In the news
Control How Your GPU Shares Work with Green Contexts
NVIDIA Developer · Published · 3 min read
In 30 seconds
- What happened
- CUDA 13.1 adds green contexts, letting applications partition GPU resources like SMs and workqueues within a single process for explicit control.
- Why it matters
- Matters for engineers running multiple concurrent workloads on one GPU who need predictable latency or better resource isolation than stream priorities alone provide.
- Watch out
- Green contexts require explicit API calls and resource planning; they trade total throughput for latency guarantees by dedicating SMs to specific workloads.
- kernel
- inference
- throughput
- latency
The patterns behind this
- Agentic Context Engineering (Evolving Playbook)
- Concurrent Orchestration
- Context Engineering Frameworks
Each one covers how the technique works, when it earns its cost, and where it breaks.
The Agent Architect
One pattern, one tradeoff, one production failure story. A short weekly briefing for people building agentic systems.
Weekly email, one-click unsubscribe. We only use your address to send the briefing.