Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP Path: utzoo!mnetor!uunet!husc6!think!ames!amdahl!oliveb!intelca!rod From: rod@intelca.UUCP (Rod Skinner) Newsgroups: comp.sys.ibm.pc Subject: Re: Re: Intel Microprocessors Message-ID: <4381@intelca.UUCP> Date: Mon, 24-Aug-87 16:05:44 EDT Article-I.D.: intelca.4381 Posted: Mon Aug 24 16:05:44 1987 Date-Received: Wed, 26-Aug-87 01:12:18 EDT References: <1112@lznv.ATT.COM> <399@aucs.UUCP> <3225@cucca.columbia.edu> <789@unccvax.UUCP> <1924@Shasta.STANFORD.EDU> <138@ts <7042@steinm24 Aug 87 20:05:44 GMT Organization: Intel, Santa Clara, Ca. Lines: 63 >I will quote some figures from Byte magazine, July 1987 issue. I >personally do not doubt the Intel benchmarks at all. If they chose to >select benchmarks which show the best points of their product, can you >seriously believe that Motorola is so dumb that they don't do the same >thing? According to COMPAQ's documentation, the 287 is executing at 8MHz with the appropriate logic to slow down the CPU access to the 287. The Model 80 numbers from the August issue of BYTE have been added (as well as the corrections discovered, even impartial benchmarkers make mistakes). COMPAQ Fibo was 53.1 seconds and the Float was 4.41 seconds. FROM BYTE: vvvvv vvvvv vvvvv test 68010 68020 68020 68020 80386 80386 7.8MHz 16MHz 12.5MHz 15.7Mhz 16MHz 16Mhz MAC-II COMPAQ IBM 1w/s 1w/s 0w/s 1w/s 0.5w/s 1w/s -- 881@8 881@12 881@15 287@8 387@16 Fibonacci 264.0 71.6 70.2 83.7 53.1 57.4 time in sec Float 230.0 4.2 2.9 2.7 4.4 0.5 Sieve 64.7 14.9 12.8 16.7 6.0 6.5 Sort 111.3 19.8 12.6 22.4 9.7 9.5 Savage 1884.3 8.8 24.8 5.4 35.1 19.2 Dhrystones 574.0 2114.0 2702.0 2083.0 3703.7 3125.0 dhrys/sec >The only comment I would make on this is that the 80287 was used rather >than the 80387. This typically would improve the f.p. performance by at >least 2:1. The 80387 has much faster trig functions (I have seen figures >indicating 10:1), which would improve the Savage benchmark performance. >-- > bill davidsen (wedu@ge-crd.arpa) > {chinet | philabs | seismo}!steinmetz!crdos1!davidsen >"Stupidity, like virtue, is its own reward" -me The Float numbers that BYTE measured were very close to the 10x numbers that you might see moving from the 287-8 to a 387-16. The Savage numbers are however puzzling. The numbers are not even "frequency scaled". So when we pulled out our version of the Metaware High C compiler and executed the Savage benchmark, our numbers did not match. Without optimizations, the number was 6.0 seconds on the Model 80 only using the "387" switch. The same Savage benchmark executing on the Intel 386/24 Multibus I board with UNIX/386 System V and the Greenhills C 1.8.2H compiler took 3.9 seconds. The difference in these two compilers is their ability to generate inline transcendentals versus calling a subroutine. The Metaware High C compiler makes a call to process the exponential used in the Savage code while the Greenhills compiler processes it inline. The last time I saw the Mot Benchmark Report BR322, it was dated October 1986 and used information that Intel published in April 1985. Not what I would call current information. Rod Skinner Santa Clara Microcomputer Division Intel Corp. MS SC4-40 3065 Bowers Avenue Santa Clara, CA 95051 {mipos3,hplabs,oliveb,qantel}!intelca!rod