home *** CD-ROM | disk | FTP | other *** search
-
-
- > Well , I can't say that I care that much about MiNT either. What about MagiC 4 then?
- > Have anyone tried it? Daniel said that MagiC works but MagiC Desk doesn't.
-
- I think the AB is fine so long as you don't touch the MMU - which is a problem since
- setting up the cache and FastRAM kind of depends on that...
-
- It just means you need to be careful when adapting the driver software to work
- with Magic & MINT. I'll worry about all that when I am happy with the drivers
- on a non-multitasking machine. It's quite hard to patch all the incompatible parts
- of TOS as it is...
-
- > A little boring to burn along at lightspeed and not be able to multitask! :-|
-
- Multitasking is fine if you can rely on it, but it seems MiNT & Magic still need
- work in this department. I heard things like LHarc don't like Magic very much,
- which is strange considering how simple it is system-wise.
-
- > That's why I wonder if MiNT works. Even MultiTOS shoule be usable on the AB.
-
- I tried it without memory protection, and it does seem to work quite well - but
- my patches will need more work before Mint is happy to run. I was only able to
- test it with the original AB software which I really don't trust at all.
-
- > It's slow compared with some other things, but I guess you are right. It's a little
- > more complex than just drawing some text..
-
- You are right about the bus being a problem, as it's mainly screen-writing and
- copying that's going on there. This is why the figure seems slow, but it will always
- look like that unless you get a new screen bus. Even with a 68060 fitted. :)
-
- > I guess a graphics card would do wonders. :-)
-
- Yep.
-
- > > Integer is slow too.
-
- > What? At 410% for a divide? That's amazing Magnus! The divide is the one thing
- > that's almost impossible to optimise as you can't do a parallel version. If the
- > divide is this fast, it's a good sign for everything else! :)
-
- > But... okey. I just thought it was slow because of the other tests... but it's only
- > about twice the speed of an TT, or?
-
- Divides are misleading - you can't easily parellelise them in hardware, so the best
- you can expect is 1 bit per cycle. Occasionally, you will find a chip optimised to
- perform 2 bits per cycle or more, but that takes a lot of silicon and would waste
- space needed for more frequent opcodes. Integer divides are relatively infrequent.
-
- The 68040 either takes 1.5 cycles per bit or 0.75 cycles per bit, depending on
- whether you are counting the source or the destination bits. If you are counting
- the source bits for a 64/32=32 bit divide, then the divide takes 44 cycles to
- finish - this is 44/64 cycles, or exactly 0.68 cycles per bit, which would indicate
- superscalar speeds, or pretty close anyway.
-
- If you are counting the destination bits, then the answer is 44/32, or 1.37 cycles
- per bit - which is pretty much what you would expect from RISC, or at least very
- close to this.
-
- What I'm trying to say is that you can't really improve much on the 68030's
- divide - beyond doubling it's speed. The 68040 doubles it's speed! GEMBench
- shouldn't be basing it's integer tests on a divide, as it's the least likely
- instruction to change between chips. The fact that it has doubled means the
- chip is damned impressive. :)
-
- Do you see what I mean? The details are not clear, but GEMBench is just not
- a very well balanced way of judging speeds of chips. If you know about the
- divide, you can see that the improvement is actually quite significant.
-
- > I think the divide is quite impressive myself...
-
- > I take your word for it.
-
- Thanks. :)
-
- > I'm still running mine at 32Mhz as I don't have my RC-33 yet... :(
-
- RC-33? Do you mean a 33MHz 040??
-
- Yep. I still only have an over-clocked RC-25. :(
-
- Doug.
-
-