Keynote: Paving the Road to Datacenter-Scale RISC-V
summary
Google describes a playbook: port 30,000+ packages to multiarch using CI/CD and automation, feed ~40,000 real code edits into Gemini to categorize fixes, and deploy via automated canary rollouts that ramp 1%→4% before scale. Concrete needs: ratified RVA23 server spec, server SoCs with ≥64 cores and ≈4 GB/core, side‑channel‑hardened crypto and MMU for DMA. Encountered real issues—flaky tests, NaN/FP differences, a handful (~3) memory‑ordering bugs—and publishes a preprint and blog while collaborating with RISE and RISC‑V International. CI/CD pull requests AI code edits
AI-written summary · about
chapters (13)
- 0:00Introduction and vision
- 0:26Historical context and ARM
- 1:29Warehouse scale foundations
- 2:29Key ingredients for success
- 4:51Workload migration strategy
- 5:58Automation and scalability
- 7:06Road testing and developer feedback
- 8:20Overcoming technical potholes
- 10:28Ecosystem and collaboration
- 11:12AI-assisted porting journey
- 11:53Applying AI to scale
- 12:40Seamless deployment strategies
- 14:16The road ahead for RISC-V
related talks
23mEmpower Upstream ML Frameworks on RISC-V
09mDriving the Future: Semiconductor Innovation, AI, and the Rise of RISC-V
30mKeynote Panel: The Future of...
45mVerified seL4 on secure RISC-V processors
21mThe RISC-V Vector Cryptography Extensions
17mLLM Inference on RISC-V Embedded CPUs
57mPervasive and Sustainable AI with Adaptive Computing
23mKeynote: gRPC, A Foundation for AI