oolidjr@hubcap.UUCP (09/15/87)
Hello out there in NET land ...
I am sorry if this has been discussed before on the NET, but I am
bumfuzzled about this one, I also prob. spelled that word wrong, but ...
What is the deal with the clock speed rating on the INTeL co-processor
chips??? My brother just bought a 10Mhz, 0 wait state AT clone. The
dealer claims (along with a few people at our institution) that the
coprocessor runs at (2/3) * main processor clock speed.
My question is: What is the correct answer??? Seems to me that the
clock speed is distributed out of the clock generator to both the
coprocessor and the main processor (8086, 80286, what have you ...).
If you put a coprocessor that is rated (2/3) the speed of the input
clock, you introduce alot of room for timing errors, do you not??
Or am I missing something here??
Thanx for any help. If this is in the interest of the net, post.
Otherwise mail me and I will post a summary.
Joe Moll
Mail Boxes:
ARPA: oolidjr@hubcap.clemson.edu
Internet UUCP: (any major hub)!gatech!hubcap!oolidjr
or ARPA: jlmoll@prism.clemson.edu
jlmoll@eureka.clemson.eduperkins@bnrmtv.UUCP (Henry Perkins) (09/15/87)
In article <454@hubcap.UUCP>, oolidjr@hubcap.UUCP (Joe Moll) writes: > What is the deal with the clock speed rating on the INTeL co-processor > chips??? My brother just bought a 10Mhz, 0 wait state AT clone. The > dealer claims (along with a few people at our institution) that the > coprocessor runs at (2/3) * main processor clock speed. This is correct, but is a "feature" of IBM's AT architecture and not of the Intel chips used. IBM used a 12 MHz crystal in the original AT, with a divide-by-2 for the 80286 and a divide-by-3 for the 80287. This is possible on the 80286 machines because the CPU and FPU are asynchronous, whereas they weren't on the 8086 machines. IBM did this to allowed their ATs to use cheaper lower-speed 80287 chips. Consequently my original Compaq DeskPro (with 8086 and 8087 at 7.14 MHz) is a better number cruncher than most AT-type machines (with 80287 running at 4 or 5.33 MHz). -- {hplabs,amdahl,ames}!bnrmtv!perkins --Henry Perkins It is better never to have been born. But who among us has such luck? One in a million, perhaps.
coffee@aero.ARPA (Peter C. Coffee) (09/16/87)
In article <454@hubcap.UUCP> oolidjr@hubcap.UUCP (Joe Moll) writes: >My dealer claims (along with a few people at our institution) that the >coprocessor runs at (2/3) * main processor clock speed. > >My question is: What is the correct answer??? Seems to me that the >clock speed is distributed out of the clock generator to both the >coprocessor and the main processor (8086, 80286, what have you ...). 8088, 8086, and 8087 run at board clock rate. 80286 runs at 1/2 board clock rate. I don't know about the 80386. 80287 runs at 1/3 board clock rate. I don't know about the 80387. Does this mean an 8 MHz Compaq DeskPro (8086/8087) runs intensive floating point faster than a 10 MHz 80286 machine with the same waits? Yes. Shrewd assembly code that really uses the numeric chip's stack, or even reasonably well-compiled high level languages, show this ratio in a matrix multiply benchmark that we swiped from Digital Review. You can, in principle, feed the numeric chip its own clock rate, since the main chip uses an FWAIT to force synchronization when needed; at least one add-on product does precisely this.
davidsen@steinmetz.steinmetz.UUCP (William E. Davidsen Jr) (09/16/87)
In article <17263@aero.ARPA> coffee@aero.UUCP (Peter C. Coffee) writes: This was the question: |In article <454@hubcap.UUCP> oolidjr@hubcap.UUCP (Joe Moll) writes: |>My dealer claims (along with a few people at our institution) that the |>coprocessor runs at (2/3) * main processor clock speed. And this was the answer: |8088, 8086, and 8087 run at board clock rate. |80286 runs at 1/2 board clock rate. I don't know about the 80386. |80287 runs at 1/3 board clock rate. I don't know about the 80387. This isn't correct. The speed of the chip is not provided by the clock fed to it, and there is no reason why the 80?87 can't be run at the same speed as the 80?86, or even faster. For example, there is a company which sells a product which plugs an 80287 runing at 10MHz into the socket of an existing 80287. This allows an original AT (6 MHz 80286, 4MHz 80287) to run the 80287 60% faster than the main CPU. This can be done because the 287 and 387 are run independently of the main CPU. The 8087 (as I read the manual) is run at the same speed as the CPU. About the 80387: it can be run at 16MHz. Faster parts are promised. What you are calling "board clock" is the speed of the crystal. Although it is traditionally divided by 2 using an Intel clock chip or builtin clock logic, the "speed of the chip" is rated by the output clock signals. -- bill davidsen (wedu@ge-crd.arpa) {uunet | philabs | seismo}!steinmetz!crdos1!davidsen "Stupidity, like virtue, is its own reward" -me
coffee@aero.ARPA (Peter C. Coffee) (09/19/87)
In article <7385@steinmetz.steinmetz.UUCP>...bill davidsen writes: >In article <17263@aero.ARPA> coffee@aero.UUCP (Peter C. Coffee) writes: > >|8088, 8086, and 8087 run at board clock rate. >|80286 runs at 1/2 board clock rate. I don't know about the 80386. >|80287 runs at 1/3 board clock rate. I don't know about the 80387. > >This isn't correct. The speed of the chip is not provided by the clock >fed to it...For example, there is a company which plugs an 80287 running >at 10MHz into the socket of an existing 80287. You're kidding, right? You never noticed the CRYSTAL(!!!!) on the little daughterboard that holds said 80287??? I specifically said, you can feed the numeric chip its own clock if you want to; and yes, the 8087's execution unit can also run asynchronously with its control unit. Geez Louise--plug the chip in to 5VDC and off it goes at 10MHz...wait til the crystal manufacturers find out about this...mutter, mutter.
coffee@aero.ARPA (Peter C. Coffee) (09/19/87)
Munching... Let me abandon anecdote and go to the source. From Intel Doc.210920-002, p.4: "The 80287 can operate either directly from the CPU clock or with a dedicated clock. For operation with the CPU clock (CKM=0), the 80287 works at one-third the frequency of the system clock (i.e., for an 8 MHz 80286, the 16 MHz system clock is divided down to 5.3 MHz). The 80287 provides a capability to inter- nally divide the CPU clock by three to produce the required internal clock (33% duty cycle). To use a higher performance 80287..., an 8284A clock driver and appropriate crystal may be used to directly drive the 80287 with a 1/3 duty cycle on the CLK input (CKM=1)." This explains the picture on p.128 of the June 1987 Byte showing a MicroWay daughterboard with a 12 MHz crystal nestled up against an 80287. The CKM line, pin 39 if anyone cares, would appear to be being driven high, an option described on p.2 of the aforementioned Intel document: "a HIGH input will cause CLK to be used directly." OK, to this degree the decision to let the chip divide by three *was* an AT design decision, but it sure was imitated with enthusiasm...
davidsen@steinmetz.steinmetz.UUCP (William E. Davidsen Jr) (09/22/87)
In article <17428@aero.ARPA> coffee@aero.UUCP (Peter C. Coffee) writes: |In article <7385@steinmetz.steinmetz.UUCP>...bill davidsen writes: |>In article <17263@aero.ARPA> coffee@aero.UUCP (Peter C. Coffee) writes: |> |>|8088, 8086, and 8087 run at board clock rate. |>|80286 runs at 1/2 board clock rate. I don't know about the 80386. |>|80287 runs at 1/3 board clock rate. I don't know about the 80387. |> |>This isn't correct. The speed of the chip is not provided by the clock |>fed to it...For example, there is a company which plugs an 80287 running |>at 10MHz into the socket of an existing 80287. | |You're kidding, right? You never noticed the CRYSTAL(!!!!) on the little |daughterboard that holds said 80287??? I specifically said, you can feed |the numeric chip its own clock if you want to; and yes, the 8087's execution |unit can also run asynchronously with its control unit. | |Geez Louise--plug the chip in to 5VDC and off it goes at 10MHz...wait til |the crystal manufacturers find out about this...mutter, mutter. Not only was this "quote" taken out of context, but it was also *edited* to insure giving the incorrect information. I would have replied by mail, but the offending posting has no signature and the purported poster's address is returned to me as "no such user". My original text is available by mail if anyone missed it. -- bill davidsen (wedu@ge-crd.arpa) {uunet | philabs | seismo}!steinmetz!crdos1!davidsen "Stupidity, like virtue, is its own reward" -me
coffee@aero.ARPA (Peter C. Coffee) (09/28/87)
In article <7427@steinmetz.steinmetz.UUCP> davidsen@crdos1.UUCP (bill davidsen) writes: : > >Not only was this "quote" taken out of context, but it was also *edited* >to insure giving the incorrect information. : I don't know about your postnews program, but mine specifically asks me to "trim the quoted article down as much as possible." I'm sorry that you feel the result presented your comments out of context, but I vigorously object to the assertion that this was deliberate. Why should I reconstruct your message into a form that makes my stomach churn merely for the dubious pleasure of replying to the result? A sense of proportion, please. So far, I have provided a short answer to a short question; I have been told, without substantiation, that I was wrong; I have provided an extensive sup- porting excerpt from the relevant Intel document; and now I am being accused of the worst sort of intellectual dishonesty. No wonder it's so hard to get a question answered anymore. My use of sarcasm may have been inappropriate, but it's a long way from there to the sort of ad hominem attack in your reply. How about a presumption of good faith and a mutual interest in getting the *right* answer all around?