The acceleration race extends from GPUs to NPUs, photonics and edge/space systems, with new drivers, runtimes and production milestones for AI infrastructure.
NVIDIA has released CUDA 13.5 together with the Linux driver 620.70.06. The central change is general availability support for Vera Rubin, the NVIDIA GPU architecture. For teams working on self-hosted infrastructure and LLMs, the toolchain becomes av...
An open source project called npu-driver is trying to move Windows 11 desktop composition from the GPU to the NPU. It currently runs at 2–3 Hz but targets 30 Hz. This is not a performance win: it shows that NPUs can step beyond their AI inference rol...
Intel has released version 26.40.40196.8 of its open-source Compute Runtime for Windows and Linux. The release prepares support for future D Series GPUs and OpenCL LEO extensions targeting Nova Lake P. More than a maintenance update, it signals that ...
An open-source USB tool removes the software limit that cuts AI performance of Nvidia GPUs on Windows to half speed. The news reopens the debate on artificial segmentation of consumer hardware and on the choices of those evaluating on-prem deployment...
As first production units ship, NVIDIA revises some specifications for its Rubin GPU platforms. The timing signals a shift from validation to supply and forces AI infrastructure planners to reassess integration, power, and cooling assumptions.
Google's first orbital data center satellite is operational: four TPUs on board, up to 1 kW of power generation, but power limits allow only one accelerator to be switched on for about 20 minutes. The news, picked up ironically on Reddit, highlights ...
NVIDIA has released CUDA Oxide 0.3, an update to its experimental Rust-to-CUDA compiler. The project aims to let developers write GPU kernels in Rust within the CUDA ecosystem. It opens an alternative path to C++/CUDA for on-prem and self-hosted stac...
Google has brought online its first data center satellite, carrying the first TPU in space and running a Google DeepMind Gemma 4 model that writes cheesy poetry. The experiment shows LLM inference can operate in a remote, constrained, self-contained ...
PsiQuantum has built an 8x8 optical switch prototype for AI data center GPU networks. The company draws on its photonics expertise and reports lower light loss than rival solutions, with plans to develop larger devices. The project highlights the gro...
The first Loom SDK v0.1 from AMD marks the start of the Hyperloom roadmap announced at Advancing AI, aiming to optimize agentic workloads across CPUs, GPUs and NPUs. For local deployment, the signal is twofold: AMD wants to unify the ROCm ecosystem, ...
