01 What happened
Google Cloud has made Confidential Computing on G4 instances with NVIDIA RTX PRO 6000 Blackwell Server GPUs generally available, offering encrypted AI inference across 36 zones.
02 Key details
- Confidential G4 instances with NVIDIA Blackwell GPUs are GA on Google Cloud, covering 36 zones.
- Benchmarking shows less than 3.5% throughput degradation for Llama 3, Llama 3.3, Resnet-50, SDXL.
- No code changes required; uses standard NVIDIA drivers and CUDA-X libraries.
- Customer Vertebrae reported 5.2% latency increase with hardware-level privacy assurance.
- Confidential Space for secure multi-party data sharing is rolling out to general availability in the coming weeks.
03 Why it matters
Organizations can protect model weights and user data during computation with verifiable hardware isolation, meeting compliance without significant performance trade-offs.
04 Who it matters to
Enterprises and AI developers.
Original source