AI and SDR: Software Meets Hardware Again...
summary
Claims concrete wins: a real Versal ACAP family ships with heterogeneous AI engines + adaptable fabric + Arm scalars, a programmable network-on-chip and on‑chip SRAM that keeps tensors off DRAM, and toolpaths from Vivado HLS to a new Scout frontend plus TensorFlow/PyTorch support. Quantified pressure points: 5G needs 20× peak rates and 100× capacity, ResNet50 costs 7.6B ops and OpenAI notes ~300,000× compute growth versus early CNNs. Built outcomes target beamforming, anti‑jam and SIGINT inference at the edge with better perf/W; launches include AI Core, Prime and an AI Edge edition expected in November.
AI-written summary · about
chapters (5)
- 0:00Start
- 5:34Cognitive Radio
- 9:29Domain Specific Architectures
- 12:26The Adaptive Compute Acceleration Platform
- 12:58Heterogeneous Processing
related talks
14mArm: One Platform for Agentic AI, from Cloud to the Edge
52mPhysical AI: Crafting Resilient Systems with Emotional Intelligence
24mScalable AI Server
42mAI and Machine Learning BoF
50mAI acceleration at the edge
40mFueling the AI Revolution with Gaming
48mThe Power of Edge AI: Revolutionizing Future of Real-Time, Autonomous Processing
22mAdvancing the Human Experience