Amiga 3000UX, X, OpenLook, Motif, Color, A2410, Etc. (somewhat long)
Matthew Dillon
dillon at overload.Berkeley.CA.US
Sun Mar 24 03:31:10 AEST 1991
In article <1991Mar21.085254.5325 at kessner.denver.co.us> david at kessner.denver.co.us (David Kessner) writes:
>In article <EACHUS.91Mar20181104 at aries.mitre.org> eachus at aries.mitre.org (Robert I. Eachus) writes:
>> I thought your system didn't have an XT bus? The problems "only"
>>occur when DMA can choke the CPU by taking over the bus.
>>Unfortunately, most PC compatable '386 machines currently have XT
>>busses.
>
>I have the standard "PC AT" bus which is called the ISA bus, for Industry
>Standaed Arch. Then there is MCA or Microchanel-- where IBM clames that
>you cannot do 'multitasking' without it (and only thet have it!). There is
>also EISA, for Extended ISA-- wich is like MCA, but isnt owned by IBM and is
>compatable with normal ISA cards.
>
>I have the ISA bus, but have never experienced the jerkyness that you
>describe...
The fact is, doing DMA on an ISA bus is *SLOWER* than doing it with a
CPU. Same goes with the EISA bus. The reason is due to the way
arbitration works. That plus the fact that the overall backplane speed
is, well, snails pace, and you get a real mess. Even accelerated
busses don't fix the problems... they're just accelerated slow busses.
The only reason PC's are able to claim any performance at all is
because their main memory isn't anywhere near that bus! This is a
great idea, but also exasperates the speed difference between memory
and IO. While you may not necessarily see any jerkiness, overall
performance is easily cut in half during disk io, or worse. PC based
UNIX platforms never work well for performance seekers and the faster
procssors (33MHz 486) only make the differences worse --- you are going
on at a fine clip but the moment you go to disk, POOF, the machine
stops. The moment you start to page, the thing becomes jerky.
I haven't looked at the MCA spec, but from what I've heard IBM is more
mouth than brain in terms of its capabilities.
>> But not the same code generators! The cheats for Dhrystone (as opposed to
>>optimizations) were all in the code generators.
>
>These results are consistent with EVERY dhrystone test I have seen, including
>those in UNIX Review, where they list 1.1 results as well as 2.x.
>
>The same Dhrystone program when compiled with Lattice C under AmigaDOS (with
>the 030 optimization turned on) got 7600 dhry/sec. I would not expect
>any Dhrystone program (1.1 or 2.x) to get more than 8500 using any compiler.
>
>I would not expect the Dhrystone results to get better with Amiga UNIX v2,
>but OS overhead will improve signifigantly.
>
>I've hinted enough: DOES ANYONE HAVE DHRYSTONE 2.X OR SPECMARK CODE????
>(and now back to the normally schedualed program...)
A 25MHz 68030 without a cache, as in the A3000, will get 6000-8000
Drys. With a 16K cache you can get 9000-12000. Most high-end 386
boxes have at least a 16K cache and this is why they generally get
8000-9000 drystones, it has NOTHING to do with the processor.
>For me, the raw CPU power is important since I do a lot of CPU bound
>tasks. Disk I/O is also important, but to a lesser extent. I also have
>12 meg of RAM, so swapping is not a problem. All of the terminals (one
>local, another via 9600 V.42bis modem) are connected via 38400 baud
>connections and are not a bottle-neck themselves. All in all, I am only
>I/O bound when I get a large Netnews batch-- but that has to do with the
>28ms drives I have...
>
>FYI, I have found a BIG case for text only displays. On the 386, I can
>'cat' a 80K text file in about 6 seconds. It even supports ten virtual
>terminals. It makes X windows look snail'ish...
Well, a 386 ISA/EISA isn't a bad choice if all you are looking for is
raw CPU power, assuming the thing doesn't have to go to disk you are
O.K. I.E. as long as you don't need to do a lot of disk IO and have
enough RAM so you don't have to page, you are ok.
> - David K
>--
>David Kessner - david at kessner.denver.co.us | do {
>1135 Fairfax, Denver CO 80220 (303) 377-1801 (p.m.) | . . .
>If you cant flame MS-DOS, who can you flame? | } while( jones);
--
Matthew Dillon dillon at Overload.Berkeley.CA.US
891 Regal Rd. uunet.uu.net!overload!dillon
Berkeley, Ca. 94708
USA
More information about the Comp.unix.amiga
mailing list