Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!zaphod.mps.ohio-state.edu!swrinde!cs.utexas.edu!natinst!davoli
From: davoli@natinst.com (Russell Davoli)
Newsgroups: comp.sys.mac.hardware
Subject: Re: Graphics accelerator theory of operation question
Message-ID: <22883@natinst.natinst.com>
Date: 6 May 91 23:29:28 GMT
References: <1991May2.085707.11955@cascade.Stanford.EDU>
Organization: National Instruments, Austin, TX
Lines: 39

In article <1991May2.085707.11955@cascade.Stanford.EDU>, simoni@strat.Stanford.EDU (Richard Simoni) writes:
> 
> How do graphics accelerators for the Mac (say for instance the Apple card
> with an AMD 29000) gain their graphics speedup over a stock Macintosh?  That
> is, do they use:
> 
> 1.  Parallelism:  Graphics operations are off-loaded to separate processor,
> allowing main processor to continue with other tasks.  (In the Mac world,
> would this mess up the serialization of operations intended by the
> programmer?)  For entirely graphics-bound operations, this parallelism
> obviously won't buy much.

The 8-24 GC does this.  It apparently queues up a number of operations then
performs them when it decides it has queued enough.  When an operation is
queued, the program gets control back to continue processing while the
operation is performed.
> 
> 2.  Faster Graphics Processor:  Graphics operations are off-loaded to
> separate processor which is faster than main processor.  Main processor
> waits for graphics CPU to complete before proceeding.
>
The graphics processor on the 8-24GC is a RISC chip (an AM29000 I think).
It's probably a lot faster at most operations than a 68000 series processor,
but 1. is the way to go.

> 3.  Other factors: e.g., block transfers on bus to increase available
> bandwidth to memory, etc.
> 
The 8-24GC is supposed to be able to perform block transfers, but I don't
know whether they are really used.  The target memory must be capable of
block mode access for this to work, so I'm guessing that most of the video
cards out there don't have this hardware.  Block transfers are a big plus
since they take significantly fewer cycles than arbitrating on every transfer
then performing the transfer.

Hope this helped.

Russell Davoli