rbannon@cosmos.acs.calpoly.edu (Roy Bannon) (09/27/89)
I was thinking of getting a FPE. I was wondering how much it would help? Does anyone out there in netland have a gs with a transwarp and a FPE. If so, can you notice a mesurable difference with and without the FPE? Can it be placed in any slot? Specifically, can it go in slot 6? (I don't use my 5.25" drives anymore). Is the sane tool file compatible with sys 5.0? Replies can either be posted or emailed to me. Thanks a lot! Roy rbannon@cosmos.acs.calpoly.edu
gwyn@smoke.BRL.MIL (Doug Gwyn) (09/28/89)
In article <1989Sep27.024212.1824@polyslo.CalPoly.EDU> rbannon@cosmos.acs.calpoly.edu.UUCP (Roy Bannon) writes: >I was thinking of getting a FPE. I was wondering how much it would help? It makes a TREMENDOUS difference in floating-point computation, especially with a compiler like Orca/C generating in-line code to use it directly (as a slot peripheral) rather than via the SANE tools. >Does anyone out there in netland have a gs with a transwarp and a FPE. >If so, can you notice a mesurable difference with and without the FPE? Well, you'd get a dramatic speed-up in floating-point operations, although the integer-CPU SANE emulations obviously would run faster using the TransWarp GS. The FPE is still many times faster. However, you should be advised that the FPE doesn't work with the TransWarp GS as it is shipped. One of the AE technical support people told me of a mod that AE will make to your TransWarp GS to prevent the problem. I made the mod myself (technically voiding my warranty) but for some unknown reason mine still are incompatible. Maybe U22 pin 1 isn't really on the upper left as I had thought? Anybody who knows for sure or who can look at their AE-modified TWGS, please let me know which pin of U22 was lifted, or if some other mod was made. >Can it be placed in any slot? Specifically, can it go in slot 6? Yes, it's just a slot peripheral. The installation disk that comes with the FPE includes SANE takeover INIT files for all slots. The installation program peeks at the slots and tells you where it found the FPE card, then when you tell it to install the INIT it copies the appropriate one onto your system disk. >Is the sane tool file compatible with sys 5.0? Yes, I was using it fine via both SANE and direct slot code before I installed the TransWarp GS. If you can ever get them to coexist, you'll have a real cruncher.
jabernathy@pro-houston.cts.com (Joe Abernathy) (10/03/89)
Network Comment: to #652 by adm!smoke!gwyn@nyu.edu I use the two together. I can tell you that my computer's extremely fast, but just how fast is hard to say. Particularly where a math coprocessor is concerned, you get into all kinds of subjectives ... what kind of work do you do with your machine, how often does your work involve number-crunching, and is the potential for time savings worth the considerable cost. The answer to every question is different with every user. The FPE speeds up things like spreadsheets, some graphics, and some things that are really hard to measure, such as FOR loops in programs written in certain compilers. The TWGS is a more broad-based accelerator, but as someone mentioned, its acceleration in any one area of system performance is pretty wimpy compared to the FPE. The thing is that the TWGS accelerators almost everything -- so it is the first accelerator most people should buy. If you're into fractals, or astronomy, or math, or anything similar, the FPE could well be a better investment. For amusement, I designed some loose -- very loose -- benchmarks to try to compare my system's performance with other machines. I used the FPE, TWGS, Multi-Kache disk accelerator w/SCSI option (getting an effective disk read of about 10 ms), and caching. My GS is comparable to, if not a little faster than, a Mac II, and it runs circles around an IBM clone 386 with a clock speed of 20 MHz. Curiously, however, a IIe running with a 10 MHz Rocket Chip is faster than my machine. (Actually, this isn't so curious. My tests were based on the Finder environment, and there just isn't any IIe software that takes up comparable resources to the Finder ... so there can be no valid comparison.) As for incompatibility between FPE and TWGS, there is. Applied Engineering did the fix, and at the time I discovered the imcompatibility, they set a policy of making the upgrade free of charge. Knowing AE, however, it would be worth a call before sending in your card to see what the policy is today. Another compatibility problem with FPE occurs if you try to compile programs written under Micol Advanced BASIC. This problem arose once, was fixed by innovative Systems, and then came back up under v3.0 of Micol BASIC. Somebody is ignoring programming standards ... and at this date, the problem has not been fixed. UUCP: crash!pro-houston!jabernathy ARPA: crash!pro-houston!jabernathy@nosc.mil INET: jabernathy@pro-houston.cts.com
paul@pro-europa.cts.com (Paul Hutmacher) (10/05/89)
Comment to message from: jabernathy@pro-houston.cts.com (Joe Abernathy) > Curiously, however, a IIe running with a 10 MHz Rocket Chip > is faster than my machine. (Actually, this isn't so curious. My tests were > based on the Finder environment, and there just isn't any IIe software that > takes up comparable resources to the Finder ... so there can be no valid > comparison.) Aw come on Joe, we're not talking about any //e. We're talking about pro-europa, the fastest //e machine on the block! Actually, I feel the comparisons were valid for eight bit programs and will challenge any IIgs with whatever speedup device to a launch and run an eight bit program. Those 10 MHz chips are fast. | UUCP: [ucsd, nosc] !crash!pro-europa!paul | "All that is necessary for the | | INET: paul@pro-europa.cts.com | triumph of evil, is for good | | ARPA: crash!pro-europa!paul@nosc.mil | men to do nothing." |
philip@pro-generic.cts.com (Philip McDunnough) (10/07/89)
Network Comment: to #5787 by jabernathy@pro-houston.cts.com
In what sense have you found the GS+TWGS+FPE to be faster than a 20MHz
386(with a 387 I assume) as well as being faster than the MacII? Can
you provide some standard benchmarks(eg.MIPS or MFlops or...)? Which
compilers are you using?Do you know whether the TWGS has a problem
with the ROM3.0 GS's?Finally,is there a 16 bit Fortran for the GS?
(forgot:is the data path between the cpu and the FPE a 16 bit one or
8,because if it is 8 then your results hard to accept given the 32
bit data path between the cpu's and the math chips on both the MacII and
a 386).
Philip McDunnough
philip@utstat.toronto.edu
[standard disclaimer]jabernathy@pro-houston.cts.com (Joe Abernathy) (10/11/89)
Network Comment: to #1054 by philip@pro-generic.cts.com
> In what sense have you found the GS+TWGS+FPE to be faster than .. etc.
Philip, you have to keep in mind first and foremost that speed is a subjective
thing. The big thing affecting actual speed -- and actual speed is all that
matters -- is what kind of load the CPUs are being asked to pull. I try to
design tests that reflect this, but don't even pretend they're perfect. Most
benchmarks are perfectly useless, and I'm aware of this.
The next thing that's important to remember that speed here isn't speed there.
I spent over an hour on the phone with Bill Mensch yesterday, and one of the
things he stressed at length was that 20 MHz of Macintosh is equal to 5 MHz of
Apple II, owing to design considerations in the main microprocessor the
machines are built around. (In particular, the formula is always Macintosh
MHz/4 = Apple II.)
Finally, you have to look at what a particular architecture is designed to do.
Steve Jobs wanted a desktop publishing system, and he came up with the finest
one available. But do anything else with a Mac and you've got one overburdened
machine on your hands. Sure, you can buy a Mac IIci with supporting
peripherals for major megabucks and get around the problem, but I don't regard
that as an elegant solution. In almost everything you do -- other than desktop
publishing -- machines such as the Apple II and IBM clones have a dramatic
performance advantage over Macintosh.
As recently as a year ago, believed a lot of the heresay that says you can't
do serious work with an old Apple II, or IIGS, but after taking the time to
find out the truth, I think the only thing you can't do is trust the marketing
departments of Apple and IBM. (The formula for measuring IBM performance
against that of any Apple II, by the way, is generally: IBM MHz/5 = Apple II
... meaning you're gonna have to come up with something like a 60 MHz 80386
box in order to outrun me.
The point here is don't believe everything you here in ComputerCraft. Most of
those guys used to sell shoes for a living, and in my opinion they should
never have abandoned a trade for which they obviously are so well-suited.
Joe Abernathy
UUCP: crash!pro-houston!jabernathy
ARPA: crash!pro-houston!jabernathy@nosc.mil
INET: jabernathy@pro-houston.cts.comphilip@pro-generic.cts.com (Philip McDunnough) (10/12/89)
Network Comment: to #6055 by jabernathy@pro-houston.cts.com Thank you for the comments Joe.I am aware(even more now!) that clock speed is not everything.For instance one has RISC computers with relatively slow clock speeds with high MIPS/MFLOPS ratings.There is the cycle/instruction issue as you have alluded to.However,the 68881/68882 are very good math coprocessors which function at their best if they BYPASS SANE.I don't no if the FPE does that.(no->know).For floating point operations clock cycles aren't that important.The speed of the coprocessor is however. A good way of seeing what happens is to generate a Mandelbrot set.Even then you would have a problem as the GS has fewer pixels to worry about than a MacII.Would you have any reference indicating timing for such operations with the GS(+TWGS+FPE). Philip McDunnough -> philip@utstat.toronto.edu University of Toronto
gwyn@smoke.BRL.MIL (Doug Gwyn) (10/14/89)
In article <8910130419.AA19703@trout.nosc.mil> philip@pro-generic.cts.com (Philip McDunnough) writes: >However,the 68881/68882 are very good math coprocessors >which function at their best if they BYPASS SANE.I don't no if the FPE does >that.(no->know). The Innovative Systems FPE card knows nothing about SANE; it works in any slot-based Apple II, not just the IIGS. When you buy the FPE card you also receive some software that: (1) patches BASIC.SYSTEM to use the FPE card instead of emulating floating-point internally (2) installs an INIT file for GS/OS that patches the toolbox to redirect SANE calls to code that uses the FPE card (3) patches AppleWorks (I think) to use the FPE card ByteWorks' ORCA/C will generate direct calls to the FPE card (slot) if you invoke a magic #pragma. Otherwise it uses the SANE tools, which may be sped up if you installed the FPE's INIT file.
jabernathy@pro-houston.cts.com (Joe Abernathy) (10/15/89)
Network Comment: to #1223 by philip@pro-generic.cts.com > Would you have any reference indicating timing for (generating the Mandel- > brot set) with the GS(+TWGS+FPE). Sure. You can get the specifics in my August 1989 article on this subject in inCider magazine. Roughly, and depending very much on which fractal program you're using and how it's set up, the time is reduced from about 12 hours on a plain IIGS to about 40 minutes with both accelerators installed. Speaking of accelerators, I have the parts coming in to do the 12 MHz upgrade on my TWGS. I would appreciate your E-mailing me if you have done this upgrade on your GS. I'm writing an article about it, so you'll get to see your pontifications in print ... assuming my editors don't laugh me out of the room. :) Joe Abernathy UUCP: crash!pro-houston!jabernathy ARPA: crash!pro-houston!jabernathy@nosc.mil INET: jabernathy@pro-houston.cts.com
philip@pro-generic.cts.com (Philip McDunnough) (10/16/89)
Network Comment: to #6215 by jabernathy@pro-houston.cts.com Sorry Joe, Haven't gotten around to putting all the pieces together.Is this 12MHz upgrade frome AE?Be interesting to compare the same fractal program on a variety of platforms.12 hours to 40 minutes is quite something!Whay not compare it with a Mac+(since this seems to be a favorite with some users here),an 8086+8087(use PC-Transporter),a MacII(),a 20MHz 386(with cache)+387. Philip McDunnough ->philip@utstat.toronto.edu ::University of Toronto : [my own opinions]
jabernathy@pro-houston.cts.com (Joe Abernathy) (10/26/89)
Network Comment: to #1453 by philip@pro-generic.cts.com
>Why not compare it with ... (everything).
Wow. I'd like to do that. Really. But I don't have the time or expertise to
write equivalent code on seven platforms, and I sure don't have the money to
buy them.
But if you get any ideas, let me know.
UUCP: crash!pro-houston!jabernathy
ARPA: crash!pro-houston!jabernathy@nosc.mil
INET: jabernathy@pro-houston.cts.comjabernathy@pro-houston.cts.com (Joe Abernathy) (11/12/89)
In-Reply-To: message from philip@pro-generic.cts.com > ... concerning the data bus of the Floating Point Engine as compared to > the 68030. The FPE employs a 16-bit data bus. > I'm somewhat confused about these comparisons. It's pretty easy to get that way when you start thinking about all of the issues that are concerned in obtaining valid performance comparisons of hardware. In fact, the staff of Byte has been consumed with this very pursuit for years. The only problem is that I'm not sure how many people care or even should care about that level of detail. My post was intended only to point out that the performance of the Apple IIGS is quite a bit better than what it gets credit for ... and that this performance can't be measured in megahertz alone. UUCP: crash!pro-houston!jabernathy ARPA: crash!pro-houston!jabernathy@nosc.mil INET: jabernathy@pro-houston.cts.com