CUDA

Evidence (9)

• 8The New Stack·8/10/2026breakthrough
  EKS speeds up multi-gigabyte container image pulls

  The article describes how large ML inference container images on Amazon EKS were taking several minutes to pull, delaying pod readiness and leaving GPUs idle. By reworking the image pull pipeline to better use available network, storage, and compute resources, the team reduced multi-gigabyte pulls to seconds; the improvements are now available by default in EKS Auto Mode and were contributed upstream to containerd and the SOCI snapshotter.

• 8NVIDIA GenAI·7/22/2026open_source
  NVIDIA Open Sources GPU-Accelerated Medical Physics Simulation Framework

  NVIDIA has open sourced Medical Physics Simulation, a GPU-accelerated framework within Isaac for Healthcare for simulating anatomy-device interactions, sensor inputs, and difficult edge-case scenarios. The framework is designed to help medical robotics teams train, test, and evaluate policies in simulation before moving to hardware-heavy validation.

• 7Hacker News·7/14/2026product_launch
  Spectral Compute Pushes a CUDA-Free Path for Non-NVIDIA GPUs

  The article covers Spectral Compute’s effort to make CUDA-style GPU programming work on non-NVIDIA hardware, potentially reducing dependence on NVIDIA’s software stack. The Hacker News discussion reflects interest in whether a credible alternative can emerge for developers and HPC users who want broader hardware compatibility.

• 7Hacker News·6/30/2026open_source
  ZLUDA 6 adds PhysX, Blender texture support, and better Windows compatibility

  ZLUDA has released version 6, a new major update for its project that enables unmodified CUDA applications to run on non-NVIDIA GPUs. The update adds pre-alpha 32-bit PhysX support, basic texture support that also enables Blender, and improved Windows handling for performance libraries and loaders.

• 4Hacker News·6/29/2026research
  A Deep Dive Into What Happens When a CUDA Kernel Runs

  This article traces a simple CUDA vector-add program from source code through compilation and GPU execution, explaining how nvcc, PTX, and SASS fit into the pipeline. It also describes low-level runtime details such as warps, device memory access, and the mechanisms used to launch and complete the kernel on an RTX 4090.

• 8NVIDIA GenAI·6/2/2026product_launch
  NVIDIA Adds Agentic AI Support to Jetson Edge Platform

  NVIDIA announced JetPack 7.2 and NemoClaw support for Jetson at COMPUTEX, bringing agentic AI capabilities to edge devices. The update adds Yocto support, CUDA 13 on Jetson Orin, MIG on Jetson Thor, and performance gains for Jetson AGX Orin 32GB, while enabling deployment of agentic workflows for robotics, inspection, and industrial automation.

• 4Hugging Face Blog·5/29/2026framework_update
  A Beginner’s Guide to torch.profiler in PyTorch

  Hugging Face Blog introduces the first part of a series on profiling in PyTorch, focused on helping beginners read profiler traces and understand how Python calls map to CPU and CUDA activity. The article explains how to use torch.profiler, interpret its tables and traces, and observe what changes when torch.compile is applied.

• 7Hugging Face Blog·2/13/2026open_source
  Claude and Codex Gain CUDA Kernel Writing Skill

  Hugging Face introduced an agent skill that teaches coding agents how to write production CUDA kernels and integrate them with PyTorch-based projects. The blog says Claude and Codex were used to generate working kernels for a diffusers pipeline and a transformers model, including bindings and benchmarks end to end.

• 6Hugging Face Blog·1/28/2026research
  Hugging Face shows how Claude can teach smaller models CUDA kernel writing

  Hugging Face describes a workflow using Claude Opus 4.5 and a new tool called upskill to turn an agent trace into a reusable skill for specialized tasks like writing CUDA kernels. The post argues that this approach can transfer hard domain knowledge from a stronger model to smaller or cheaper models, while also measuring whether the skill actually improves performance.