eriks@yetti.UUCP (Eriks Rugelis) (01/30/86)
we received a brand new IBM RT PC on a one week demo from IBM; i thought
i'd run a quick series of simple benchmarks for rough speed comparisons
the machine is an IBM RT PC running the Interactive Systems port plus
modifications of
AT&T System V.1 that IBM has named AIX (Advanced Interactive eXecutive)
the machine is configured with:
2 MB and the floating point accelerator board on the system bus
a PC/AT co-processor board with its own 1/2 MB memory on the PC bus
various other peripherals (asynch. terminal mux, disk controller..etc
on the PC bus)
a 70 MB hard disk and an IBM 1.2 MB floppy drive
i believe that this is known as a Model 20 in the RT PC series (maybe a Model
25, i don't have my RT PC glossies handy)
i ran a copy of the dhrystone benchmark, put together by Rick Richarson, and
labeled as Version C/1 dated 12/01/84
as AIX booted it reported its version as 'B47 release 1 version 0' (for all
i know this might just as well be referring to the version of the Virtual
Resource Manager... i don't know... does ANYONE?)
at 50K iterations
the no-register benchmark clocked in at 1736 dhrystones/second
the register benchmark produced 1879 dhrystones/second
at 500K iterations the register benchmark produced 1883 dhrystones/second
compiling with the fcc command that allows use of the floating point accelerator
did not yield any different numbers
i also compiled and ran David Hinnants C language port of the single
precision whetstone benchmark
using the admittedly crude technique of using a time command to gather
numbers, 1 million iterations took about 5 seconds or less to complete
the whetstone benchmark was run with the floating point board enabled
observations and comments:
IBM claims that the RT PC is a '2 MIPS' machine... the dhrystones seem to
bring to the fore the contrast between CISC and RISC mips... a VAX 780
clocks around 1500 dhrystones and a 785 clocks around 2100...
i have heard these to be described as about 1 and 1.5 mips machines respectively
IBM claims about 200K single whetstones/second... my gross measurements appear
to bear this out
trademarks:
this article is full of references to trademarks owned by several corporate
entities; the reader is hereby advised not to take these tm's in vain
more comments:
the RT that i used today spent the previous night in a truck in -12C weather...
it was delivered in the morning and allowed warm-up through-out the day before
being turned on.... if this example is any indication, these machines seem
to stand up well to freeze-drying
a flame:
AIX continues the IBM tradition of:
distributing UNIX like systems BUT,
not associating the name UNIX with said systems AND,
- - -> not including machine-readable documentation with the distribution
of the operating system; not even committing to EVER providing
machine readable documentation of ANY sort
i like on-line manuals; i bet you do too; if you have any contact with IBM,
tell them.. then tell them again
--
Voice: Eriks Rugelis
Ma Bell: 416/665-8585 x6308
USENET: {allegra|decvax|ihnp4}!utzoo!yetti!eriks
NETNORTH/BITNET: eriks@yuorion
QUOTE: Of all the things I've lost, I miss my mind the mostmcgrath@uicsrd.CSRD.UIUC.EDU (02/03/86)
>/* Written 11:52 pm Jan 29, 1986 by eriks@yetti.UUCP in uicsrd:net.arch */ >/* ---------- "dhrystones on the IBM RT PC" ---------- */ >we received a brand new IBM RT PC on a one week demo from IBM; i thought >i'd run a quick series of simple benchmarks for rough speed comparisons > .... >compiling with the fcc command that allows use of the floating point accelerator >did not yield any different numbers We also saw a demo. The IBM rep "didn't want to talk about floating point" with the implication that it didn't really exist yet. Who knows? R. E. McGrath mcgrath%uicsrd@uiuc.arpa uiucdcsa!uicsrd!mcgrath
rentsch@unc.UUCP (Tim Rentsch) (02/07/86)
I just read an IBM product announcement for the RT PC. The bottom line on floating point: 200,000 whetstones (compared to 750,000 on a vax780, or 1,150,000 for 780 with FPA). well, it's a benchmark. # "i can't resist adding the following -- just this once, i promise" # # self standardDisclaimer. # self signature. # # "take that, you C fanatics"
stubbs@ncr-sd.UUCP (Jan Stubbs) (02/08/86)
In article <1000008@uicsrd> mcgrath@uicsrd.CSRD.UIUC.EDU writes: > >We also saw a demo. The IBM rep "didn't want to talk about floating point" >with the implication that it didn't really exist yet. Who knows? > I saw the IBM/RT at uniforum. A friendly person who claimed to head up the chip development team at Austin, Texas said the following: The floating point chip is a National Semiconductor NS32????. The ROMP is not an 801 (which was a discrete system) but used some similar concepts. It has a 23Mhz clock but uses four clocks per cycle which gives it a 170NS cycle time. 80 out of 120 instructions execute in 1 cycle, but the average cycles per instruction was >3. Load and store take 5 cycles each. There is no cache. There is a 16 word instruction buffer. The MMU was added late in the design (presumably to run UNIX??). There are 16 General purpose registers. The operating system emulates DOS commands, so I believe that copy a b does the same as cp a b if you are in DOS mode. Only one person can use the 80286 at a time. The 80286 is an option. Xenix was also shown in the booth along with some applications such as accounting and engineering stuff. I also saw a 12 cpu Encore Multimax running a similated load with 128 processes. While this was running I logged on and ran a 'C' compile, which completed in a very reasonable time considering the load. Jan Stubbs ....sdcsvax!ncr-sd!stubbs NCR Corp. San Diego