Hardware acceleration for unikernels
summary
Examines using unikernels (rumprun) to reduce virtualization overhead when offloading work to specialized hardware for ML inference. Describes porting the vAccel hardware-acceleration framework to rumprun and QEMU/KVM, including frontend/backend driver changes and the runtime support needed to execute inference on accelerators, and presents a demo of two unikernel setups running image ML inference.
AI-written summary · about
related talks
24mUniversal Tools for Acceleration, Timing, Integration & Machine Enhancement
10mDemo: Nested Virtualization Acceleration
41mQEMU: Emulating Your Way to Success
32mDeveloping the RISC-V Hypervisor Extensions in QEMU
40mHands-on TinyML Deployment on RISC-V: Building an Optimized Edge AI Pipeline - Corrected Audio
18mZVM: Zephyr-based Virtual Machine
34mTailor-Made Security: Building a Kubernetes Specific Hypervisor
22mPreview: Hardware and Offloads