Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!sdd.hp.com!caen!uwm.edu!rutgers!mcdchg!heiby From: heiby@mcdchg.chg.mcd.mot.com (Ron Heiby) Newsgroups: comp.sys.m68k Subject: Re: 68040 status Message-ID: <55919@mcdchg.chg.mcd.mot.com> Date: 7 Feb 91 22:12:03 GMT References: <38696@cup.portal.com> Organization: Motorola Computer Group, Schaumburg, IL Lines: 41 Note that I am not part of Motorola's Semiconductor Sector. I'm basically a customer, just like everybody else. I have a copy of the errata sheet for the 18D34B. If you can get a part, you can get the sheet (I believe). (Perhaps members of the press could contact the appropriate office at Semiconductor to get one?) Almost all of the items seemed to me to be related to the FPU or the MMU. Most seem to have workarounds. Many are marked as being fixed in the next mask. I have a customer who has ported a real-time kernel to our first 68040 based board. He reports that running his software compiled with a 68020 compiler, taking *NO* advantage of optimizations in the newer processor and *NOT* taking into consideration alignment issues for performance optimization, with Icache on and Dcache in copy-back mode, he is seeing 3 to 3.5 times the performance of the 68020 board running at the same clock speed, same quantity of (on-board and total) DRAM in some of his benchmark tests. That brings us to what I believe are the reasons we've seen very little in the way of benchmarks. The compiler vendors have not yet produced compilers that understand and take advantage of the optimizations that are possible with the 68040. The 3 to 3.5 X figures can be taken as *minimums*, with much better performance expected when the compilers are there. (I've heard that a major C compiler vendor is in Beta with its compiler right now.) The problem is that once "official" numbers are published, people tend to forget all of the caveats about how they came from an ancient compiler that had never heard of MOVE16 instructions (for example). As always, to find out what kind of performance boost your application will see, the best thing to do is to try it. Second best is to *deeply* understand exactly how your benchmarks have been constructed, compiled, loaded, and executed. Beyond that, all you've got (as my figures above) are general ballpark figures with mucho caveats. I have heard no rumor about cancelling the 040. It sounds completely ridiculous to me. I *have* seen detailed plans for our second 68040 board offering. I am *VERY* impressed. -- Ron Heiby, heiby@chg.mcd.mot.com Moderator: comp.newprod "Wrong is wrong, even when it helps you." Popeye