home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!elroy.jpl.nasa.gov!swrinde!zaphod.mps.ohio-state.edu!uwm.edu!rpi!masscomp!calvin!mark
- From: mark@calvin..westford.ccur.com (Mark Thompson)
- Newsgroups: comp.sys.amiga.graphics
- Subject: Re: Vivid 24 graphics card
- Message-ID: <1992Aug18.190314.25964@westford.ccur.com>
- Date: 18 Aug 92 19:03:14 GMT
- References: <64064@cup.portal.com> <andrey.713925244@harry> <1992Aug17.164849.5029@eagle.lerc.nasa.gov>
- Sender: usenet@westford.ccur.com (UNIX news)
- Organization: Concurrent Computer Corp. Westford MA.
- Lines: 48
-
- In article <1992Aug17.164849.5029@eagle.lerc.nasa.gov> fsset@bach.lerc.nasa.gov (Scott Townsend) writes:
- >In article <1992Aug15.003630.23701@mnemosyne.cs.du.edu> dkennedy@nyx.cs.du.edu (Don Kennedy) writes:
- >>Note: the full blown system is rated at 28 MIPS, 160 Mflops. As a
- >>Comparison, the Silicon Graphics 4K/480 VGX is rated at 280 MIPS, 70
- >>Mflops. The SGI system costs $194,000. An Amiga 3000, 040 processor, and
- >>Vivid 24 (full blown) costs $12,325.
-
- >This looks like CPU rates for the SGI.
-
- As Scott pointed out, these are the CPU rates and have nothing to do with the
- graphics engine. The graphics engine is a finely tuned architecture designed
- specifically for fast polygon rendering. This is in strong contrast to the DMI
- design which is a general purpose graphics engine. While general purpose is
- becoming more and more popular in the graphics hardware world (it allows
- a great deal more algorithm flexibility), it does not compare to the speed and
- efficiency of specialized designs like SGI's. You need ASICs to achieve that
- kind of efficiency. 160 MFLOPS of peak performance has little bearing on
- how much rendering capability you will get out of a system.
-
- >I'll admit, I don't know where this 40 MFLOPS number came from
- >(peak pipeline speed?, LINPACK? SLALOM? single/double precision?)
-
- The 40MFLOPS comes from an uachieveable peak theoretical pipelined throughput.
- The actual useable amount is substantially less and will vary depending on
- the application and the code efficiency.
-
- > but for graphics pipeline operations, which tend toward 'embarassing'
- > parallelism, you can get close to linear speedup.
- ^^^
- Yes, you CAN get close to linear speedup for graphics pipeline operations
- but only in a system designed for it. But taking the whole rendering task
- into perspective there are many things that will prevent such linear increases
- and this is especially true of a general purpose design.
-
- >I don't think quoting 160 MFLOPS for the processors on the board is
- >out of line. A bit agressive perhaps ;-)
-
- Marketing people will always tout the largest number they can lay their
- slimy hands on. I know for our quad i860 based 3D graphics card it says
- 320 MFLOPS peak in the marketing blurb. This peak rate is clearly not
- attainable by a long shot though and the same holds true with the DMI card.
- %~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~%
- % ` ' Mark Thompson CONCURRENT COMPUTER %
- % --==* RADIANT *==-- mark@westford.ccur.com Principal Graphics %
- % ' Image ` ...!uunet!masscomp!mark Hardware Architect %
- % Productions (508)392-2480 (603)424-1829 & General Nuisance %
- % %
- ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-