TigerSHARC BDTI score

Does anyone have seen the BDTI score for the TigerSHARC?

TigerSHARC (600MHz): 6280.
BlackFin (600MHz):   3360.
TMS320C64x (720Mz):  6480.

A far as I know the TigerSHARC is almost four times faster than the
BlackFin and two times faster than the C64x at 16 bit operations (for
the same clock frequency). So, why these scores? What am I missing?
Was used 32 bit math for scoring the TigerSHARC?

Luiz Carlos

Luiz Carlos wrote:

> Does anyone have seen the BDTI score for the TigerSHARC?

I didn't.

> 
> TigerSHARC (600MHz): 6280.
> BlackFin (600MHz):   3360.
> TMS320C64x (720Mz):  6480.
> 
> A far as I know the TigerSHARC is almost four times faster than
> the BlackFin and two times faster than the C64x at 16 bit
> operations (for the same clock frequency). So, why these scores?
> What am I missing? Was used 32 bit math for scoring the
> TigerSHARC?
> 
> Luiz Carlos

Just a guess:
If BDTI is an abbrev. - does it mean anything with ...Tex..Instr.. ?
If so, I wouldn't wonder why TMS performs best...

Bernhard

-- 
before sending to the above email-address:
replace deadspam.com by foerstergroup.de

Bernhard Holzmayer wrote:
> 
> Luiz Carlos wrote:
> 
> > Does anyone have seen the BDTI score for the TigerSHARC?
> 
> I didn't.
> 
> >
> > TigerSHARC (600MHz): 6280.
> > BlackFin (600MHz):   3360.
> > TMS320C64x (720Mz):  6480.
> >
> > A far as I know the TigerSHARC is almost four times faster than
> > the BlackFin and two times faster than the C64x at 16 bit
> > operations (for the same clock frequency). So, why these scores?
> > What am I missing? Was used 32 bit math for scoring the
> > TigerSHARC?
> >
> > Luiz Carlos
> 
> Just a guess:
> If BDTI is an abbrev. - does it mean anything with ...Tex..Instr.. ?
> If so, I wouldn't wonder why TMS performs best...
> 
> Bernhard
> 
> --
> before sending to the above email-address:
> replace deadspam.com by foerstergroup.de

Bernhard,

BDTI (Berkeley Design Technology, Inc.) is an independent business that
sells advice. They couldn't survive with a reputation for bias. 
The comp.dsp FAQ is on their server.    http://www.bdti.com/

Jerry
-- 
Engineering is the art of making what you want from things you can get.
�����������������������������������������������������������������������

"Luiz Carlos" <oen_br@yahoo.com.br> wrote in message
news:8471ba54.0308050257.5407f62a@posting.google.com...
> Does anyone have seen the BDTI score for the TigerSHARC?
>
> TigerSHARC (600MHz): 6280.
> BlackFin (600MHz):   3360.
> TMS320C64x (720Mz):  6480.
>
> A far as I know the TigerSHARC is almost four times faster than the
> BlackFin and two times faster than the C64x at 16 bit operations (for
> the same clock frequency). So, why these scores? What am I missing?
> Was used 32 bit math for scoring the TigerSHARC?

Because raw MIPS is only a number and doesn't mean jack when it comes to
real world operations. The BDTI folks use a bunch of benchmarking
methodology to come up with this score. BTW, I just looked at their July
newsletter and they were just getting started at testing the TigerSHarc 21x
series, which they claim will beat the C64x. So perhaps your numbers are for
the prev. version. In any case, go to their web site and understand what
they do in order to get these scores - perhaps that will give you enough
details to answer the 'why'.

Cheers
Bhaskar


>
> Luiz Carlos

Jerry Avins <jya@ieee.org> wrote in message news:<3F2FBF0F.65FD2052@ieee.org>...
> Bernhard Holzmayer wrote:
> > 
> > Luiz Carlos wrote:
> > 
> > > Does anyone have seen the BDTI score for the TigerSHARC?
> > 
> > I didn't.
> > 
> > >
> > > TigerSHARC (600MHz): 6280.
> > > BlackFin (600MHz):   3360.
> > > TMS320C64x (720Mz):  6480.
> > >
> > > A far as I know the TigerSHARC is almost four times faster than
> > > the BlackFin and two times faster than the C64x at 16 bit
> > > operations (for the same clock frequency). So, why these scores?
> > > What am I missing? Was used 32 bit math for scoring the
> > > TigerSHARC?
> > >
> > > Luiz Carlos
> > 
> > Just a guess:
> > If BDTI is an abbrev. - does it mean anything with ...Tex..Instr.. ?
> > If so, I wouldn't wonder why TMS performs best...
> > 
> > Bernhard
> > 
> > --
> > before sending to the above email-address:
> > replace deadspam.com by foerstergroup.de
> 
> Bernhard,
> 
> BDTI (Berkeley Design Technology, Inc.) is an independent business that
> sells advice. They couldn't survive with a reputation for bias. 
> The comp.dsp FAQ is on their server.    http://www.bdti.com/
> 
> Jerry

Good to remember that. But, any further opinion about the scores? I am
also surprised. I would have expected something better.

Luiz: have you asked BDTI people?

JaaC

> Because raw MIPS is only a number and doesn't mean jack when it comes to
> real world operations. The BDTI folks use a bunch of benchmarking
> methodology to come up with this score. BTW, I just looked at their July
> newsletter and they were just getting started at testing the TigerSHarc 21x
> series, which they claim will beat the C64x. So perhaps your numbers are for
> the prev. version. In any case, go to their web site and understand what
> they do in order to get these scores - perhaps that will give you enough
> details to answer the 'why'.
> 
> Cheers
> Bhaskar
> 

Hi Bhaskar,

I�ve read the BDTI white paper for scoring the DSPs, and the tests
include, real and complex FIRs, IIR biquads, vector dot products,
vector add, vector max, 256 point FFT, etc. In almost all these
routines the TIGERSharc is (or should be) four times faster (at 16
bit) than the BlackFin and two times faster than the C64x (same clock
frequency). So, my question remains: what am I missing?
I�m not talking about MIPs, but real 16 bit MACs, 16 bit max/min, 16
bit add, etc.
I know the three architectures and I was really expecting a bigger
score for the TIGERSharc. I�m planning using it in one design and I
would like to know if there is something hidden.
I�m new to this newsgroup. Does people from ADI, TI, posts here?

Regards

Luiz Carlos

"Luiz Carlos" <oen_br@yahoo.com.br> wrote in message
news:8471ba54.0308051415.4d32e9c5@posting.google.com...
> > Because raw MIPS is only a number and doesn't mean jack when it comes to
> > real world operations. The BDTI folks use a bunch of benchmarking
> > methodology to come up with this score. BTW, I just looked at their July
> > newsletter and they were just getting started at testing the TigerSHarc
21x
> > series, which they claim will beat the C64x. So perhaps your numbers are
for
> > the prev. version. In any case, go to their web site and understand what
> > they do in order to get these scores - perhaps that will give you enough
> > details to answer the 'why'.
> >
> > Cheers
> > Bhaskar
> >
>
> Hi Bhaskar,
>
> I&#4294967295;ve read the BDTI white paper for scoring the DSPs, and the tests
> include, real and complex FIRs, IIR biquads, vector dot products,
> vector add, vector max, 256 point FFT, etc. In almost all these
> routines the TIGERSharc is (or should be) four times faster (at 16
> bit) than the BlackFin and two times faster than the C64x (same clock
> frequency). So, my question remains: what am I missing?
> I&#4294967295;m not talking about MIPs, but real 16 bit MACs, 16 bit max/min, 16
> bit add, etc.
> I know the three architectures and I was really expecting a bigger
> score for the TIGERSharc. I&#4294967295;m planning using it in one design and I
> would like to know if there is something hidden.
> I&#4294967295;m new to this newsgroup. Does people from ADI, TI, posts here?

Not that many. Keith from TI posts fairly often. I suspect there are others
who probably read the posts but never write.

If you plan to use it in a design, I suggest that you prototype what matters
to your design and check to see how the TigerSharc fares.
That way you know *exactly* how it performs for your application and since
you did the test yourself, you don't have to second guess yourself.
As for the BDTI scores, ask BDTI for an explanation (first offer your theory
as to why you expect something different) - more than likely, they will give
you references to their methodology. You can also ask ADI reps about this
and let them do the hard work of justifying why the BDTI score is inaccurate
(or otherwise).

>
> Regards
>
> Luiz Carlos

Jerry Avins wrote:

> Bernhard,
> 
> BDTI (Berkeley Design Technology, Inc.) is an independent business
> that sells advice. They couldn't survive with a reputation for
> bias.
> The comp.dsp FAQ is on their server.    http://www.bdti.com/

> Jerry

Right. I should have remembered that! 
Thanks for the hint.
Bernhard

Luiz Carlos wrote:

>> Because raw MIPS is only a number and doesn't mean jack when it
>> comes to real world operations. The BDTI folks use a bunch of
>> benchmarking methodology to come up with this score. BTW, I just
>> looked at their July newsletter and they were just getting
>> started at testing the TigerSHarc 21x series, which they claim
>> will beat the C64x. So perhaps your numbers are for the prev.
>> version. In any case, go to their web site and understand what
>> they do in order to get these scores - perhaps that will give you
>> enough details to answer the 'why'.
>> 
>> Cheers
>> Bhaskar
>> 
> 
> Hi Bhaskar,
> 
> I?ve read the BDTI white paper for scoring the DSPs, and the tests
> include, real and complex FIRs, IIR biquads, vector dot products,
> vector add, vector max, 256 point FFT, etc. In almost all these
> routines the TIGERSharc is (or should be) four times faster (at 16
> bit) than the BlackFin and two times faster than the C64x (same
> clock frequency). So, my question remains: what am I missing?
> I?m not talking about MIPs, but real 16 bit MACs, 16 bit max/min,
> 16 bit add, etc.
> I know the three architectures and I was really expecting a bigger
> score for the TIGERSharc. I?m planning using it in one design and
> I would like to know if there is something hidden.
> I?m new to this newsgroup. Does people from ADI, TI, posts here?
> 
> Regards
> 
> Luiz Carlos

Hi Luiz,
I was comparing TIs processors with Sharc some time ago, and I found 
that the performance is depending on what I'm going to do with it.
Check the relevance of 
- floating point processing
- 32bit/40bit width
- command pipelining
- DMA capabilities
- built-in interfaces/protocols
Depending on what you really do you'll find the one or the other 
better.
If the one or the other is faster, sometimes depends on that.
If the processor has to "manually" implement what the DMA processor 
of the other does in the background, this may cost you a lot of 
performance. 
If the command pipeline holds a complete loop and saves you memory 
access this might increase speed - but what if pipelining fails in 
your case? 

Without knowing your application, it's difficult to give an advice.
Nevertheless, I'd say: TigerSharc is a 32bit-DSP.
If you really need a 16bit-DSP there might be better choices out 
there. 

I talked to a couple of developers who started with TI processors 
(62xx/67xx) and ended up with their "better" choice Sharc.
Others seem to be happy with TI and the CodeComposer Studio.
By the way: don't forget that the programming tools are also very 
important - you'll have to deal with them all the time.   

Bernhard

-- 
before sending to the above email-address:
replace deadspam.com by foerstergroup.de

"Bhaskar Thiagarajan" <bhaskart@deja.com> wrote 
> 
> Not that many. Keith from TI posts fairly often. I suspect there are others
> who probably read the posts but never write.
> 
> If you plan to use it in a design, I suggest that you prototype what matters
> to your design and check to see how the TigerSharc fares.
> That way you know *exactly* how it performs for your application and since
> you did the test yourself, you don't have to second guess yourself.
> As for the BDTI scores, ask BDTI for an explanation (first offer your theory
> as to why you expect something different) - more than likely, they will give
> you references to their methodology. You can also ask ADI reps about this
> and let them do the hard work of justifying why the BDTI score is inaccurate
> (or otherwise).

In the FPGA newsgroups the manufacturers (development engineers from
Xilinx, Altera, ...) and other very skilled people participate
actively. So, asking for information in the newsgroup is normally very
interesting: solid information and lots of opinions/ideas.

I thought somebody of ADI would like to enlighten this!

Luiz Carlos