) DEC press conf. (2) all other outlets of >DEC - controlled information (3) No DEC advertising.... Actually, it turns out that DR got "reinstated" pretty quickly: too many DEC product managers complained up the line that DR was an important channel for them. [source: DR people] Maybe this does show it's possible, anyway. There certainly is a fairly symbiotic, often love-hate relationship amongst vendors and trade press. -- -john mashey DISCLAIMER: <generic disclaimer, I speak for me only, etc> UUCP: {ames,decwrl,prls,pyramid}!mips!mash OR mash@mips.com DDD: 408-991-0253 or 408-720-1700, x253 USPS: MIPS Computer Systems, 930 E. Arques, Sunnyvale, CA 94086