dbert@pogo.ai.mit.edu (Douglas Siebert) (02/28/91)
With all the talk about the 68040 accelerator boards and such, I was wondering if someone could maybe post a quick summary of the differences between the 68030 and 68040. I know that the 40 has a built-in Floating-Point co-processor (though without all the functionality of the 68882, from what I hear) and a built in cache...what other features does it have? And if you had a Mac with a 50MHz 040, how fast would your memory have to be to avoid wait states? Just curious... -- ________________________________________________________________________ Doug Siebert dbert@albert.ai.mit.edu MBA Student (2nd year) The University of Iowa
rfischer@Neon.Stanford.EDU (Ray Fischer) (03/01/91)
dbert@pogo.ai.mit.edu (Douglas Siebert) writes ... >With all the talk about the 68040 accelerator boards and such, I was wondering >if someone could maybe post a quick summary of the differences between the >68030 and 68040. I know that the 40 has a built-in Floating-Point >co-processor (though without all the functionality of the 68882, from what >I hear) and a built in cache...what other features does it have? And if Separate integer & floating point units allowing concurrent execution of floating point and integer instructions. Independent instruction and data memory managment units, each with a 64 entry 4-way set associative address translation cache. 4k byte instruction cache and 4k byte data cache, each 4-way set associative with 16 byte line size. Multiple independent execution pipelines, full Harvard internal architecture, and multiple internal buses. Dynamic bus sizing is NOT supported. All of this means roughly that the '040 is a fast little puppy. A 25MHz 68040 gets about as much work done as a 150MHz 68030. A floating point multiply that took about 70 clocks on a 68882 takes about 7 on the '040. So although most transedentals aren't directly supported in microcode, it should be possible to get them done in less time than on a 68882 anyway. Overall the '040 executes one instruction for every 1.4 clocks. Otherwise most people won't see much difference. >you had a Mac with a 50MHz 040, how fast would your memory have to be to >avoid wait states? Just curious... The highest transfer rate for the '040 is one long-word per clock. At 50MHz this is one read per 20ns. However, part of the advantage of having pipelining and multiple caches is that you don't need to have fast and expensive memory. The idea being that most memory references will be supplied from the cache. Ray Fischer rfischer@cs.stanford.edu
piper@s5000.rsvl.unisys.com (Piper Keairnes) (03/02/91)
rfischer@Neon.Stanford.EDU (Ray Fischer) writes: >All of this means roughly that the '040 is a fast little puppy. A 25MHz >68040 gets about as much work done as a 150MHz 68030. Where's the sign-up sheet? :-) -- Piper Keairnes * piper@rsvl.unisys.com * Purdue Univ. Unisys Corporation * uunet!rsvl.unisys.com!piper * Computing Center Open Software Products * 1410 Carling Dr, St Paul, MN 55108 * Consultant