Forum Post: Exhausting the C66x DSP core - what DSPLIB example is the toughest?

Hi,

For customers to benchmark max power consumption hands-on for C66x devices (to plan for cooling HW), please advice what DSPLIB math function that could generate the most heat when run from L2RAM?

The DSPLIB in MCSDKv3 comes with a single FFT example where some studies in the simulator reveils the following stats for 'core efficiency':

            /* Call the FFT routines */
   DSP_fft16x16(w_16x16, N, x_16x16, y_16x16); // cycle.CPU=778  CPU.stall.summary=286  Load=73%
   DSP_fft16x32(w_16x32, N, x_16x32, y_16x32); // cycle.CPU=1514  CPU.stall.summary=439  Load=77%
   DSP_fft32x32(w_32x32, N, x_32x32, y_32x32); // cycle.CPU=1493  CPU.stall.summary=602  Load=71%

Q1 - Are there any better algorithms that for instance uses more of the .M units in parallel than the one above? 
Q2 - Will the 32x32 generate more heat (.M unit works harder?) than the 16x32 even if it shows more stalls?

Grateful for your advice.

/Magnus