cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 
Highlighted
Visitor
Visitor
388 Views
Registered: ‎02-27-2019

Multiplier IP V12.0 documentation not aligned with switching characteristics

I am looking at the best performance that I can get for a 18x18 signed multiplier on Kintex-1 device.

On that documentation [ https://www.xilinx.com/support/documentation/ip_documentation/ru/mult-gen.html ] Xilinx claims that the IP should run at 866Mhz:

xcku13pffve900-1kup_1_dsp_18x18Parallel_MultiplierSigned18Signed18  Use_MultsSpeed 3503CLK86600100PRODUCTION 1.23 03-18-2019

 

But on the kintex datasheet [ https://www.xilinx.com/support/documentation/data_sheets/ds922-kintex-ultrascale-plus.pdf  page 35 ] it seems to me that the DSP cannot run faster than 645Mhz

kintex.png

 

What value should I consider and why?

Thanks a lot

 

0 Kudos
2 Replies
Highlighted
Mentor
Mentor
350 Views
Registered: ‎02-24-2014

It's true that they don't match,  but even the lower figure of 645 MHz will be insanely difficult to reach in a real design.   The purpose of the high speed on the DSP48 is not to run at those speeds, but to NOT be the bottleneck for larger designs that use the DSP48.     You will find that in real world designs, the DSP48 speed doesn't matter, as long as there is enough pipelining enabled.    Of course,  in IIR filter designs, the DSP48 is very likely to be a bottleneck, because of the feedback path.    

Don't forget to close a thread when possible by accepting a post as a solution.
0 Kudos
Highlighted
Visitor
Visitor
340 Views
Registered: ‎02-27-2019

 I am interrested in simple FIR designs.

If the true speed of the DSP is 866Mhz I can hope to have a FIR working at 600Mhz

If the true speed of the DSP is 645Mhz then it will never work at 600 Mhz

That is why I ask the question

0 Kudos