EOAHMAD%NTUVAX.BITNET@cunyvm.cuny.edu (06/14/91)
NL%"adrie@PHILICA.ICA.PHILIPS.NL" Adrie Koolen wrote >I must by saying that I'm no 486 expert, but if I've understood it well, >the `one cycle' is execution time. Of course, the instruction first has to be >fetched from cache or external memory (which can also be externally cached). >Only when fetched from the internal cache, the pipeline is not stalled. >I don't know how fast a burst mode memory access is. Maybe someone can clarify. Read the article In the IEEE MICRO FEB 1990 by John H. Crawford, Intel titled The i486 cpu: Executing Instructions in One Clock Cycle. RISC microprocessors and i486 can produce results in one cycle consecutively but each one would still neet at least 3 cycles. If you have instruction cache, all instructions must be fetched from the cache. In case of a miss then Intel uses burst mode. Burst mode is twice faster because it does not have to send the upper half of the addresses which do not change. Even RISC requires fast program memory in order to execute in one cycle. >Are you sure that both processors mentioned run at 25MHz? I doubt that! >Not that I want to praise my SparcStation IPC into heaven, but in general I double checked, I am sure they all run at 25MHz. >tests, my 25MHz SparcStation IPC is as fast as a 33MHz 486, of course >depending on the speed of the disk subsystem, etc. On some tests, the 486 >is faster, in other tests the Sparc. Anyway, I've yet to meet the PC expert, If we want to compare microprocessors only, leave out the disk oriented tests. >who can sit behind an unknown PC, work for an hour with MS windows, >Borland C++ and WordPerfect and tell exactly what processor on what speed >is in it and what the seek and data transfer numbers of the disk are :-) We can find out, but what is the point? >I think you're comparing apples with oranges. The SparcStations have a frame >buffer with only one display mode, which is bit-mapped. They're not designed >for fast text scrolling. I've recently compared the speed of the graphical Typing make is faster than going through all the menu routines. Most of the people that I saw use the X-window systems use them as text based windows. >display systems of my color SparcStation and 33MHz 386 PC with super VGA in >1024*768*256 mode under MS Windows. I looked at display operations like >refreshing the screen and copying rectangular areas on the screen. The Why do you use 256 colour mode when you need only 16 most of the time. My MS windows with 640X480X16 on 386-25 is as fast as the Sparcstation. There is no point to use 1024X768X256 mode because my software does not support it. >SparcStation managed to copy some 4MB per second (scrolling in an xterm >window), the PC stopped at 500KB per second. If you've used MS Windows at >this resolution, you'll know that it's terribly slow on display operations. >That's not the fault of the 33MHz 386 nor the ET4000 VGA chip, but from the >AT-bus. The AT-bus can go at 16Mbit/s. The 32 bit EISA can do 33Mbyte/s. It is not the AT bus, it is the VRAM. You need dual ported VRAM. Why don't you try VGA display cards with cache from Trident. >BTW. Framebuffer access on the SparcStation IPC takes some 7 cycles, so you >can access 4 pixels (32 bits) more than 3 times per us. Does it have 4 bit mode? >I won't argue on that, but such a generalized statement doesn't say much. >If the systems running dos and running Minix were compareble, then you've >got a point, but so many programs only run under dos and not under Minix >that it's useless to say that "my Informix SQL database query under dos >runs more slowly that under Minix, if it would exist". Don't get me wrong, >you might be right, but don't underestimate the importance of the availability >of applications under certain operating systems! Dos does not have elle. Some source codes are written for Unix systems only. It would be hard to compile it under Dos. I had to use generalised term, because there is no other way. Would you like me to say, "no comment", because of insufficient data. In that way we can't even express a single opinion ever because there is no instance of there ever be absolutely fair and true comparison. It applies to the theory of relativity, or more relevantly, any benchmark. >What work??? Professional work? Database work? CAD work? Minix is a nice OS, >but don't lose your head please. Please view it in my and the people in this discussion list. Most of us write programs which are text based. Or write articles where the text entry is the most time consuming one. I spent 3 hours on this article. The setup time is only 15 minutes, which includes switching on the computer, logging in, and entering the commands. >(BTW, I don't know version 4.1.1 of BSD. I guess that you mean SunOS 4.1.1. >While it is not perfect (it's BIG), it certainly is not slow in everything!) It is slow in everything that matter to me the most: swithing on and off, responding to simple commands such as ls. I admit I don't use it a lot and try to avoid it as much as I can. BSD Unix is never designed to be fast. It is designed to help "PROGRAMMERS" write reasonably fast programs easily, without worrying too much about memory management, memory and disk space savings. You can't port it to Super Computers but someday all PC's would be Super Computers because super fast computers cannot afford to accommodate extra fat, but Unix System 7 can. Unfortunately ast failed to follow the strength of Unix System 7, the most serious is the relocating object format of Unix 7. If it had been supported, the 64K barrier can be broken by now. After all, the a.out format for Unix 7 is similar to MSDOS .EXE format. For PCB design I can use ORCAD PCB. For PAL programming I use ABEL. For PGA design I use XILINK. For networking connection I use Dec terminal emulator. For software simulation I may use Turbo C++. All running on MSDOS on IBM PC. With 486-33Mhz and 600Mbyte, I have no need for Sun Workstation, at least for the moment and I notice that a lot of CAD software can be used on the IBM PC even for the most sophisticated ones and they are cheaper. Sparc has a wonderful architecture. In fact overall it is faster than 80486 because Sparc can now operate at 50Mhz whereas 486 has not got there yet. The advantage of Sparc is that the design is simpler so the process technology need not be sophisticated but who can beat Intel Process Technology, and IBM Marketing muscle. MIPs is even simpler, thats why it is faster, but you need the optimising compiler. The Intel 80860 is even faster because it has built in floating point processor without hardware interlock, but it does not have an optimising compiler. I heard that i586 would combine the best features of 486(software portability) with 80860(hardware simplicity). The only way it can do it, is by disabling some hardware hazard interlock. I don't really like the x86 architecture because it is dirty, with lots of options, but refer to ast paper titled: Implications of Structured Programming for Machine Architecture, in Communications of the ACM, Volume 21, Number 3, March 1978. His conclusion was "an encoding scheme must be devised in which the most commonly occurring cases are assigned the shortest bit patterns, and the least commonly occuring cases are assigned the longest bit patterns." Othman Ahmad, School of EEE, Nanyang Technological University, Singapore 2263
adrie@philica.ica.philips.nl (Adrie Koolen) (06/17/91)
In article <56298@nigel.ee.udel.edu> EOAHMAD%NTUVAX.BITNET@cunyvm.cuny.edu writes: >NL%"adrie@PHILICA.ICA.PHILIPS.NL" Adrie Koolen wrote >>I think you're comparing apples with oranges. The SparcStations have a frame >>buffer with only one display mode, which is bit-mapped. They're not designed >>for fast text scrolling. I've recently compared the speed of the graphical > >Typing make is faster than going through all the menu routines. Most of the >people that I saw use the X-window systems use them as text based windows. Then you just have to believe me when I say that typing/editing in an 80*25 character XTerm window on a Sun SparcStation is instantenuous. In fact, the speed of the computer really doesn't matter because display actions in such a small window are much faster than my typing. The advantage I have on my Sun is that I can open a lot of those small windows and view/edit several files simultaneously. I know that on a PC, you can do that as well with the proper software and hardware, but my Sun is by no means worse to use than a PC. >>display systems of my color SparcStation and 33MHz 386 PC with super VGA in >>1024*768*256 mode under MS Windows. I looked at display operations like >>refreshing the screen and copying rectangular areas on the screen. The > >Why do you use 256 colour mode when you need only 16 most of the time. You said it yourself: sometimes, I need more colors. Your question seems an acknowledgement of the fact that 256 colors on a SVGA is considerably slower than 16 colors. >My MS windows with 640X480X16 on 386-25 is as fast as the Sparcstation. May I quote from your first message: >Surprisingly, I find that the Sun Sparc Station 1, even without windows, >at 25 Mhz is so slow at screen updates compared to Minix on my 386/25 without >cache with super VGA display. Seems to me that you're not very consistent in your remarks! BTW, without windows, the SparcStation uses a very big bit-mapped screen (1MB) with very big letters to emulate a 80*34 terminal. Character IO is done via Forth routines which are stored in the boot PROM. Of course this is slow. Almost nobody uses the `console' mode on a SparcStation. Use a windowing system like OpenWindows or XWindows; they have direct access to the frame- buffer and thus are much faster. >There is no point to use 1024X768X256 mode because my software does not >support it. The only thing I can conclude from that is that that's the fault of your software. But if you're ever able to get software that supports this mode (almost every SVGA card comes with MS windows 3 driver for all high res modes), please take a look and you'll find that screen IO is slow. >The AT-bus can go at 16Mbit/s. The 32 bit EISA can do 33Mbyte/s. It is not >the AT bus, it is the VRAM. You need dual ported VRAM. Why don't you try >VGA display cards with cache from Trident. First, 16Mbit/s is 2MB/s and that means max 1MB/s scrolling, unless you use the special tricks in VGA plane modes. That's not fast. Second, I've no PC at home, only one at work. My boss won't pay for SVGA cards just to test the speed of them. Finally, are the Trident cards really that much faster than ET4000 cards in 256 color graphics mode? Can you give examples of how fast you can move a screen rectangle to another part of the screen in 256 color mode? I'm curious! >>BTW. Framebuffer access on the SparcStation IPC takes some 7 cycles, so you >>can access 4 pixels (32 bits) more than 3 times per us. > >Does it have 4 bit mode? No, the color frame buffer of the SparcStation (cg3) doesn't know different modes. Only one: 1152*900*256. >For PCB design I can use ORCAD PCB. For PAL programming I use ABEL. For PGA >design I use XILINK. For networking connection I use Dec terminal emulator. > For software simulation I may use Turbo C++. All running on MSDOS on > IBM PC. With 486-33Mhz and 600Mbyte, I have no need for Sun Workstation, >at least for the moment and I notice that a lot of CAD software can be used >on the IBM PC even for the most sophisticated ones and they are cheaper. Let me quote from your first article one more time: >Minix is not only for learning >but also for serious users who wants work done quickly. You just indicated a reason why serious users don't use Minix: (almost) all serious PC programs run under DOS, not Minix. I've nothing against Minix, I just observe and take note. Adrie Koolen (adrie@ica.philips.nl) Philips Innovations Center Aachen