uace0@uhnix2.uh.edu (Michael B. Vederman) (07/25/89)
Well, here is a follow-up to my previous (obviously harsh) posting.
*** FLAME SORTA ON ***
Is the CMI accelerator a 16 Mhz upgrade? Yes, it is.
Do you actually get improved system performance? Yes and no...
What's the real scoop? I'll answer that after this installation update...
After completely trashing the 68000 on MY ST (after an unsuccessful install-
ation) and subsequently replacing the 68000, only to discover my video had
gone out somehow, and after fiddling with it some more, only to have the machine
completely dead now...
I have successfully installed the CMI upgrade to the company's ST (all beit
it oply has 512K - barely usable :-), I can say that I have been a victim of
poor documentation.
Hear is an excerpt of what is clearly an unacceptable explanation of installing
the unit *in place of* the 68000 using a socket:
"There are two ways to install the Processor Accelerator into the Atari ST. One
is to piggyback the Accelerator to the original 68000 processor soldered to the
motherboard of your Atari ST, the other is to desolder the original processor
and replace it with a socket which you would solder to the motherboard.
[their emphasis]
THE FIRST AND PREFERABLE WAY TO INSTALL THE PROCESSOR ACCELERATOR IS TO
PIGGYBACK IT ON THE ORIGINAL 68000 PROCESSOR ON THE MOTHERBOARD. Piggybacking
is preferable because ...
A paragraph on why you should do it this way, plus a plug for a
"high reliability" socket which can be obtained from CMI for $5.00. "
That's it folks... Now there is a page dedicated to piggyback installation,
with no reference what-so-ever to the socket method.
As it turns out, even if you *do* remove the 68000 and install a socket, you
must *still* clip 3 pins on the socket and attach the "fly-wires." Am I being
a stickler about words? Ask my dead upgraded 520 (a REV A board at that!) that
question!
[ATARI - does the $95 exchange work on any trade-in?]
So now I have the board installed in Double Cick Software's 520 ST, in fact I am
staring at it right now. I fortunately had enoug fore-sight to wire-wrap all
the components, as well as making a standoff socket so I could plug in the
"fly wires" instead of munging the motherboard. I even wire-wrapped my eproms,
although I did manage to break two pins off on two chips...
So now for the results:
These results are using the benchmark program Quik Index 1.5 by Darek Mihoka
and Ignac Kolenko. An excellent program I use all the time during development
so I can tweek performance. The results...
16 mhz+tst 1.6 16 mhz alone 8 mhz+tst 1.6 FR 8 mhz w/tst
Cpu memory read 99 99 99 99 | 99
CPU register 99 100 99 99 | 99
CPU division 181 181 99 99 | 99
CPU shift 177 178 99 99 | 99
DMA 3201 3219 3219 3166! | 3166!
GEMDOS I/O 600 600 600 600 | 600
RPM 3600 3600 3600 3600 | 3600
BIOS bconout txt 337 106 335 99 | 335
BIOS cconws 1440 104 1410 99 | 1410
Screen Scroll 134 100 133 99 | 133
GEM Draw 200 104 194 99 | 194
Legend:
16 mhz = CMI accl in 16 mhz mode
tst/tst 1.6 = Turbo ST 1.6 (I borrowed a friends copy - but I am convinced now)
8 mhz = CMI accl in 8 mhz mode
FR = FAST ROM option
I should note that my system would not operate in the 16 mhz FAST ROM mode!!
Also, the results obtained in the 'FAST ROM' mode are normal in non-FAST ROM
machines! I am using TOS 1.4 with 512K cmos eproms (27C512).
As you can see, the FAST ROM seems to be a hoax, seeing as I could not test it
at 16 mhz, I can not say that with complete certainty. However, the results
obtained indicate the same performance as a plain 8 mhz machine.
I do not have a blitter at this time, but I am certainly convinced that
Turbo ST 1.6 is !phenomenal! (at least this experience is not a complete
waste :-)
For a matter of reference, I am posting the 'results' posted by CMI in their
media hype...
CMI Proc Accel Atari-8 V JATO-16 FAST-T16
Price($U.S.) $299 $339 $99 $399/299?
(Blitter)
CPU memory (3) 101 101 100 100 130
GEM draw (3) 220 279 158 151 187
BIOS text (3) 201 210 143 162 177
GEMDOS I/O (2) 112 112 112 112 75
DMA read (2) 161 161 161 161 161
CPU divide (1) 183 183 100 183 202
CPU shifts (1) 180 180 100 179 205
Mandelbrot (FPU,1) 153 976 100 111 **
--- --- --- --- ---
Weighted Average: 164 228 128 141 157
Forget the weighted average... when using Turbo ST, the only advantage
offered is the CPU register operations. To test that out, I assembled a
3200 line 68000 assembly program in both the 8 and 16 mhz mode (fast rom
has no effect in assembling) using GenST. The file took 6.0 seconds in
8 mhz mode and 5.9 seconds in 16 mhz mode.
So what is the bottom line? Turbo ST more than offsets any advantages gained
for GEM, but has no effect on CPU effectiveness. If my understanding is
correct, the FAST-16 board which offers a RAM cache will be the real winner
because it has a 30% improvement in ram access, which means that programs
like Turbo ST will in all probability run faster, and seeing that the cpu
operations as posted above are better than CMI, the FAST-16 should be the
real winner (when it makes it to market...)
The only problem will be of course also the advantage, the RAM cache, which
may cause software incompatibility if the program uses self-modifying code.
This is my understanding, I could be unclear on it, though.
So, In my (humble) opinion, the CMI board should be called a blitter and
FPU upgrade (albeit expensive). Get a blitter though...
I should note that the CMI board does have an expansion connector which is
not even alluded to in the docs. One could only guess what it could be used
for.
(I should post note that I was using an Atari SH204 hard drive in the test)
So what now? Anyone want an accelerator board? It is ashame...
*** FLAME SORTA OFF ***
The opinions expressed are solely my own and *DO NOT* reflect those of
Double Click Softare, it's management, or any living person.
- michael vederman
--
for (;;) : Use ATARINET, send an interactive
do_it(c_programmers); : message such as:
: Tell UH-INFO at UHUPVM1 ATARINET HELP
University Atari Computer Enthusiasts : University of Houston UACE