home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.sys.amiga.hardware
- Path: sparky!uunet!zaphod.mps.ohio-state.edu!rpi!sarah!albnyvms.bitnet!TW5232
- From: tw5232@albnyvms.bitnet (Teddy Wang)
- Subject: Re: MIPS results: Which one is true ???
- Message-ID: <1992Jul22.153637.5219@sarah.albany.edu>
- Sender: news@sarah.albany.edu (News Administrator)
- Reply-To: tw5232@albnyvms.bitnet
- Organization: University of NY at Albany
- References: <92196.134124PROHVK38@TREARN.BITNET>,<33228@cbmvax.commodore.com>
- Date: Wed, 22 Jul 92 15:36:37 GMT
- Lines: 53
-
- In article <33228@cbmvax.commodore.com>, mks@cbmvax.commodore.com (Michael Sinz) writes:
- [Useless Stuff Deleted]
- >This is really very uninteresting. For example, the on the 68040,
- >the NOP instruction is actually sometimes rather slow. (Very slow)
- >yet the time to execute it is basically 0, but the fact that it forces
- >pipeline sync and pending writes to finish means that it can take many
- >cycles. On a 68000, a DIVU or DIVS takes up to 168 cycles. At 7MHz
- >this comes out to 41,700 DIVS per second (0.041 MIPS) but then a
- >MOVE register to register takes 4 cycles (1.75 MIPS) but then doing
- >just one or the other is not what sofeware does. Plus, given that
- >sometimes you get wait states to memory but you may have a cache on
- >the CPU but it may be off for certain memory areas, etc, all make MIPS
- >figures for processors useless. Plus, when you look at the difference
- >between the instuctions and what they did, if it takes 5 instructions
- >to do something on one processor but it take 10 instructions on the other,
- >who cares if the second processor has 50% more "MIPS" since you need
- >twice as many MIPS just to make up for the fact that it take more
- >instructions to get the job done.
- >
- >Lets just say that the 68040 is *very* fast. The 68030 is fast (as is the
- >68020 at the same clock rate unless you add the external MMU) and the 68000
- >is rather slow (when put beside these others).
-
- More importantly, I think, is the performance of these chips in the REAL WORLD.
- How many people run speed-test and comparison programs day in - day out*. Put
- it plainly, doing a DPaint IV move animation, an 030 equipped 2500 was 6 times
- quicker than a standard 2000. An 040 equipped 2000 was 17 times faster than
- the standard 2000.
-
- Doing screen magnification (from 400% to 800%) on a heavily bitmap-imported
- PageStream screen took 2 minutes 35 seconds on a standard 68000 based A2000.
- On the 030, the same process took 25 seconds. The 040 took slightly less than
- 1 second to do the same.
-
- A Turbo Silver ray-tracing which took 14 hours on the standard A2000, took 30
- minutes on the 030 equipped 2500 using the floating point version of the
- program. Since the floating point version of Turbo Silver crashed on the 040,
- I used the standard non-floating which took 45 minutes.
-
- You must also realize that many programs are not compiled to optimize the
- modern chips (come on Lattice and Manx, get your act together and release a
- compiler which is 040 specific.)
-
- * Actually, the first couple of days I had my Fusion Forty, I ran every
- speed-testing program available. I was real curious at how much performance,
- $995 got me.
-
- ////////
- // twang@thor.albany.edu
- // TWANG@ALBNYVMS.BITNET
- // eddy Wang TW5232@ALBNYVMS.BITNET
- // Computer Art and Animation at SUNY Albany
- New Illusions Albany, NY "Increase the Peace..."
-