Enhancing Transformers: Accelerating Nonlinear Function Computation On RISC-V Vector Processor

on YouTube ↗24m

" summary

Describes a method and custom RISC‑V vector extension instructions for faster exponential function evaluation used inside nonlinearities in transformer models (GeLU, Softmax). Presents an algorithmic approximation with tunable precision parameters plus a hardware microarchitecture implementation (about 95k gates per fp32 exp) and reports speedups versus the standard RISC‑V vector extension (≈4.44× for GeLU, ≈10.7× for Softmax) supporting FP16/BF16/FP32 data types.

AI-written summary · about

" related talks

• 24mMoving to RISC-V Vector: A Practical Journey of AI Operator OptimizationGuodong Xu · RISC-V Summit North America 2025
• 28mAn Efficient Implementation of TensorFlow Lite for RISC-V VectorsMostafa Hagog · RISC-V Forum: Vector & Machine Learning 2021
• 13mUnleashing ML Processing Power Through RISC-V Vectors: Applications, Implementation and...B. Barker · RISC-V Summit North America 2025
• 11mDemo: Running Transformers on Semidynamic's "All-In-One" Vector and Tensor UnitRoger Espasa · RISC-V Summit North America 2024
• 18mLessons Learned in Using RISC-V for Generative AI and Where We Can Go from HereJayesh Iyer, Josep M Perez · RISC-V Summit North America 2024
• 10mOptimize Openblas by RISC-V "V" Vector ExtensionXianyi Zhang · RISC-V Global Forum 2020
• 20mBoosting Video Codec With RISC-V Vector ExtensionJing Qiu, Jiayan Qian · RISC-V Summit North America 2025
• 34mRISC V Vector Extension ProposalRoger Espasa · 7th RISC-V Workshop 2017