home *** CD-ROM | disk | FTP | other *** search
- Xref: sparky comp.sys.ibm.pc.hardware:21729 comp.sys.intel:1485 comp.sys.ibm.pc.misc:11435
- Newsgroups: comp.sys.ibm.pc.hardware,comp.sys.intel,comp.sys.ibm.pc.misc
- Path: sparky!uunet!math.fu-berlin.de!informatik.tu-muenchen.de!roell
- From: roell@informatik.tu-muenchen.de (Thomas Roell)
- Subject: Re: More on Local Bus - What about Unix?? S3 vs.
- In-Reply-To: adams@pdv2.fmr.maschinenbau.th-darmstadt.de's message of Thu, 13 Aug 1992 00:48:12 GMT
- References: <0105010F.ac48r9@mprnews.mpr.com>
- <1992Aug10.092255.118@vulcan.resmel.bhp.com.au>
- <Bss82v.LJM@usenet.ucs.indiana.edu>
- <ADAMS.92Aug13004812@PDV2.pdv2.fmr.maschinenbau.th-darmstadt.de>
- Sender: news@Informatik.TU-Muenchen.DE (USENET Newssystem)
- Organization: Inst. fuer Informatik, Technische Univ. Muenchen, Germany
- Date: Thu, 13 Aug 1992 06:41:56 GMT
- Message-ID: <1992Aug13.064156.2539@Informatik.TU-Muenchen.DE>
- Lines: 48
-
- >1) Without a coprocessor, graphic in a PC means bit manipulation by CPU.
- >This may eat up most of the available bandwidth.
-
- Yes, but in localbus designs you have the bandwidth available. And if
- you have a simple framebuffer directly attached to the CPU it is
- *very* fast.
-
- >2) As a coprocessor allows to use protocols at a much higher level,
- >like X or Phigs, communication overhead _shrinks_ tremendously.
- >This has been proven with PC-graphic adapters running Xserver themselves,
- >no CPU had to write to video memory anymore.
-
- Well, then you hit another effect. The CPUs are getting faster faster
- than the graphics chips do. Like the 34020 is about 6 MIPS, while a
- 486 is much higher (think around 20 on a 486/50). That means that the
- main CPU is just waiting for the Graphics board the do something ...
- Another performance wall you hit is the communication between adaptor
- and CPU. Normally they want to have all graphics requests in a special
- format encoded, which means the 486 is quite busy to translate the
- users requests. And not to forget you have to go over a bus, which
- means you have to do busarbitrating (like queue polling).
-
- My personal oppinion is that a graphics chips should have the best of
- both worlds:
-
- a) Memory mapped, both framebuffer and graphics engine
-
- b) The GE should have a deep queue for commands
-
- c) The GE should be optimized to do only simple commands, but
- these hardcoded in silicon to drive the RAMs at threir top speed.
-
- The commands would be simply COPY_RECT, FILL_RECT,
- DRAW_LINE.
-
- d) Let the rest do the main CPU, which has the horsepower for
- clipping and rendering in lowend systems.
-
- An approch like this will buy you much more than a highpriced special
- Graphics board. Or why do you think are these RISCy GEs faster under X
- than the 340x0 based boards ?
-
- - Thomas
- --
- -------------------------------------------------------------------------------
- Don't drink and drive ! e-mail: roell@sgcs.com
- First drink, then drive ... #include <sys/pizza.h>
-
-