FPGA DSP Speed at Low $/W
Versal/Zynq DSP speed at a fraction of the cost & power
10s of mm2 of InferX in N7/N5/N3 exceeds Zynq/Versal DSP performance at a fraction of the cost/power.
Each InferX tile can do 2 TeraMACs/second (N5, INT16 inputs, INT40 accumulation).
InferX is a scalable architecture built on the InferX tile which form arrays: N tiles run N times faster.
InferX is reconfigurable in microseconds so you can configure for different workloads or process multiple workloads in one chip.
InferX can also run AI Neural Network models: simultaneous or alternating.
InferX does very high performance streaming DSP.
We do Complex INT16 FFT of various sizes, switchable instantly with no symbol loss.
FIR INT16 real and complex are supported with any number of taps.
Matrix Inversions of real or complex of varying sizes run efficiently on InferX.
Also, IIR, Beam Forming, Matrix/Vector operations, Kalman functions and more.
Multiple DSP streams can run independently simultaneously OR chained.
More benchmarks and PPA under NDA.