Cash Flow Over Ten Years to $389B… 113 *Note: Big Six USA technology companies include Apple, Nvidia, Microsoft, Alphabet / Google, Amazon, & Meta Platforms / Facebook. FCF calculated as cash flow from Amazon Meta 2014 2019 2024 Free Cash Flow, $B Tech Big Six (USA) = Loaded With Cash to Spend on AI & CapEx Big Six* Public Technology Companies – Free Cash Flow ($B) – 2014-2024, per Capital IQ…Big Six* with market capitalizations in excess of $1 trillion – most with gross margins of +50% plus free cash flow – attacking the same opportunity at the same time in a relatively transparent world, adding in high

FPGA ASIC Optimization AutoTVM Device FleetExisting Deep Learning Frameworks High-level data flow graph Hardware Primitive Tensor operators such as Conv2D eg. cuDNN Offload to heavily optimized intensiveMachine Learning based Program Optimizer TVM: Learning-based Learning System High-level data flow graph and optimizations Directly generate optimized program for new operator workloads and hardware of New IR Infra Single unified module/pass, type system, with function variants supportCompilation Flow under the New Infra IRModule (relay::Function) IRModule (te::Function, ExternFunc, …) runtime::Module

Control flow (if, loop, etc) ● Dynamic shapes ○ Dynamic inputs: batch size, image size, sequence length, etc. ○ Output shape of some ops are data dependent: arange, nms, etc. ○ Control flow: concatenate

WISE ... Efficiency > 50% for mainstream neural networks >> 4© Copyright 2018 Xilinx Inference Flow >> 5 MxNet CPU Layers FPGA Layers Runtime Image Model Weights Calibration Set Quantizer Compiler

for runtime dispatching ● Codegen path: src/relay/backend/contrib//codegen.cc ● Flow overview data weight1 weight3 weight2 output Build() Your graph representation (e.g., JSON