alan@mq.UUCP (Alan H. Mintz) (09/05/90)
A while back, someone posted a "Whetstone" benchmark to comp.sys.dec in both FORTRAN and C (from AURORA::HALLYB). I compiled said benchmark for two machines: 1) MicroVax3100 / VMS5.3 / VAXC 2.2 2) Dell Sys310 (386/20) w/80387 FPU / Dell UNIX SVR3.2 (ISC 2.0.?) / "Stock" C The puzzling thing is that performance on the 386 was 20% SLOWER than the MV3100. From what I have been told, the 3100 is a relatively low-end machine and a 386 should outperform it (the Dell is a pretty fast machine). The general "response" of the MV3100 seems to go along with this. Is there something unusual about VAXC or the architecture of the machine that could cause this ? -- < Alan H. Mintz | Voice +1 714 980 1034 > < Micro-Quick Systems, Inc. | FAX +1 714 944 3995 > < 10384 Hillside Road | uucp: ...!uunet!mq!alan > < Alta Loma, CA 91701 USA | Internet: alan@MQ.COM >
tihor@acf4.NYU.EDU (Stephen Tihor) (09/05/90)
General explainations that leap to mind are (a) benchmarks of this sort are pretty bogus, consider the SPEC suite instead; (b) DEC builds better I/O systems that most microvendors.
nichols@vms.macc.wisc.edu (Dennis Nichols, MACC) (09/06/90)
In article <53@mq.UUCP>, alan@mq.UUCP (Alan H. Mintz) writes... >1) MicroVax3100 / VMS5.3 / VAXC 2.2 >2) Dell Sys310 (386/20) w/80387 FPU / Dell UNIX SVR3.2 (ISC 2.0.?) / "Stock" C >The puzzling thing is that performance on the 386 was 20% SLOWER than the >MV3100. Maybe not so puzzling. From the February 1989 and April 1989 issues of MIPS, in round numbers: Dhrystones Whetstones-single Whetstones-double VAXstation 3100 4000 3000 2000 Dell 310-20+80387 4700-4900 1600 1400 Further details available from the issues cited above. Dennis Nichols 1210 West Dayton Street 608-262-1936 Academic Computing Center Madison, WI 53706 nichols@macc.wisc.edu Univ. of Wisconsin-Madison UNITED STATES nichols@wiscmacc.bitnet
moss@cs.umass.edu (Eliot Moss) (09/06/90)
Sometimes this tells a lot more about a *compiler* than it tells about hardware. My guess is that the 80x86 compilers may not be as good at some fancy optimizations, whereas the DEC C compiler is pretty good. If you want to compare the hardware alone, code in assembly (but that does not give a fair comparison if your programming will be in C). Another thing to do is to use more similar compilers, e.g., the GNU C compielr, which I believe is available for both of these machines/OSs. I also agree with the other respondent that it is not necessarily a very informative benchmark; be very careful in this kind of comparison! I would also not call a 3 MIP (or so) VAX "low-end"; one VAX instruction can do a lot more than one 80x86 instruction in many cases, and workstation I/O and memory systems are perhaps better designed than PC class boxes (and mainframes even better than that, etc.). -- J. Eliot B. Moss, Assistant Professor Department of Computer and Information Science Lederle Graduate Research Center University of Massachusetts Amherst, MA 01003 (413) 545-4206; Moss@cs.umass.edu
meissner@osf.org (Michael Meissner) (09/06/90)
In article <MOSS.90Sep5193730@ibis.cs.umass.edu> moss@cs.umass.edu (Eliot Moss) writes: | Sometimes this tells a lot more about a *compiler* than it tells about | hardware. My guess is that the 80x86 compilers may not be as good at some | fancy optimizations, whereas the DEC C compiler is pretty good. If you want to | compare the hardware alone, code in assembly (but that does not give a fair | comparison if your programming will be in C). Another thing to do is to use | more similar compilers, e.g., the GNU C compielr, which I believe is available | for both of these machines/OSs. I also agree with the other respondent that it | is not necessarily a very informative benchmark; be very careful in this kind | of comparison! I would also not call a 3 MIP (or so) VAX "low-end"; one VAX | instruction can do a lot more than one 80x86 instruction in many cases, and | workstation I/O and memory systems are perhaps better designed than PC class | boxes (and mainframes even better than that, etc.). Two notes: 1) the MIPS assembler does backend optimizations -- if you use GCC with it, you get some of the optimizations normally done with the MIPS compiler suite; 2) If you are using GCC, I did a lot of tuning and patching, so you probably want to start with my patches or wait to 1.38. The experience of tuning and hacking the 88k and MIPS ports for GCC has showed me that there is a lot of stuff that can be done in the machine dependent layer of the compiler that affects performance drastically. Sometimes changing one byte (such as in a constraint) can mean quite a bit. This means that you still have an apples vs. oranges comparison if you compare GCC ports (though not as much of one). It depends on how much time & effort has been spent tuning things for each port. A young port of GCC will tend to be less optimal than it could be. -- Michael Meissner email: meissner@osf.org phone: 617-621-8861 Open Software Foundation, 11 Cambridge Center, Cambridge, MA, 02142 Do apple growers tell their kids money doesn't grow on bushes?
alan@mq.UUCP (Alan H. Mintz) (09/08/90)
In response to an article I posted wrt benchmark result diffs between a MV3100 and 386 UNIX, In article <13310052@acf4.NYU.EDU>, tihor@acf4.NYU.EDU (Stephen Tihor) writes: > General explainations that leap to mind are (a) benchmarks of this sort > are pretty bogus, consider the SPEC suite instead; (b) DEC builds better I/O > systems that most microvendors. This is pretty much the consensus of the reponses I got. It does confirm what I have been told from the DECies, as opposed to PCers that indicated that the 386 should perform better. Are the SPECmarks commercial, or are they available electronically (free) ? -- < Alan H. Mintz | Voice +1 714 980 1034 > < Micro-Quick Systems, Inc. | FAX +1 714 944 3995 > < 10384 Hillside Road | uucp: ...!uunet!mq!alan > < Alta Loma, CA 91701 USA | Internet: alan@MQ.COM >
tihor@acf4.NYU.EDU (Stephen Tihor) (09/09/90)
try contacting mashey@mips.com.
khb@chiba.Eng.Sun.COM (Keith Bierman - SPD Advanced Languages) (09/11/90)
Rather than bother John, you might contact SPEC directly...
Kim Shanley Waterside 415-792-2901
415-792-4748(Fax)
shanley@cup.portal.com
Their address is:
WATERSIDE ASSOCIATES, ATTN: KIM SHANLEY
39510 Paseo Padre Pkwy
Suite 350
Freemont, CA 94538
(USA)
SPEC codes are themselves freely available, but SPEC puts them,
makefiles and reports available for a fee.
--
----------------------------------------------------------------
Keith H. Bierman kbierman@Eng.Sun.COM | khb@chiba.Eng.Sun.COM
SMI 2550 Garcia 12-33 | (415 336 2648)
Mountain View, CA 94043