home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.arch
- Path: sparky!uunet!charon.amdahl.com!pacbell.com!sgiblab!zaphod.mps.ohio-state.edu!sol.ctr.columbia.edu!ira.uka.de!gmd.de!Germany.EU.net!rrz.uni-koeln.de!unidui!math.fu-berlin.de!news.th-darmstadt.de!adams
- From: adams@pdv3.fmr.maschinenbau.th-darmstadt.de (Adams)
- Subject: Re: RTX and SC32
- Sender: news@news.th-darmstadt.de (The News System)
- Message-ID: <ADAMS.92Nov7002912@PDV2.pdv3.fmr.maschinenbau.th-darmstadt.de>
- In-Reply-To: doconnor@sedona.intel.com's message of 5 Nov 92 08: 54:08 GMT
- Date: Sat, 7 Nov 1992 00:29:12 GMT
- References: <17102@mindlink.bc.ca> <1992Nov3.144748.21826@sobeco.com>
- <ADAMS.92Nov5001909@PDV2.pdv3.fmr.maschinenbau.th-darmstadt.de>
- <DOCONNOR.92Nov5085408@potato.sedona.intel.com>
- Nntp-Posting-Host: pdv2.fmr.maschinenbau.th-darmstadt.de
- Organization: TH-Darmstadt
- Lines: 66
-
- In article <DOCONNOR.92Nov5085408@potato.sedona.intel.com> doconnor@sedona.intel.com (Dennis O'Connor) writes:
-
- ... and I was cited:
-
- > adams@pdv3.fmr.maschinenbau.th-darmstadt.de (Adams) writes:
- > ] The FRP1600 does a 16x16 multiply in one cycle (100ns--80ns--66ns).
- > ]
- > ] A 10MHz FRP1600 does 16x16bit multiply at least 5 times faster than
- > ] a 33MHz AMD29000.
- >
-
- To say the least, I was comparing CPU architectures costing about the
- same. A 16 MHz AMD29k and an evaluation board of FRP1600 were priced
- the about same, less than 100 US$.
-
- > The Intel i960(R) CA microprocessor does a 32x32->32 bit multiply in
- > 4 or 5 cycles, depending on the data. It does the multiply in
-
- .... Remarks about Intel 960 discarded.
-
- One should remember, i960 was out of discussion because of price, at
- least here in Germany at evaluation time (about 1990). 800 US $ /chip
- were too much, cost for development systems not included.....
-
- > And remember, it takes 4 16x16 multiplies and 3 32-bit adds
- > ( minimum ) to do a 32x32 bit multiply. How long does that
- > take on the FRP1600 ?
-
- They had well known requirements, multiplying 12-14 bit operands with
- 12-14 bit coeffizients. 32 bit multiplications were not necessary.
- Summing up 3 products, 16 bit were sufficient.
-
- > Some older RISC architectures didn't implement fast multiply
- > at first but added it in later products. There's no causal
- > relationship between architectural class ( RISC, CISC, Stack, VLIW )
- > and multiply speed.
-
- No, but silicon already gone for cache is not available for
- multipliers ;-<.
-
- > ] This example was not dedicated to any particular advantage of stack
- > ] machines, but to an area RISC chips did tend to loose.
- >
- > The i960 CA RISC (mainly) microprocessor has been around for years.
- > You didn't compare to the right RISC chip.
-
- PRICE in Germany. To say it frankly, i960 and development system were not
- affordable.
-
- > ] For realtime and controller applications, FORTH chips and stack machines
- > ] may blow RISCs out of the water, yes, they run much faster....
- >
- > The i960 family has delivered over 1 million units into embedded
- > and control applications. In typical Intel fashion, I tend to
- > measure an arhcitectures success by the revenue it generates ...
- > after all, the silicon itself is only one peice of the solution.
-
- An i960 was neither required nor economical feasible. The guys implementing
- it took the fastest and cheapest path. They started in FORTH with
- Harris and went on with FRP1600.
-
- Yet to be convinced , adams
-
-
-
-
-