http://www.tomshardware.com/hardnews...02_135943.html

"JoVee" wrote in message
...
http://www.tomshardware.com/hardnews...02_135943.html


I saw that yesterday, as a matter of fact. It's a neat idea, but how cost
effective is it when you can get a DSP upgrade card for 500 bucks?

jb

"JoVee" wrote in message
...
http://www.tomshardware.com/hardnews...02_135943.html


I saw that yesterday, as a matter of fact. It's a neat idea, but how cost
effective is it when you can get a DSP upgrade card for 500 bucks?

jb

Anybody besides myself see the UAD-2 card with an Nvidia 3d chip set
installed?
Max Arwood

"reddred" wrote in message
...

"JoVee" wrote in message
...
http://www.tomshardware.com/hardnews...02_135943.html


I saw that yesterday, as a matter of fact. It's a neat idea, but how cost
effective is it when you can get a DSP upgrade card for 500 bucks?

jb

Anybody besides myself see the UAD-2 card with an Nvidia 3d chip set
installed?
Max Arwood

"reddred" wrote in message
...

"JoVee" wrote in message
...
http://www.tomshardware.com/hardnews...02_135943.html


I saw that yesterday, as a matter of fact. It's a neat idea, but how cost
effective is it when you can get a DSP upgrade card for 500 bucks?

jb

In article writes:

http://www.tomshardware.com/hardnews...02_135943.html

I saw that yesterday, as a matter of fact. It's a neat idea, but how cost
effective is it when you can get a DSP upgrade card for 500 bucks?

How much does the graphics card in question cost?

The thing is that a UA or t.c. DSP card is a fairly limited application
thing and therefore has a fairly limited market. A graphics board for
a consumer-level computer (even a hot-rod consumer) has potential for
far greater sales, so it has potential for far lower cost.

Which do you think is more likely - that UA or t.c. will enter the
consumer market so they can make their $500 cards to sell for $100, or
that an already established graphics card maker will sell more cards
because of the audio processing power available on them?

I don't (at least at this point) see this as a way to run serious
pro audio applications, but rather as a way to enhance consumer audio
coming out of a computer - much in the way that improved graphics
cards provide better video (as in "movie") capability than ever
before.


--
I'm really Mike Rivers )
However, until the spam goes away or Hell freezes over,
lots of IP addresses are blocked from this system. If
you e-mail me and it bounces, use your secret decoder ring
and reach me he double-m-eleven-double-zero at yahoo

In article writes:

http://www.tomshardware.com/hardnews...02_135943.html

I saw that yesterday, as a matter of fact. It's a neat idea, but how cost
effective is it when you can get a DSP upgrade card for 500 bucks?

How much does the graphics card in question cost?

The thing is that a UA or t.c. DSP card is a fairly limited application
thing and therefore has a fairly limited market. A graphics board for
a consumer-level computer (even a hot-rod consumer) has potential for
far greater sales, so it has potential for far lower cost.

Which do you think is more likely - that UA or t.c. will enter the
consumer market so they can make their $500 cards to sell for $100, or
that an already established graphics card maker will sell more cards
because of the audio processing power available on them?

I don't (at least at this point) see this as a way to run serious
pro audio applications, but rather as a way to enhance consumer audio
coming out of a computer - much in the way that improved graphics
cards provide better video (as in "movie") capability than ever
before.


--
I'm really Mike Rivers )
However, until the spam goes away or Hell freezes over,
lots of IP addresses are blocked from this system. If
you e-mail me and it bounces, use your secret decoder ring
and reach me he double-m-eleven-double-zero at yahoo

"Mike Rivers" wrote in message
news:znr1094384426k@trad...

In article
writes:

http://www.tomshardware.com/hardnews...02_135943.html

I saw that yesterday, as a matter of fact. It's a neat idea, but how
cost
effective is it when you can get a DSP upgrade card for 500 bucks?

How much does the graphics card in question cost?

The thing is that a UA or t.c. DSP card is a fairly limited application
thing and therefore has a fairly limited market. A graphics board for
a consumer-level computer (even a hot-rod consumer) has potential for
far greater sales, so it has potential for far lower cost.

Which do you think is more likely - that UA or t.c. will enter the
consumer market so they can make their $500 cards to sell for $100, or
that an already established graphics card maker will sell more cards
because of the audio processing power available on them?

I don't (at least at this point) see this as a way to run serious
pro audio applications, but rather as a way to enhance consumer audio
coming out of a computer - much in the way that improved graphics
cards provide better video (as in "movie") capability than ever
before.


My understanding from the article is that the software will sell for between
2 and 8 hundred dollars. Add that to the cost of a recent high-end video
card (which can cost 500 dollars in and of themselves) and you're looking at
a lot of dough for what I see as a 'neat' but imperfect solution.

It seems to me like it is something that needs to run from startup at a
deeper level than your regular app. If it can be turned on and off
painlessly, and costs around 1 or two hundred bucks, he may have a winner on
his hands.

jb




--
I'm really Mike Rivers )
However, until the spam goes away or Hell freezes over,
lots of IP addresses are blocked from this system. If
you e-mail me and it bounces, use your secret decoder ring
and reach me he double-m-eleven-double-zero at yahoo

"Mike Rivers" wrote in message
news:znr1094384426k@trad...

In article
writes:

http://www.tomshardware.com/hardnews...02_135943.html

I saw that yesterday, as a matter of fact. It's a neat idea, but how
cost
effective is it when you can get a DSP upgrade card for 500 bucks?

How much does the graphics card in question cost?

The thing is that a UA or t.c. DSP card is a fairly limited application
thing and therefore has a fairly limited market. A graphics board for
a consumer-level computer (even a hot-rod consumer) has potential for
far greater sales, so it has potential for far lower cost.

Which do you think is more likely - that UA or t.c. will enter the
consumer market so they can make their $500 cards to sell for $100, or
that an already established graphics card maker will sell more cards
because of the audio processing power available on them?

I don't (at least at this point) see this as a way to run serious
pro audio applications, but rather as a way to enhance consumer audio
coming out of a computer - much in the way that improved graphics
cards provide better video (as in "movie") capability than ever
before.


My understanding from the article is that the software will sell for between
2 and 8 hundred dollars. Add that to the cost of a recent high-end video
card (which can cost 500 dollars in and of themselves) and you're looking at
a lot of dough for what I see as a 'neat' but imperfect solution.

It seems to me like it is something that needs to run from startup at a
deeper level than your regular app. If it can be turned on and off
painlessly, and costs around 1 or two hundred bucks, he may have a winner on
his hands.

jb




--
I'm really Mike Rivers )
However, until the spam goes away or Hell freezes over,
lots of IP addresses are blocked from this system. If
you e-mail me and it bounces, use your secret decoder ring
and reach me he double-m-eleven-double-zero at yahoo

In article writes:

My understanding from the article is that the software will sell for between
2 and 8 hundred dollars. Add that to the cost of a recent high-end video
card (which can cost 500 dollars in and of themselves) and you're looking at
a lot of dough for what I see as a 'neat' but imperfect solution.

Yeeouch! That being the case, they'd better be able to show me that
they have more to offer than the current crop of DSP cards for audio
applications. I'll bet it's not trivial to port a popular auido
plug-in to the video card's GSP, so for a while anyway, all the
software will be coming from this company.


--
I'm really Mike Rivers )
However, until the spam goes away or Hell freezes over,
lots of IP addresses are blocked from this system. If
you e-mail me and it bounces, use your secret decoder ring
and reach me he double-m-eleven-double-zero at yahoo

In article writes:

My understanding from the article is that the software will sell for between
2 and 8 hundred dollars. Add that to the cost of a recent high-end video
card (which can cost 500 dollars in and of themselves) and you're looking at
a lot of dough for what I see as a 'neat' but imperfect solution.

Yeeouch! That being the case, they'd better be able to show me that
they have more to offer than the current crop of DSP cards for audio
applications. I'll bet it's not trivial to port a popular auido
plug-in to the video card's GSP, so for a while anyway, all the
software will be coming from this company.


--
I'm really Mike Rivers )
However, until the spam goes away or Hell freezes over,
lots of IP addresses are blocked from this system. If
you e-mail me and it bounces, use your secret decoder ring
and reach me he double-m-eleven-double-zero at yahoo

I just like that 40gflps!
Max Arwood

"Mike Rivers" wrote in message
news:znr1094418196k@trad...

In article
writes:

My understanding from the article is that the software will sell for
between
2 and 8 hundred dollars. Add that to the cost of a recent high-end video
card (which can cost 500 dollars in and of themselves) and you're
looking at
a lot of dough for what I see as a 'neat' but imperfect solution.

Yeeouch! That being the case, they'd better be able to show me that
they have more to offer than the current crop of DSP cards for audio
applications. I'll bet it's not trivial to port a popular auido
plug-in to the video card's GSP, so for a while anyway, all the
software will be coming from this company.


--
I'm really Mike Rivers )
However, until the spam goes away or Hell freezes over,
lots of IP addresses are blocked from this system. If
you e-mail me and it bounces, use your secret decoder ring
and reach me he double-m-eleven-double-zero at yahoo

I just like that 40gflps!
Max Arwood

"Mike Rivers" wrote in message
news:znr1094418196k@trad...

In article
writes:

My understanding from the article is that the software will sell for
between
2 and 8 hundred dollars. Add that to the cost of a recent high-end video
card (which can cost 500 dollars in and of themselves) and you're
looking at
a lot of dough for what I see as a 'neat' but imperfect solution.

Yeeouch! That being the case, they'd better be able to show me that
they have more to offer than the current crop of DSP cards for audio
applications. I'll bet it's not trivial to port a popular auido
plug-in to the video card's GSP, so for a while anyway, all the
software will be coming from this company.


--
I'm really Mike Rivers )
However, until the spam goes away or Hell freezes over,
lots of IP addresses are blocked from this system. If
you e-mail me and it bounces, use your secret decoder ring
and reach me he double-m-eleven-double-zero at yahoo

Mike Rivers wrote:

In article writes:


My understanding from the article is that the software will sell for between
2 and 8 hundred dollars. Add that to the cost of a recent high-end video
card (which can cost 500 dollars in and of themselves) and you're looking at
a lot of dough for what I see as a 'neat' but imperfect solution.


Yeeouch! That being the case, they'd better be able to show me that
they have more to offer than the current crop of DSP cards for audio
applications. I'll bet it's not trivial to port a popular auido
plug-in to the video card's GSP, so for a while anyway, all the
software will be coming from this company.

As far as the convolution power, which is what is really
needed today, it says that they will be eschewing the
inaccurate FFT method for the direct one. That will make it
run out of Hz far sooner than the gflop numbers alone would
indicate, as much as 1000 times sooner for a reasonably long
FIR. That is simply a reflection of the lack of flexibility
in reconfiguring or programming these things. If they could
use a partitioned FFT you can damn betcha there would be
_no_ reason not to.


Bob
--

"Things should be described as simply as possible, but no
simpler."

A. Einstein

Mike Rivers wrote:

In article writes:


My understanding from the article is that the software will sell for between
2 and 8 hundred dollars. Add that to the cost of a recent high-end video
card (which can cost 500 dollars in and of themselves) and you're looking at
a lot of dough for what I see as a 'neat' but imperfect solution.


Yeeouch! That being the case, they'd better be able to show me that
they have more to offer than the current crop of DSP cards for audio
applications. I'll bet it's not trivial to port a popular auido
plug-in to the video card's GSP, so for a while anyway, all the
software will be coming from this company.

As far as the convolution power, which is what is really
needed today, it says that they will be eschewing the
inaccurate FFT method for the direct one. That will make it
run out of Hz far sooner than the gflop numbers alone would
indicate, as much as 1000 times sooner for a reasonably long
FIR. That is simply a reflection of the lack of flexibility
in reconfiguring or programming these things. If they could
use a partitioned FFT you can damn betcha there would be
_no_ reason not to.


Bob
--

"Things should be described as simply as possible, but no
simpler."

A. Einstein

In article writes:

As far as the convolution power, which is what is really
needed today, it says that they will be eschewing the
inaccurate FFT method for the direct one. That will make it
run out of Hz far sooner than the gflop numbers alone would
indicate, as much as 1000 times sooner for a reasonably long
FIR. That is simply a reflection of the lack of flexibility
in reconfiguring or programming these things. If they could
use a partitioned FFT you can damn betcha there would be
_no_ reason not to.

You lost me in the first sentence, because I don't know how these
things work, nor do I feed a need to know. My thing is external
interfaces, trusting that what's inside the "black box" (be it
hardware, software, or soupware) will either do its job or not.

So is this a good thing, a bad thing, a potentially good thing if they
do it right (what isn't?), or a marketing thing?



--
I'm really Mike Rivers )
However, until the spam goes away or Hell freezes over,
lots of IP addresses are blocked from this system. If
you e-mail me and it bounces, use your secret decoder ring
and reach me he double-m-eleven-double-zero at yahoo

In article writes:

As far as the convolution power, which is what is really
needed today, it says that they will be eschewing the
inaccurate FFT method for the direct one. That will make it
run out of Hz far sooner than the gflop numbers alone would
indicate, as much as 1000 times sooner for a reasonably long
FIR. That is simply a reflection of the lack of flexibility
in reconfiguring or programming these things. If they could
use a partitioned FFT you can damn betcha there would be
_no_ reason not to.

You lost me in the first sentence, because I don't know how these
things work, nor do I feed a need to know. My thing is external
interfaces, trusting that what's inside the "black box" (be it
hardware, software, or soupware) will either do its job or not.

So is this a good thing, a bad thing, a potentially good thing if they
do it right (what isn't?), or a marketing thing?



--
I'm really Mike Rivers )
However, until the spam goes away or Hell freezes over,
lots of IP addresses are blocked from this system. If
you e-mail me and it bounces, use your secret decoder ring
and reach me he double-m-eleven-double-zero at yahoo

Mike Rivers wrote:

In article writes:


As far as the convolution power, which is what is really
needed today, it says that they will be eschewing the
inaccurate FFT method for the direct one. That will make it
run out of Hz far sooner than the gflop numbers alone would
indicate, as much as 1000 times sooner for a reasonably long
FIR. That is simply a reflection of the lack of flexibility
in reconfiguring or programming these things. If they could
use a partitioned FFT you can damn betcha there would be
_no_ reason not to.


You lost me in the first sentence, because I don't know how these
things work, nor do I feed a need to know.

Ok. Nonetheless the point was of some real signifigance.


Bob
--

"Things should be described as simply as possible, but no
simpler."

A. Einstein

Mike Rivers wrote:

In article writes:


As far as the convolution power, which is what is really
needed today, it says that they will be eschewing the
inaccurate FFT method for the direct one. That will make it
run out of Hz far sooner than the gflop numbers alone would
indicate, as much as 1000 times sooner for a reasonably long
FIR. That is simply a reflection of the lack of flexibility
in reconfiguring or programming these things. If they could
use a partitioned FFT you can damn betcha there would be
_no_ reason not to.


You lost me in the first sentence, because I don't know how these
things work, nor do I feed a need to know.

Ok. Nonetheless the point was of some real signifigance.


Bob
--

"Things should be described as simply as possible, but no
simpler."

A. Einstein

Mike Rivers wrote:
Yeeouch! That being the case, they'd better be able to show me that
they have more to offer than the current crop of DSP cards for audio
applications. I'll bet it's not trivial to port a popular auido
plug-in to the video card's GSP, so for a while anyway, all the
software will be coming from this company.

the chip used on the UAD-1 started life as a graphics processor.

--
Aaron J. Grier | "Not your ordinary poofy goof." |
"someday the industry will have throbbing frontal lobes and will be able
to write provably correct software. also, I want a pony." -- Zach Brown

Mike Rivers wrote:
Yeeouch! That being the case, they'd better be able to show me that
they have more to offer than the current crop of DSP cards for audio
applications. I'll bet it's not trivial to port a popular auido
plug-in to the video card's GSP, so for a while anyway, all the
software will be coming from this company.

the chip used on the UAD-1 started life as a graphics processor.

--
Aaron J. Grier | "Not your ordinary poofy goof." |
"someday the industry will have throbbing frontal lobes and will be able
to write provably correct software. also, I want a pony." -- Zach Brown

(Mike Rivers) wrote in message news:znr1094469385k@trad...
In article writes:

As far as the convolution power, which is what is really
needed today, it says that they will be eschewing the
inaccurate FFT method for the direct one. That will make it
run out of Hz far sooner than the gflop numbers alone would
indicate, as much as 1000 times sooner for a reasonably long
FIR. That is simply a reflection of the lack of flexibility
in reconfiguring or programming these things. If they could
use a partitioned FFT you can damn betcha there would be
_no_ reason not to.

You lost me in the first sentence, because I don't know how these
things work, nor do I feed a need to know. My thing is external
interfaces, trusting that what's inside the "black box" (be it
hardware, software, or soupware) will either do its job or not.

So is this a good thing, a bad thing, a potentially good thing if they
do it right (what isn't?), or a marketing thing?



i think what he's saying is that they are dropping the fft (fast
fourier transform) for a direct transform, right? the fft does some
mathematical compromises to get the computational load down into a
realistic realm of current computers -- and that they're *really*
doing this not because of the "inaccuracies", but because the
video-optimized GPU isn't flexible to do the mx+b type operations that
audio requires.

however, i understood that convolution was a rather "simple" process
compared to fft -- and only presented more cpu load with longer
impulse response samples . . .



at any rate, 200-800 is completely unrealistic -- it's band-aid
technology to begin with that smacks of vaporware -- the overhead of
audiovideopcigpupcivideoaudio processing alone will eat up
most of the gains -- just buy another P4 box for $500!


cheers,
chris deckard
saint louis mo

(Mike Rivers) wrote in message news:znr1094469385k@trad...
In article writes:

As far as the convolution power, which is what is really
needed today, it says that they will be eschewing the
inaccurate FFT method for the direct one. That will make it
run out of Hz far sooner than the gflop numbers alone would
indicate, as much as 1000 times sooner for a reasonably long
FIR. That is simply a reflection of the lack of flexibility
in reconfiguring or programming these things. If they could
use a partitioned FFT you can damn betcha there would be
_no_ reason not to.

You lost me in the first sentence, because I don't know how these
things work, nor do I feed a need to know. My thing is external
interfaces, trusting that what's inside the "black box" (be it
hardware, software, or soupware) will either do its job or not.

So is this a good thing, a bad thing, a potentially good thing if they
do it right (what isn't?), or a marketing thing?



i think what he's saying is that they are dropping the fft (fast
fourier transform) for a direct transform, right? the fft does some
mathematical compromises to get the computational load down into a
realistic realm of current computers -- and that they're *really*
doing this not because of the "inaccuracies", but because the
video-optimized GPU isn't flexible to do the mx+b type operations that
audio requires.

however, i understood that convolution was a rather "simple" process
compared to fft -- and only presented more cpu load with longer
impulse response samples . . .



at any rate, 200-800 is completely unrealistic -- it's band-aid
technology to begin with that smacks of vaporware -- the overhead of
audiovideopcigpupcivideoaudio processing alone will eat up
most of the gains -- just buy another P4 box for $500!


cheers,
chris deckard
saint louis mo

mr c deckard wrote:


i think what he's saying is that they are dropping the fft (fast
fourier transform) for a direct transform, right?

Close. More specifically they are dropping fast convolution
which employs the FFT for direct convolution.

the fft does some
mathematical compromises to get the computational load down into a
realistic realm of current computers --

The only compromise is latency with a payoff of reducing the
computation from an O(N*N) problem to an O(N*log2(N)) where
N is the length of what you are convolving with the audio
stream. For even a 1024 sample filter IR, it can be 100+ to
1 improvement in throughput done in special hardware. Get
up to say a 16,384 sample reverb (and they can be a lot
longer than that) and the improvement is up to 1170 to 1.
There aren't really any "accuracy issues." These ratios are
actually somewhat smaller because there is more overhead for
a less hardware streamlined FFT that eats into the ratio
some but that's still how fast the difference grows.

And the latency issue is tractable by using a more complex
"partitioned" convolution algorithm which also uses the FFT.

and that they're *really*
doing this not because of the "inaccuracies", but because the
video-optimized GPU isn't flexible to do the mx+b type operations that
audio requires.

Not quite, the b-mx+b is the core operation of the direct
convolution. The kernal operation of the FFT is what is
called a "complex butterfly" using complicated routing as
you go or reordering logic at the end of each block.


however, i understood that convolution was a rather "simple" process
compared to fft -- and only presented more cpu load with longer
impulse response samples . . .

Direct convolution is definitely simpler than fast
convoultion using FFT but the results are identical and it
is only really practical for very short IR's. You run out
of HzPower _really_ fast using the simple direct method.

For the kind of convolutional filtering used in video, the
IR's are very short and are probably down in the range where
the direct form is faster when overhead is taken into
account so there is little motivation to design in the data
paths and control sequences to do an FFT form convolution.


Bob
--

"Things should be described as simply as possible, but no
simpler."

A. Einstein

mr c deckard wrote:


i think what he's saying is that they are dropping the fft (fast
fourier transform) for a direct transform, right?

Close. More specifically they are dropping fast convolution
which employs the FFT for direct convolution.

the fft does some
mathematical compromises to get the computational load down into a
realistic realm of current computers --

The only compromise is latency with a payoff of reducing the
computation from an O(N*N) problem to an O(N*log2(N)) where
N is the length of what you are convolving with the audio
stream. For even a 1024 sample filter IR, it can be 100+ to
1 improvement in throughput done in special hardware. Get
up to say a 16,384 sample reverb (and they can be a lot
longer than that) and the improvement is up to 1170 to 1.
There aren't really any "accuracy issues." These ratios are
actually somewhat smaller because there is more overhead for
a less hardware streamlined FFT that eats into the ratio
some but that's still how fast the difference grows.

And the latency issue is tractable by using a more complex
"partitioned" convolution algorithm which also uses the FFT.

and that they're *really*
doing this not because of the "inaccuracies", but because the
video-optimized GPU isn't flexible to do the mx+b type operations that
audio requires.

Not quite, the b-mx+b is the core operation of the direct
convolution. The kernal operation of the FFT is what is
called a "complex butterfly" using complicated routing as
you go or reordering logic at the end of each block.


however, i understood that convolution was a rather "simple" process
compared to fft -- and only presented more cpu load with longer
impulse response samples . . .

Direct convolution is definitely simpler than fast
convoultion using FFT but the results are identical and it
is only really practical for very short IR's. You run out
of HzPower _really_ fast using the simple direct method.

For the kind of convolutional filtering used in video, the
IR's are very short and are probably down in the range where
the direct form is faster when overhead is taken into
account so there is little motivation to design in the data
paths and control sequences to do an FFT form convolution.


Bob
--

"Things should be described as simply as possible, but no
simpler."

A. Einstein