The war for the AI driver stack is just beginning. Stay tuned.
The first stable release, committing to semantic versioning. Includes new "green contexts" allowing partitioning of GPU SMs to shield latency‑sensitive kernels from long‑running workloads, and process checkpointing to snapshot the full CUDA state of a running process. cuda driver release news exclusive
Workload Type | Performance Increase vs. Previous Driver ----------------------------------------------------------------------- LLM Training (Multi-Node) | ▲ 28% Faster Throughput Generative AI Inference | ▼ 35% Latency Reduction Molecular Dynamics | ▲ 22% Higher Simulation Steps/Sec Real-Time Path Tracing | ▲ 18% Frame Rate Stability 🤖 LLM and Generative AI Enhancements The war for the AI driver stack is just beginning
Our exclusive CUDA driver release news pipeline continues. We have seen early staging branches of the R560 driver, which contains a flag called --kernel-mode-only . This suggests NVIDIA is preparing a driver that can run entirely in user space, bypassing the OS kernel entirely for AI workloads—a "micro-driver" to fight back against AMD’s ROCm and Intel’s SYCL. Includes new "green contexts" allowing partitioning of GPU
Driven primarily by the Windows Subsystem for Linux (WSL 2), Windows driver updates target developers working on local workstations. A major breakthrough in recent driver lineages is the radical optimization of memory virtualization, allowing local developers to prototype models that technically exceed the physical VRAM limits of a single desktop GPU by gracefully paging assets to system RAM.
Our sources inside three independent AI hardware labs have confirmed that the R570.100 driver branch is not incremental. It is foundational. While the public-facing changelog will mention “stability improvements and new GPU support,” the private developer preview tells a different story.