Path: utzoo!attcan!uunet!zaphod.mps.ohio-state.edu!julius.cs.uiuc.edu!apple!sun-barr!rutgers!cbmvax!daveh From: daveh@cbmvax.commodore.com (Dave Haynie) Newsgroups: comp.sys.amiga.hardware Subject: Re: RISC Amiga Keywords: A RISC Mac? So what? Message-ID: <15627@cbmvax.commodore.com> Date: 6 Nov 90 00:33:53 GMT References: <1156@iceman.jcu.oz> <1990Oct30.173846.6928@idt.unit.no> <1990Oct31.012758.26467@jarvis.csri.toronto.edu> <1990Oct31.170315.25808@idt.unit.no> <1990Oct31.165728.3196@jarvis.csri.toronto.edu> Reply-To: daveh@cbmvax.commodore.com (Dave Haynie) Organization: Commodore, West Chester, PA Lines: 52 In article <1990Oct31.165728.3196@jarvis.csri.toronto.edu> leblanc@eecg.toronto.edu (Marcel LeBlanc) writes: >Another thing to note is that based on published performance claims, >Motorola's statements of VAX-MIPS are not nearly as trustworthy as MIPS'. >Here are a few numbers comparing measured SPEC data vs. Published MIPS: >(these numbers were compiled by John Mashey) >SPEC SPEC SPEC Published System >Integer Float mark mips >14.6 10.8 12.2 17.0 Motorola 8864SP, 20MHz, 128K >18.3 13.5 15.2 21.0 Motorola 8864SP, 25MHz, 128K >21.4 15.8 17.8 25.0 Motorola 8612, 33MHz, 32K >19.4 16.8 17.8 20.0 MIPS Magnum, 25MHz, 64K >Motorola's idea of MIPS is not as inflated as other companies, but you do >have to take their claims with a grain of salt. Your point is correct, but lots of folks seem to get confused these days over the terminology in it. SPECmarks aren't a true measure of any "instructions/ second", but in fact a measure of preformance relative to a VAX 11/780 on a series of integer and floating point benchmarks. SPECmarks are certainly more accurate than any MIPS claim, especially since the SPEC series benchmarks are more "real-world" than programs like Dhrystone, often used to calculate MIPS. That's why marketing folks continue to speak in terms of MIPS and quote Dhrystone 1.1 numbers when these come out favorable (perhaps due to weak but fast instructions in the first case and optimizing compilers and/or Dhrystone-sized caches in the second case). The other main problem is that Motorola's claim of MIPS for any chip is just a claim for that Chip, everything being perfect. SPECmarks are a test of a real living system, which includes the memory bus performance, compiler performance, etc. Between real live 68030 systems on the market, you can see more than a 4:1 variation in performance based on memory and system architecture and on compiler performance. Benchmark suites like SPEC and AIM were designed to make an attempt at measuring this true system performance, unmasked by cache or cheap compiler tricks. So you would expect any chip claim to be the best possible performance with that system, which would usually be an impossible situation. These are usually generated with simulators even before the chip works. You don't want to pay for the 25MHz 68040 system with 0 wait state main memory that would be required to get Moto's MIPS claim for the 25MHz 68040, assuming that those claims are valid. Even with a big external cache, real live systems will fall short of the ideal, at least until we get some main memories that can keep up with these modern CPUs. >Marcel A. LeBlanc -- Electrical Eng. Computer Group, Univ. of Toronto -- Dave Haynie Commodore-Amiga (Amiga 3000) "The Crew That Never Rests" {uunet|pyramid|rutgers}!cbmvax!daveh PLINK: hazy BIX: hazy Standing on the shoulders of giants leaves me cold -REM