home *** CD-ROM | disk | FTP | other *** search
- Xref: sparky comp.sys.atari.st.tech:6395 comp.sys.atari.st:18915 comp.sys.amiga.advocacy:32154
- Newsgroups: comp.sys.atari.st.tech,comp.sys.atari.st,comp.sys.amiga.advocacy
- Path: sparky!uunet!cs.utexas.edu!usc!elroy.jpl.nasa.gov!hanauma.jpl.nasa.gov!hyc
- From: hyc@hanauma.jpl.nasa.gov (Howard Chu)
- Subject: Re: Falcon BUS..
- Message-ID: <1992Dec24.234910.28952@elroy.jpl.nasa.gov>
- Followup-To: comp.sys.atari.st.tech
- Sender: news@elroy.jpl.nasa.gov (Usenet)
- Nntp-Posting-Host: hanauma.jpl.nasa.gov
- Organization: SAR Systems Development & Processing, JPL
- References: <1992Dec14.235558.17175@news.th-darmstadt.de>
- Date: Thu, 24 Dec 1992 23:49:10 GMT
- Lines: 47
-
- Well, I finally got around to compiling this with GNU C. Here are some
- numbers for the Falcon in a couple different graphics modes:
-
- Test 640x480x2 640x480x256 320x240x16
-
- 16 bit read 6206061 3413333 5610959
- 16 bit write 16384000 6400000 14124138
- 32 bit read 6301538 3442017 5688889
- 32 bit write 31507692 7876923 24094118
-
- Here's a comparison from my Mega with Turbo16:
-
- 640x200x4
-
- 16 bit read 7876923
- 16 bit write 2884587
- 32 bit read 7876923
- 32 bit write 2884587
-
- First, looking at the Mega-ST, the numbers clearly agree with the fact that
- the ST has only a 16 bit bus. The read numbers appear exaggerated to me,
- probably due to how the Turbo-16 cache works.
-
- The numbers for the Falcon are a little less clear, but I'd figure that the
- read speeds for 16 and 32 bit cases indicate a 16 bit bus, with the slightly
- higher speed in the 32 bit case due to having to execute one fewer instruction
- in the actual test loop. (Well, it's still the same instruction count, but does
- a NOP really consume as much time as a post-increment?) The write speeds again
- show the effect of caching, which makes this test a little suspect. As you
- can see in the 640x480x2 case, the video doesn't slow down the processor at
- all in the 16 bit write case, the write speed is the maximum achievable on
- a 16 MHz processor with a 2 cycle write. Or rather, the 68030 cache is able
- to make up for any delay that might be present. But for the 32 bit case,
- the impact of the video becomes noticeable, being about 4% slower than the
- theoretical maximum.
-
- This test program only used a 2K data set, so the cache effects are too
- significant in these results. I guess I should try it again with a 64K
- data set, to override the Turbo-16's 32K cache and the 68030's on-chip
- caches...
-
- In the meantime, I'd be curious to see the results of this program run on
- an Amiga, since we've been bandying around so much stuff about all this ...
- --
- -- Howard Chu @ Jet Propulsion Laboratory, Pasadena, CA
-
- There's a narrow border between genius and insanity, but I'm a dual citizen.
-