Path: utzoo!utgpu!news-server.csri.toronto.edu!mailrus!wuarchive!zaphod.mps.ohio-state.edu!rpi!crdgw1!antarctica!davidsen From: davidsen@antarctica.crd.GE.COM (william E Davidsen) Newsgroups: comp.sys.ibm.pc Subject: Re: Whis is fastest 386/33 or 486/25 ? Message-ID: <9684@crdgw1.crd.ge.com> Date: 12 Jul 90 17:42:19 GMT References: <1990Jul11.161138.13630@dvinci.usask.ca> <217@news.nd.edu> Sender: news@crdgw1.crd.ge.com Reply-To: davidsen@crdos1.crd.ge.com Distribution: na Organization: GE Corporate R&D Center, Schenectady NY Lines: 108 In article <217@news.nd.edu>, laughner@news.nd.edu (Tom laughner) writes: |> There would be no difference in speed between a 386 with a math |> coprocessor and a 486. The 486 chip is a 386 + the math coprocessor in |> one. Intel considers the 486 as a part of the 386 family. A popular misconception, but untrue. Let's look at some actual test results which will show that the 486 *at the same speed* is about twice as fast as the 386. These tests were run on 25MHz machines. Note that while integer mpy and divide are about the same, add/sub are faster, bringing the weighted average up. The test and branch, which is a pretty good indicator of performance for non-math stuff such as compilers, shows better than 2:1. Float is about twice as fast. Overall on a 48 program suite I got a2.6 times faster for the 486. System id: HP 486-25, SCO ODT 1.0, 10MB, 300MB, cc Unoptimized math results: Math operations, effective instructions/sec (thousands) Add Sub Mpy Div Wtd. Avg. short: 17347.8 17538.5 3391.3 2845.4 11740.5 long: 19814.7 19600.0 3103.4 1978.0 12897.0 float: 4266.7 4266.7 3824.2 1500.0 3741.1 double: 4133.3 4043.5 3243.2 1360.8 3468.0 int: 19912.1 19384.6 3130.4 2000.0 12871.6 Test and branch timing: integer compare and branch 0.247 uSec, 4054.0K/sec float compare and branch 0.850 uSec, 1176.5K/sec System id: Dell 325, 4MB, 150MB, Xenix/386 2.3.3, 387 Unoptimized math results: Math operations, effective instructions/sec (thousands) Add Sub Mpy Div Wtd. Avg. short: 7088.6 7169.8 2948.7 2631.6 5409.4 long: 7378.6 7169.8 2631.6 1842.1 5298.7 float: 1097.6 1097.6 921.1 897.4 1023.4 double: 963.9 963.9 722.9 786.5 877.0 int: 7378.6 7200.0 2600.0 1818.2 5296.3 Test and branch timing: integer compare and branch 0.688 uSec, 1453.5K/sec float compare and branch 4.320 uSec, 231.5K/sec And if you like some standard benchmarks, Dhrystone 2.1 486: Dhrystone benchmark: Dhrystone Benchmark, Version 2.1 (Language: C) Program compiled without 'register' attribute Please give the number of runs through the benchmark: Execution starts, 600000 runs through Dhrystone Microseconds for one run through Dhrystone: 63.1 Dhrystones per Second: 15838.1 Dhrystone Benchmark, Version 2.1 (Language: C) Program compiled with 'register' attribute Please give the number of runs through the benchmark: Execution starts, 600000 runs through Dhrystone Microseconds for one run through Dhrystone: 61.3 Dhrystones per Second: 16304.3 386: Dhrystone Benchmark, Version 2.1 (Language: C) Program compiled without 'register' attribute Please give the number of runs through the benchmark: Execution starts, 500000 runs through Dhrystone Microseconds for one run through Dhrystone: 126.2 Dhrystones per Second: 7923.9 Dhrystone Benchmark, Version 2.1 (Language: C) Program compiled with 'register' attribute Please give the number of runs through the benchmark: Execution starts, 500000 runs through Dhrystone Microseconds for one run through Dhrystone: 120.0 Dhrystones per Second: 8333.3 Note that these figures match those reported in most of the magazines. The 486 is *not* a 386 with a built-in 387, it is a whole new chip with the same instruction set, but very diferent instruction timings, with the speedup in the common instructions. -- Bill Davidsen (davidsen@crdos1.crd.ge.com, uunet!crdgw1!crdos1!davidsen) GE Corp R&D Center, Schenectady NY Moderator of comp.binaries.ibm.pc and 386users mailing list "Stupidity, like virtue, is it's own reward" -me