[comp.sys.atari.st] The TT is actually nice

t68@nikhefh.nikhef.nl (Jos Vermaseren) (10/01/90)

With all the ongoing bickering about Atari policy I think
it is about time that someone should mention that the TT is
actually quite a nice machine.
At the moment I have at home a TT with:
32 Mhz 68030
32 Mhz 68882
4 Mbytes ST RAM
4 Mbytes TT RAM
48 Mbytes 28msec hard disk
VGA monitor (640 by 480 pixels with 4 color planes).
So of course I had to test its performance and the like.
It turns out that the performance seems to be disappointing
till you tune it the same way many people got their ST tuned
with cachexxx.prg and/or similar programs.
The speed is very much a function of what memory you use, as
can be seen in the table:

On Mega 4:                   249 sec
ST/RAM  68030 Cache off      146 sec
TT/RAM  68030 Cache off      86 sec
ST/RAM  68030 Cache on       80 sec
TT/RAM  68030 Cache on       55.5 sec
TT/RAM  68030 Cache on 68020 code 51.4 sec

The above was a FORM test program. Typical runs give about a
factor 4. I measured the same kind of speed increase for the
turbo C compiler. After turning on the appropriate bits in
the program header to force the compiler to run in the TT RAM
the speed increase was more than a factor 4.
MetaFont runs faster by a factor 8. Apparently it uses many
shifts and bitfields.

Of course many programs don't run on it, but it is possible to
switch to ST mode (choice of all three resolutions). The main
problem for me is that the Turbo-C compiler doesn't support
the 68020 modes convincingly. Especially when long multiplications
divisions and mod operations are involved it may loose track
of what it is doing. It is also necessary to use a new
floating point library. The 68882 operations are produced inline,
but the trancendental functions are done by a call to a function.
The speed is nice. I measured a Savage test time of 0.1597 sec.
Most libraries of the compilers take on the 68000 (if they are
very fast) in double precision more than 30 sec. The chip
can do 600000 80 bit multiplications per second. This means that
finally it is possible to do some serious calculations at home.

Now something about the price. There is a price range for which this
computer seems to be a good buy. This would be the range around
8000 dutch guilders (or less of course). In Holland a SUN with
proper disks etc would cost close to 15000 and an Apollo DN2500
with disk would be more than 10000 after discount. Actually the
DN2500 is slower. It runs on 20 Mhz. The DN3500 from which I type
this in runs on 25 Mhz and runs the same program as tested above in
56 sec with code that was produced for the 68020/68030 and its disks
are faster. In addition the 56 sec are CPU time.
If on the other the machine starts costing more than 10000 guilders
it may be too expensive to be attractive. The official price is
actually slightly above 10000 guilders, so there may be a problem
there. For reference: A fully equipped Mac II cx etc costs here
after discount also about 15000 guilders.

I have also seen an ethernet card for the Mega and there will be
one shortly for the TT too. Atari is mainly waiting on this TT
version of the card before they come out with it. It seems the
better policy. The price would be very reasonable.

Talking about policy:
First Atari talks much about new products that never make it to
the market or if they make it have generated such an expectations
that the actual product is disappointing or too late or whatever.
Then we hardly hear anything from them for nearly two years and
everyone thinks the ST is their last good product. Then the TT
comes out, showing that they can still do something.
I believe this last mode of operation wasn't entirely ideal
either, because they must have lost the faith of many people.
Somehow it seems to me that the proper policy should be inbetween.
This could be expressed by a better (software) policy.
Many developers have felt in the past that Atari left them out
in the cold. Of course Atari isn't a social institute, but somehow
I have a feeling that this could be beneficial to both sides.
One such act could be to see what is needed to get Simon Poole to
make uniterm run on the TT. This program alone made Atari sell many
ST computers in Germany and in Holland.


Jos Vermaseren


Disclaimer:
I my employer would have a faster computer he could reject my
opinions much faster.