CUDA

2026-10-29

REGULATION · 1 SOURCE · The Decoder

US antitrust regulators scrutinize Nvidia's Hugging Face deal and sales practices

The US Department of Justice is investigating Nvidia's planned $12.93 billion acquisition of Hugging Face, asking whether the purchase could give Nvidia control over a key hub for distributing AI and harm competitors; Nvidia expects the deal to close in the first half of 2027. Regulators are also probing Nvidia's broader acquisitions, investments and sales practices — including past deals involving SchedMD, Run:ai and a licensing arrangement with Groq — while France's competition authority is separately investigating alleged anticompetitive conduct related to CUDA dependence and Nvidia's cloud stakes such as CoreWeave.

8.0

2026-10-21

CODING · 1 SOURCE · InfoQ AI, ML & Data Engineering

TornadoVM 7.3 released with CUDA hybrid API and jitLLM v2.0

TornadoVM 7.3, an open‑source heterogeneous GPU programming framework for Java, replaces JNI with the Foreign Function & Memory (FFM) API, removes the experimental JVMCI dependency, and centralizes compiler metadata via reflection/ASM/Unsafe. The release deepens Nvidia integration (optimized CUDA backend with cuTile/TileContext and direct cuBLAS/CUTLASS calls), adds Apple MXL support and multi‑GPU execution, and ships jitLLM v2.0 — a single‑JAR Java AI inference engine; both projects are being commercialized through Beestream.ai.

6.0

2026-10-19

COMPANIES · 1 SOURCE · TechCrunch AI

AWS launches TorchNeuron to run PyTorch models on Trainium

Amazon Web Services released TorchNeuron, a PyTorch-compatible tool that AWS says lets developers run PyTorch apps on its Trainium AI chips with minimal configuration (one-line change). AWS also plans to contribute upstream changes to PyTorch to reduce framework tie‑ins to specific chips and argues the move gives startups easier access to AWS cloud and better price‑performance across training and inference.

8.0

2026-10-13

CODING · 1 SOURCE · NVIDIA Developer

NVIDIA adds 'cuDF skills' to help AI agents rewrite Spark UDFs for GPU acceleration

NVIDIA describes a workflow and a new 'cuDF skills for Apache Spark' capability that uses agent-driven tooling to rewrite Spark UDFs into GPU-accelerable forms (via Spark SQL expressions, the RapidsUDF Java interface, or custom CUDA kernels), eliminating GPU fallback and achieving 22x+ speedups in tests. NVIDIA cites a case where a large consumer tech company converted 30+ production UDFs and reduced an operator's runtime from 274.55 to 12.78 hours on a 600‑billion‑row workload.

6.0

2026-09-30

COMPANIES · 1 SOURCE · The Decoder

Deepseek open-sources TileLang and Ascend programming tools for Huawei Ascend 950

Deepseek announced an open-source release of programming libraries and tools — centered on the TileLang language (developed at Peking University) — optimized for Huawei's Ascend chips, including work on a 128‑chip Ascend 950 supernode; Deepseek says Huawei "fully supported" the effort. The package is intended to help build a domestic software ecosystem for Chinese AI chips and presents a simpler programming model the company argues can rival Nvidia's CUDA.

7.0

2026-09-24

RESEARCH · 1 SOURCE · arXiv cs.AI

AI-GRACE: a use-case operationalization framework for agentic AI (arXiv:2609.21192v1)

AI-GRACE is a proposed use-case operationalization framework that connects organizational governance to technical implementation for agentic AI. The paper defines objectives and obligations, assesses risks across seven domains, and derives assurance requirements, runtime controls, and evidence needs, introducing constructs such as an Agent Operating Envelope and Risk-Aligned Independence Levels (RAIL); a fictional retail-banking example illustrates the method and the authors note empirical evaluation is required to confirm deployment benefits.

7.0

2026-09-23

CODING · 1 SOURCE · Hugging Face

NVIDIA demonstrates using Warp and MJWarp to run MuJoCo at GPU scale (up to 2,048 envs)

NVIDIA describes migrating a MuJoCo workflow (an SO-101 follower arm) to MJWarp, an implementation of MuJoCo’s physics pipeline in NVIDIA Warp, and demonstrates advancing as many as 2,048 parallel environments on NVIDIA GPUs. The article explains Warp’s GPU kernel model, device arrays, composable launches and graph capture, and notes Warp features such as differentiable kernels and an opt-in deterministic mode (Warp ≥ 1.15); it focuses on preparing and scaling simulation environments rather than training policies.

6.0

2026-09-22

CODING · 1 SOURCE · NVIDIA Developer

NVIDIA documents TensorRT LLM adaptations for confidential Blackwell (B200) inference

NVIDIA published guidance and measured results showing how TensorRT LLM adapts to confidential computing on Blackwell B200 hardware, including CC-aware memory selection, async token readback, using the GPU %globaltimer for autotuning, and communication-path detection when NVLS multicast is unavailable. In their tests, CC-enabled runs retained about 96.1–98.2% of token throughput versus CC-off, with per-token latency (TPOT) within roughly 1.2–4.3% of baseline.

7.0

2026-09-02

CODING · 1 SOURCE · NVIDIA Developer

The Modern CUDA Toolbox in Practice: A Step-by-Step Optimization Walkthrough

NVIDIA Developer published a step-by-step walkthrough showcasing the modern CUDA toolbox for optimizing GPU-accelerated code. The piece positions CUDA as a foundation for workloads from scientific simulations to large-scale AI training and demonstrates practical optimization guidance.

6.0

2026-08-25

CODING · 1 SOURCE · NVIDIA Developer

CUDA Python 1.0 — stable APIs, one foundation, full platform access

NVIDIA has released CUDA Python 1.0, presenting stable APIs and a unified foundation intended to give Python developers fuller access to CUDA and GPU capabilities without requiring C++ extension toolchains. The release aims to simplify building GPU-accelerated Python applications across the NVIDIA platform.

7.0

2025-09-29

MODELS · 1 SOURCE · DeepSeek

DeepSeek launches experimental model V3.2-Exp with sparse attention and big API price cuts

DeepSeek introduced DeepSeek-V3.2-Exp, an experimental model built on V3.1-Terminus that debuts DeepSeek Sparse Attention (DSA) to improve long-context efficiency with minimal quality loss; published benchmarks show parity with V3.1-Terminus. DeepSeek also cut DeepSeek API prices by over 50% effective immediately, provides TileLang and CUDA GPU kernels, and keeps V3.1-Terminus available via a temporary API until Oct 15, 2025, 15:59 UTC for comparison testing.

7.0