Article View: comp.sys.dec
Article #4352Re: "Are DSP Chips Obsolete?" (was re: Alpha fft performance)
From: Joe Buck
Date: Thu, 21 Jan 1993 21:26
Date: Thu, 21 Jan 1993 21:26
57 lines
3487 bytes
3487 bytes
I would have let this one drop, but since a DEC employee wants to bring it up again ... In article <1993Jan21.172924.1979@rdg.dec.com> jpt@salama.fno.dec.com (Jari Tavi) writes: >As a beginning, my personal opinion: I don't think that DSP's are >obsolete. Good for you. >Cost, power consumtion and such other things are something that you >could try asking if or when they are addressed by Alpha AXP. 21064 >will not be the only implementation... TI, Motorola, and Analog Devices, as well as several Japanese companies, have usable products for embedded real-time DSP available now (and for much of the last decade). If DEC is planning something in the future, good for you, but the paper appeared to be an effort for DEC to excuse itself for not being in the DSP market by claiming (falsely) that current products were obsolete and that we should all switch to the Alpha. >There's nothing in Alpha AXP architecture itself which conflicts with this. >I do agree that 21064 as we see it today may not be something one would use >in cellular phone or something similar, but Alpha isn't bad in issues like >memory bandwith, context switches or interrupt latencies either... Yes, you can make a hell of a workstation with a 21064. But are you implying that DEC plans to enter the DSP market with an Alpha-based product? Somehow, I doubt it, not unless DEC wishes to lose even *more* money. I think that, if the paper from DEC's Cambridge research lab and your comments are any indication, I suggest that DEC stick to markets it knows something about. The Alpha looks like a very fine design, something to be proud of. But it isn't good for everything. Embedded real-time DSP isn't about very high performance number crunching (except, of course, when it is). It's about just enough number crunching power to get the job done, with a design that makes it easy to verify that real-time constraints are met (that is, you can count cycles accurately and know that that is exactly how many cycles the operation will take). It's about putting the whole system on a very small number of chips, or on one chip (example: if you make a call that goes over a satellite link, you're probably talking through a chip that implements the LMS echo cancellation algorithm). Lots of times, these things are ASICs with a programmable DSP core. >If you have need for High Performance embedded/real Time systems, please >contact sales for product directions. Also Aeon has already implemented >worlds fastest single boarder based on Alpha AXP :-) If you're after getting the job done with products that are available today, contact Motorola, Texas Instruments, Analog Devices, or any of a half dozen other companies who understand the DSP market and don't go around publishing poorly conceived papers like "Are DSP Chips Obsolete?". -- Joe Buck jbuck@ohm.berkeley.edu
Technical Details
Message-ID:
<1jn4dn$lbf@agate.berkeley.edu>Path:
arc1-ca.pugleaf.net!sparky!uunet!cs.utexas.edu!sdd.hp.com!swrinde!zaphod.mps.ohio-state.edu!saimiri.primate.wisc.edu!ames!agate!forney.berkeley.edu!jbuckReferences:
<1993Jan5.134034.22043@ircam.fr> <1iknq6$6vt@agate.berkeley.edu> <1993Jan21.172924.1979@rdg.dec.com>Show Full Headers
Complete Article Headers
Path: arc1-ca.pugleaf.net!sparky!uunet!cs.utexas.edu!sdd.hp.com!swrinde!zaphod.mps.ohio-state.edu!saimiri.primate.wisc.edu!ames!agate!forney.berkeley.edu!jbuck From: jbuck@forney.berkeley.edu (Joe Buck) Newsgroups: comp.sys.dec,comp.dsp Subject: Re: "Are DSP Chips Obsolete?" (was re: Alpha fft performance) Date: 21 Jan 1993 21:26:15 GMT Organization: U. C. Berkeley Lines: 57 Message-ID: <1jn4dn$lbf@agate.berkeley.edu> References: <1993Jan5.134034.22043@ircam.fr> <1iknq6$6vt@agate.berkeley.edu> <1993Jan21.172924.1979@rdg.dec.com> NNTP-Posting-Host: forney.berkeley.edu