FFT core

G

Grumps

Guest
[posted to comp.arch.fpga + comp.lang.vhdl]
Hi All
I've had a quote from a 3rd party to develop a floating point FFT core for
us (1Mpt). Probably for a Xilinx Virtex5 SXT.
Obviously I'd like to get some more quotes, but would like to know if you
have any recommendations?
Thanks.
 
Grumps wrote:
Hi All
I've had a quote from a 3rd party to develop a floating point FFT core for
us (1Mpt). Probably for a Xilinx Virtex5 SXT.
Obviously I'd like to get some more quotes, but would like to know if you
have any recommendations?
Click to expand...
Dillon Engineering has a fact sheet about a floating point core on their
web site:

http://www.dilloneng.com/documents/fpfft_fact.pdf

Maybe that would give you an alternative to your quote?

Cheers,

Guenter
 
Guenter Dannoritzer wrote:
Grumps wrote:
[posted to comp.arch.fpga + comp.lang.vhdl]
Hi All
I've had a quote from a 3rd party to develop a floating point FFT
core for us (1Mpt). Probably for a Xilinx Virtex5 SXT.
Obviously I'd like to get some more quotes, but would like to know
if you have any recommendations?

Dillon Engineering has a fact sheet about a floating point core on
their web site:

http://www.dilloneng.com/documents/fpfft_fact.pdf

Maybe that would give you an alternative to your quote?
Click to expand...
Thanks. I'll see what they have to say.
 
Grumps wrote:
Hi All
I've had a quote from a 3rd party to develop a floating point FFT core for
us (1Mpt). Probably for a Xilinx Virtex5 SXT.
Obviously I'd like to get some more quotes, but would like to know if you
have any recommendations?
Thanks.
Click to expand...
I've got a floating point FFT engine for V4 that I am porting to V5. It
is the fastest floating point FFT for FPGAs available anywhere (up to
1.2 GS for the 32-2K point FFT). It can be adapted for 1M points and
will still beat anything out there for speed/density. There is info on
my website http://www.andraka.com regarding my IP, including a paper
discussing it.
 
"Ray Andraka" <ray@andraka.com> wrote in message
news:sEbNi.9606$Bq3.5856@newsfe18.lga...
Grumps wrote:
[posted to comp.arch.fpga + comp.lang.vhdl]
Hi All
I've had a quote from a 3rd party to develop a floating point FFT core
for us (1Mpt). Probably for a Xilinx Virtex5 SXT.
Obviously I'd like to get some more quotes, but would like to know if you
have any recommendations?
Thanks.

I've got a floating point FFT engine for V4 that I am porting to V5. It
is the fastest floating point FFT for FPGAs available anywhere (up to 1.2
GS for the 32-2K point FFT). It can be adapted for 1M points and will
still beat anything out there for speed/density. There is info on my
website http://www.andraka.com regarding my IP, including a paper
discussing it.
Click to expand...
Interesting reading.
You have mail!
 
Grumps wrote:
[posted to comp.arch.fpga + comp.lang.vhdl]
Hi All
I've had a quote from a 3rd party to develop a floating point FFT core
for us (1Mpt). Probably for a Xilinx Virtex5 SXT.
Obviously I'd like to get some more quotes, but would like to know if you
have any recommendations?
Thanks.
Click to expand...
4DSP offers a floating point FFT and it seems like the only one
available today with true IEEE-754 (float) arithmetic for FPGA
devices. The same Commercial Off The Shelf core can be used for FFT
ranging from 256 points to 1M points in Virtex-4 and Virtex-5 FPGA. No
need for expensive redesign! Documentation and a bit true model are
available on 4DSP's website: http://www.4dsp.com/fft.htm

Cheers
Pierrick
 
"Pierrick" <pierrickv@4dsp.com> wrote in message
news:1191953351.407840.18230@22g2000hsm.googlegroups.com...
Grumps wrote:
[posted to comp.arch.fpga + comp.lang.vhdl]
Hi All
I've had a quote from a 3rd party to develop a floating point FFT core
for us (1Mpt). Probably for a Xilinx Virtex5 SXT.
Obviously I'd like to get some more quotes, but would like to know if
you
have any recommendations?
Thanks.

4DSP offers a floating point FFT and it seems like the only one
available today with true IEEE-754 (float) arithmetic for FPGA
devices. The same Commercial Off The Shelf core can be used for FFT
ranging from 256 points to 1M points in Virtex-4 and Virtex-5 FPGA. No
need for expensive redesign! Documentation and a bit true model are
available on 4DSP's website: http://www.4dsp.com/fft.htm
Click to expand...
Thanks. Have you used this core before?
 
Pierrick wrote:

4DSP offers a floating point FFT and it seems like the only one
available today with true IEEE-754 (float) arithmetic for FPGA
devices.
Click to expand...
That is a bold statement and I would be interested which other available
cores you compared yours to and ruled out that they do not support true
IEEE-754 arithmetic?

Cheers,

Guenter
 
On Oct 9, 12:53 pm, Guenter Dannoritzer <kratfkryk...@spammotel.com>
wrote:
Pierrick wrote:
4DSP offers a floating point FFT and it seems like the only one
available today with true IEEE-754 (float) arithmetic for FPGA
devices.

That is a bold statement and I would be interested which other available
cores you compared yours to and ruled out that they do not support true
IEEE-754 arithmetic?

Cheers,

Guenter
Click to expand...
Fair enough! The message however is related to the original thread
where the user is looking for a 1M points FFT. It looks indeed as if
there are some alternatives for shorter lengths FFTs (<32k points).
 
Pierrick wrote:
On Oct 9, 12:53 pm, Guenter Dannoritzer <kratfkryk...@spammotel.com
wrote:
Pierrick wrote:
4DSP offers a floating point FFT and it seems like the only one
available today with true IEEE-754 (float) arithmetic for FPGA
devices.
That is a bold statement and I would be interested which other available
cores you compared yours to and ruled out that they do not support true
IEEE-754 arithmetic?

Cheers,

Guenter

Fair enough! The message however is related to the original thread
where the user is looking for a 1M points FFT. It looks indeed as if
there are some alternatives for shorter lengths FFTs (<32k points).
Click to expand...
I see your point, but don't agree with that statement. When I do a
google search for floating point FFT, the first five hits show
information about floating point FFT cores from:

- 4DSP
- Dillon Engineering
- Andraka Consulting
- Sundance Multiprocessor Technology Ltd
- Altera

From all the above only on the web page of Andraka Consulting it says
that their core is limited to 2048 points.

The other hits either explicit say their core supports lengths up to 1M
or do not state any length limitations.

And they all state that they are IEEE-754 compliant.
 
There is a difference between what is available today for 1M points
(COTS that can be shipped today) and what can be designed and be
available in 6 months time.
Dillon, as a consulting company you claim you can do it. Maybe you
have it as a product but there are no figures showing number of
slices, transform time, etc... So it is questionable. Since there is
no limitation in the length of your FFT length, can you supply a 512M
points FFT today?

4DSP, as a product company we have the 1M points COTS and it can ship
it today. However, I can confirm we do not have a 512M points FFT
today. But we can do it if given enough time:)
 
Guenter Dannoritzer wrote:
Pierrick wrote:

On Oct 9, 12:53 pm, Guenter Dannoritzer <kratfkryk...@spammotel.com
wrote:

Pierrick wrote:

4DSP offers a floating point FFT and it seems like the only one
available today with true IEEE-754 (float) arithmetic for FPGA
devices.

That is a bold statement and I would be interested which other available
cores you compared yours to and ruled out that they do not support true
IEEE-754 arithmetic?

Cheers,

Guenter

Fair enough! The message however is related to the original thread
where the user is looking for a 1M points FFT. It looks indeed as if
there are some alternatives for shorter lengths FFTs (<32k points).


I see your point, but don't agree with that statement. When I do a
google search for floating point FFT, the first five hits show
information about floating point FFT cores from:

- 4DSP
- Dillon Engineering
- Andraka Consulting
- Sundance Multiprocessor Technology Ltd
- Altera

From all the above only on the web page of Andraka Consulting it says
that their core is limited to 2048 points.

The other hits either explicit say their core supports lengths up to 1M
or do not state any length limitations.

And they all state that they are IEEE-754 compliant.
Click to expand...

While my core computes up to a 2048 point size, it can be used as a
building block for much larger (up to 4M point in two passes) FFTs with
external memory. No FPGA has sufficient internal memory to support
these larger FFTs without going off-chip.
 
You must log in or register to reply here.

Welcome to EDABoard.com

Sponsor

Back
Top