home *** CD-ROM | disk | FTP | other *** search
- > > To get some more water on the wheel, I tested memspeed yesterday on my Afterburner040.
- > ...
- > > With memspeed I get 4.9 MB/s to ST-RAM read and slightly less write.
- >
- > In what graphics mode?
-
- 640x480 mono.
-
- I guess it's correct as you got around 6MB/s and access to ST-RAM is said to be
- about 25% slower.
-
- > > To TT-RAM I get 35.9 MB/s read and 16 MB/s write.
- >
- > That sounds rather low, but see below.
-
- I guess it can depend on alot of things and the program was done before 040 Ataris
- were even thought of.
-
- > What did you get on cache accesses?
-
- L1 or L2?
- L1 was way over 100MB/s (~130MB/s I think) and I don't remember L2. I didn't care to
- memories them. :-(
-
- > Only 9 million 32 bit read accesses and 4 million write accesses per second.
- > Even a standard Falcon comes close to that for writes (even if they're only
- > 16 bit in that case). I guess the '040 might have to read the entire cache
- > line first when writing, though, unless you use move16.
-
- I have no idea what memspeed is up to. :-)
-
- > Shouldn't burst mode do better than that?
- > (Assuming the AB040 does burst, which seems likely.)
-
- Dunno.
-
- > According to Motorola's 680x0 optimization document:
- > Saving/restoring registers:
- >
- > A: movem.l d4-d7,-(a7) B: move.l d7,-(a7)
- > move.l d6,-(a7)
- > move.l d5,-(a7)
- > move.l d4,-(a7)
- >
- > 68000/20/60/xx: A
- > 68040: B (if time critical)
- >
- > Which suggests that using movem for memory speed test isn't a good idea
- > on that processor. I've no idea how large the difference is, though.
- >
- > Trying to figure out the instruction timing tables is not simple, but to me
- > it looks like succeding MOVEMs with four registers will take seven cycles
- > each, while four simple MOVEs will only take four cycles.
- > That is, it _might_ be possible to almost double the figures.
- >
- > Doug has probably already figured out the correct numbers. ;-)
-
- I leave this to you guys who understand this stuff. :-)
- I rest the case. :-)
-
- > > One interresting thing is that when I run GEM-Bench and test the AB040 against
- > > the Medusa040 the AB040 is actually more than twice as fast at RAM-Access and
- >
- > Strange.
-
- Yup. :-)
-
- > > also faster than MagiC Mac running on a 040 MAC. Not a big difference though.
- >
- > What kind of Mac was that?
-
- I have no idea as it doesn't say.
-
- > > BTW. My program is called memspeed.tos, not ttp. :-)
- >
- > Well, that might be true for mine as well.
- > What version? I have 1.0.
-
- I don't know. If you come by my place today you could test it your self.
-
- //Magnus Kollberg
-
-