Why TI makes no further efforts in floating point DSP?

Every week when I receive the TI newsletter emails it is always the
same disappointment. I have not seen that TI made any progress in
floating point DSP for years.

C44: not supported
C33: 10 years old, it is still supported but no further software tools
available.
C67xx: not seen a new model for 5 years.

Can anybody bring in any encouraging news?

James
www.go-ci.com

DigitalSignal <digitalsignal999@yahoo.com> writes:

> Every week when I receive the TI newsletter emails it is always the
> same disappointment. I have not seen that TI made any progress in
> floating point DSP for years.
>
> C44: not supported
> C33: 10 years old, it is still supported but no further software tools
> available.
> C67xx: not seen a new model for 5 years.
>
> Can anybody bring in any encouraging news?

Hi James,

That's not true! They came out with the TMS320F283x fairly recently,
I believe.

  http://focus.ti.com/paramsearch/docs/parametricsearch.tsp?family=dsp&sectionId=2&tabId=2105&familyId=1413&paramCriteria=no
-- 
%  Randy Yates                  % "With time with what you've learned, 
%% Fuquay-Varina, NC            %  they'll kiss the ground you walk 
%%% 919-577-9882                %  upon."
%%%% <yates@ieee.org>           % '21st Century Man', *Time*, ELO
http://www.digitalsignallabs.com

DigitalSignal wrote:
> Every week when I receive the TI newsletter emails it is always the
> same disappointment. I have not seen that TI made any progress in
> floating point DSP for years.

Maybe there is little demand for it.

I would say that larger integer word sizes would be more
useful for most DSP problems than floating point.  For those
that do need floating point there are many very fast processors
available outside the DSP world.

What algorithm are you trying to implement?

-- glen

DigitalSignal <digitalsignal999@yahoo.com> wrote in news:d2a82475-2e00-4a52-
8bf4-451e6f97a55f@d19g2000prm.googlegroups.com:

> Every week when I receive the TI newsletter emails it is always the
> same disappointment. I have not seen that TI made any progress in
> floating point DSP for years.
> 
> C44: not supported
> C33: 10 years old, it is still supported but no further software tools
> available.
> C67xx: not seen a new model for 5 years.
> 
> Can anybody bring in any encouraging news?

Sure, use a SHARC ;-)

Al Clark
Danville Signal Processing, Inc.



> 
> James
> www.go-ci.com

Randy, good to know. Let me do some comparison to C33!

James
www.go-ci.com

On May 16, 3:10&#4294967295;pm, glen herrmannsfeldt <g...@ugcs.caltech.edu> wrote:
> DigitalSignal wrote:
> > Every week when I receive the TI newsletter emails it is always the
> > same disappointment. I have not seen that TI made any progress in
> > floating point DSP for years.
>
> Maybe there is little demand for it.
>
> I would say that larger integer word sizes would be more
> useful for most DSP problems than floating point. &#4294967295;For those
> that do need floating point there are many very fast processors
> available outside the DSP world.
>
> What algorithm are you trying to implement?
>
> -- glen

From his web site (signal analyzers) I suspect he just wants to
enhance the floating point DSP routines he already has and doesn't
want to rewrite them all in integer math.

Floating point DSP's for low volume/high cost products like signal
analyzers are much more appropriated then fix point as the saving in
software development more then make up for the slight % increase in
cost/sale price.

steve wrote:
> On May 16, 3:10 pm, glen herrmannsfeldt <g...@ugcs.caltech.edu> wrote:
>> DigitalSignal wrote:
>>> Every week when I receive the TI newsletter emails it is always the
>>> same disappointment. I have not seen that TI made any progress in
>>> floating point DSP for years.
>> Maybe there is little demand for it.
>>
>> I would say that larger integer word sizes would be more
>> useful for most DSP problems than floating point.  For those
>> that do need floating point there are many very fast processors
>> available outside the DSP world.
>>
>> What algorithm are you trying to implement?
>>
>> -- glen
> 
> From his web site (signal analyzers) I suspect he just wants to
> enhance the floating point DSP routines he already has and doesn't
> want to rewrite them all in integer math.
> 
> Floating point DSP's for low volume/high cost products like signal
> analyzers are much more appropriated then fix point as the saving in
> software development more then make up for the slight % increase in
> cost/sale price.

This is true, but economics appears to mean those low volume, high value 
users will be using Pentiums for floating point in the future.

A state of the art core costs too much to develop for the small volumes 
of the floating point market. The new floating point controller from TI 
is a relatively simple core, for lowish cost parts. I expect it built a 
lot on reuse of older designs, and so was affordable to develop. I doubt 
a 64x class floating point device will follow it, unless a dramatic 
change in use creates high volume markets for high end floating point DSP.

Steve

On Sat, 17 May 2008 20:09:50 -0700, steve wrote:

> Floating point DSP's for low volume/high cost products like signal
> analyzers are much more appropriated then fix point as the saving in
> software development more then make up for the slight % increase in
> cost/sale price.

The real question, then, is what is it about a floating point "DSP" that 
gives him an advantage over a garden-variety commodity CPU, which (these 
days) has orders of magnitude more floating point performance.  Not so 
good on the power consumption (battery life) though.  If it's just the 
old low latency for I/O issue, then perhaps he can partition the problem 
between a low-latency fixed-point I/O processor and a higher-latency 
higher-throughput display/analysis processor?  Can't see that that'd be 
too much of a problem for the OP though: it's a display device and really 
only needs human-scale latency.

I saw the other day that Marvell have announced a couple of interesting-
looking ARM-family parts, one of them a dual-core, both with vector 
floating point, at up to 1.2GHz.  Something like that, or perhaps one of 
the faster AMCC or Motorola PowerPC parts could fit the bill.  Intel's 
ATOM will be interesting in that space too, as will VIA's new Isaiah.

Cheers,

-- 
Andrew

A quick question: Can Pentium CPU directly talk to an A/D converter?
what hardware interface on Pentium you would use?

James
www.go-ci.com

Al, Is Sharc better?

James
www.go-ci.com