#68040 & VT2.0
15 messages in this thread
I just got a 68040 for my A2000 at 33MHz and 16 Megs of ram.
I expected the Toaster to go faster than my old 50MHz 68030, but it
didn't.
I rendered a frame with texture mapping and it went from 11 minutes on
my 030 to 16 minutes on the 040. SysInfo reports that the 040 is working at
24 mips and that the 030 is at 6 mips. All the 040 software has been
properly installed but I really hope that there is something I did wrong,
because I cant believe the results.
I have read messages from other people here saing that the 040 is much
faster, which is true, but in the VT2.0 is not the case for me.
Anny suggestions?
PS: my 040 is a Progreesive.
PLEASE HELP!
Javier
– Via Whap! from Costa Rica
We have experienced the same slower than supposed to factor. Our 040 board
speeds up rendering only about 35% (over 030 using lightwave).
Javier –
I'll be corrected in this if I'm wrong, but assuming that your 50MHz
'030 had a 50 MHz '882, the slowness of the '040 could be the fact that the
built-in floating point unit in the '040 is not as capable as the '882.
Therefore, some floating point operations might actually take longer using
the '040 fp unit than the equivalent operations on the '882, as I believe
some of the floating point features of the '882 simply aren't there in the
'040 and need to be done via software.
If this _is_ the case, I'm not sure if anything can be done about it
unless Motorola comes out with a seperate fp chip for the 040. Hope this
helps …
– via Whap!
Javier,
Did you rename the Lightwave progams like it suggests in the '040 manual?
Apparently, the '040's built-in math coprocessor is not like the '882. When
the Switcher starts Lightwave, it doesn't realize there's a math chip
present and loads the no-coprocessor version. You need to fake it out by
renaming Lightwave to Lightwave.slow. (Just to keep it around.) Then rename
Lightwave.FP to Lightwave. When you do that, the floating point version
gets loaded by default.
Jim
In addition – make sure you have he latest 68040 software from PP&S.
-Lou
First he needs to determine if it is loading the wrong one since I know of
several people who have not had to rename them.
He needs to look at the `About' menu and see if you are running the FP
version or the non-FP. If it is the Non-FP then he will need to rename the
files.
– comming to you from `here', where else?
Hmmm… that's interesting. When I set up an '040 Toaster system for a
customer a while back, I did have to rename the files. That was early in
the life of the progressive board, so maybe things have changed.
Ok guys, here's the scoop on the '040 problem. I am very surprised that no
one has really responded to the real problem. Here goes. Forgive my petty
arrogance.
Lightwave 2.0 was code written specifically for the '882 math chip which is
(by my understanding) very good at trigonometric functions and has calls
for them. The '040 chip has a different set of calls and does not support
the advanced trig functions (these are needed for certain textures and
ray-tracing). However, if Lightwave were written for the '040 it would run
rings around anything else. I was told that an '040 version running at
28mhz could be FOUR times faster than the '030 version 2.0 running at
33mhz. It would SMOKE!!! The reason for this is that every time Lightwave
calls for Trig stuff that the '040 does not do via hardware, it calls to a
library and this functions gets done in RAM with the 040's simpler (but
much faster) functions. It's here where the slow down occurs.
The solution? Just wait. It's coming. Trust me.
Jim.lwob
How can software functions be faster than hardware functions. A trig
function written in software will *NEVER* beat a hardware analog.
–Pete
Please read what I wrote a little closer. The '040 native software does
NOT do in software the trig finctions etc. The trig functions are carried
out by the '040 as simpler math, ie., because the '040 does not support
sine, cosine as direct functions, they are called as simpler functions and
thus the need to handle these by software is avoided. Currently, some '030
programs (Lightwave) call direct trig that the '040 CANNOT DO DIRECTLY. The
'040 then looks to Libraries for these and this slows things down.
The '040 programs avoid this by making simpler calls to the chip. They do
NOT perform these functions with software patches, that's what needs to be
avoided. Instead they utilize the very quick (3.5 to 4 mflops as opposed to
about 500 Kflops (note the K)) math in the '040 to get the final results.
I got this information from folks at Progressive Peripherals and it makes
good sense. If you like think of it as writing machine code to carry out a
task as opposed to using BASIC which has to interpret the behunkus out of
it. Machine language cannot with one command return strings which BASIC can
do with little instruction, but when the numerous little intricate steps
are taken to achieve the same result there are actually many many times
fewer clock cycles eaten up in the process.
Jim Roof
Well, the trig functions in a math coprocessor are still written in
software… They are implemented in some form of microcode running in the
ROM on the Math Co itself. You can implement the functions completely in
hardware, but I don't think you can do it without going back to analog
computers.
Brian
– via Whap!
Javier,
from what I read, Lightwave's fractal noise functions actually run
_slower_ on a 68040. Were you using lots of them in your picture?
Blaq!
There will be a new version of LightWave that will be more compatible with
the 040
JWL
You forgot to rename the LIGHTWAVE.FPU file to LIGHTWAVE before starting
the toaster… your dealer should have told you that!!
– Pat, Whapping from Quebec City!
It seems strange to us that there is only a 3 minute difference between
render times. If the failure is that the 040's coprocessor is not being
used then the time lag should be >> than 3 minutes. (I think, unless the
040 is insanely faster than the 030 (benchmarks aside)). Take the case of
the lowly XT with a math coprocessor and a brand new 386SX. If both are
given the same floating point operation, the XT is going to beat the pants
off of the 386SuX by a lot. And since XTs are somewhere around 1/10 the
instruction speed of a 386SuX, you might be suprized, but it is the case
none the less.
–Pete