grenley@nsc.nsc.com (George Grenley) (07/02/87)
About 2 months ago I started a discussion about benchmarking the current crop of microprocessors, specifically the NSC32532, the Intel 386, the Mot 68020 and 68030, Fairchild's Clipper, MIP's whatever-thet-call-it (the MIPS chip, I guess), and maybe AMD's new device (although it isn't intended to be a general purpose processor?). I still want to do it. Specifically, I want to create a SMALL working group of 1 engineer from each company, plus a couple of 'neutral' outsiders, to define a CPU intensive benchmark (as opposed to an AIM type system b'mark), code it in C (and maybe assembler) on each CPU, and run it under comparable conditions. We would then announce the results on the net. No official corporate endorsement or support is wanted or implied - it's just us engineers. I also want to avoid lengthy discussion on this net about the wisdom of this idea, the morality of b'marks, etc, or a lot of the other admittedly interesting points that were raised a coupled of months ago. John Mashey, if you're reading this, I recall you expressed considerable interest in this. Since we're neighbors, perhaps we should get together and get this thing moving. My phone is 721-8176. As far as the benchmark itself, I have two suggestions: 1. "Fix" the problems with Dhrystone. Modify it so it doesn't get ridiculously optimized, but don't cripple compilers by not permitting optimization. 2. Write a new benchmark. I suggest we write a small scheduler, some CPU and memory intensive tasks, and let it context-switch from a clock tick interrupt. This could be a prety good simulation of both Unix type applications and real time stuff. It is also a portable test. I suppose we ought to run this on a reference machine, probably the good ol' VAX780. (are there any still in service :-))? George Grenley