wcs@ho95b.UUCP (Bill Stewart) (11/19/84)
---- This is a reply to Mike Zaleski's (pegasus!mzal) article posted Nov. 2. Mike commented that the 2 was a bit slow, although his definition of slow was "doesn't respond instantly". There is a good (i.e. fixable) reason for this - the system comes configured with only 30 buffers, which means it *will* run, even if you have minimum hardware configuration on it (i.e. you ordered the 1/2 meg memory and removed one of the two boards.) However, even with the 1/2 meg configuration you should increase this (I don't know exactly how much; for a VAX you should use 25% of your memory.) To do this, you need the System Reconfiguration Package. The other reason people have called the 3B-2 slow is that a major UNIX(tm) reviewing magazine didn't read the manual before they ran their benchmarks. The AT&T 3B computers use 100 clock ticks per second instead of 60. Therefore, their benchmark, which used clock ticks for timing, reported a 66% too long time. (They were comparing supermicros, including the 3B2 and a bunch of 68000-based systems. Using the corrected numbers, the 3B2 was among the to 2 or 3 machines, instead of the bottom 2 or 3, for integer-CPU and disk-related benchmarks. It didn't have floating-point hardware, so the floating-point performance was not very exciting, nut that should be available soon. Speaking unofficially and *NOT FOR AT&T*, Bill Stewart -- Bill Stewart AT&T Bell Labs, Holmdel NJ 1-201-949-0705 ...!ihnp4!ho95b!wcs
brad@bradley.UUCP (11/24/84)
As I understand it, the floating point hardware will take away four ports from the machine. Whether or not it has hardware floating point is now reason for it to be about 6 times slower than an IBM PC with no 8087. I had the chance to see a 3b2 and it took over 18 minutes to run a simple program while on the IBM it took 3:47 and 9 sec on my 11/73 and 6 sec on my 11/44. If people are interested I can post the results. I still think that the 3b2 has a long way before it will fit into the market. BTW: We also have an 11/40 with a 'C' compiler that didn't use the hardware floating point on it (Version 6) and the 40 almost beat the 3b2. Bradley Smith UUCP: {cepu,ihnp4,noao,uiucdcs}!bradley!brad Text Processing ARPA: cepu!bradley!brad@UCLA-LOCUS Bradley University PH: (309) 676-7611 Ext. 446 Peoria, IL 61625
pedz@smu.UUCP (11/26/84)
was intended to test the cpu speed. It was very simple is an
approximate listing follows below. The same program was
run on a MacIntosh. The result was that the 3B2 took about
1.3 times the time it took the Mac. Note that this is only
a cpu speed test and not really a benchmark.
main()
{
register int i, j;
for (i = 0; i < 1000; i++)
for (j = 0; j < 1000; j++) ;
}
Perry
ps I used the Megamax C compiler for the Mac
ken@ihuxq.UUCP (ken perlow) (11/28/84)
-- >> ...The same program was >> run on a MacIntosh. The result was that the 3B2 took about >> 1.3 times the time it took the Mac. Note that this is only >> a cpu speed test and not really a benchmark. >> main() >> { >> register int i, j; >> for (i = 0; i < 1000; i++) >> for (j = 0; j < 1000; j++) ; >> } >> Perry Beware the spurious benchmark, my son! When you run null loops, what you're testing is the hardware implementation of loop-related instructions. To know whether you've really tested cpu speed you'll have to look at the assembly code generated by each compiler to see what is actually running on each machine. Efficient looping instructions are a good thing, of course, so the test is not worthless. -- *** *** JE MAINTIENDRAI ***** ***** ****** ****** 27 Nov 84 [7 Frimaire An CXCIII] ken perlow ***** ***** (312)979-7188 ** ** ** ** ..ihnp4!iwsl8!ken *** ***
Ron Natalie <ron@BRL-TGR> (11/28/84)
RE: Lousy floating point performance of AT&T machines. Phone switches don't need floating point.
perry@heurikon.UUCP (Perry Kivolowitz) (11/30/84)
Let's face it. The 3b2 is toy computer worthy of discussion only in the toy computer news group. Rotten products, obnoxious personnel, premium prices.. It just goes to show that... SHIT* *is a registered characteristic of AT&T (since the breakup).
dan@rna.UUCP (Dan Ts'o) (12/02/84)
> A sufficiently smart compiler generates > > i= 1000; > j= 1000; > > for the given code, and no loop instructions at all. Yeah! And a sufficiently smart compiler/OS would realize the entire program is a NOP, put a mark in its little black book to give you a lower priority and return with a question about your sanity...
gwyn@brl-tgr.ARPA (Doug Gwyn <gwyn>) (12/02/84)
> A sufficiently smart compiler generates > > i= 1000; > j= 1000; > > for the given code, and no loop instructions at all. And an even smarter compiler would generate no code at all.
bo@enea.UUCP (12/07/84)
A sufficiently smart compiler generates i= 1000; j= 1000; for the given code, and no loop instructions at all.