This Hardware archive tracks the practical side of local AI infrastructure: GPUs, NPUs, mini PCs, edge accelerators, memory bandwidth, and power efficiency tradeoffs that directly impact LLM inference quality. We prioritize benchmark-backed updates and deployment notes useful for real build decisions, from compact home labs to enterprise pilot clusters. Use this stream to compare total cost of ownership, thermal constraints, and model-fit scenarios across current devices, then deepen with our hardware pillar guide and connected LLM coverage.
Meta has developed four new MTIA (Meta Training and Inference Accelerator) processors to power its artificial intelligence and recommendation systems. This strategic move aims to reduce reliance on external vendors like Nvidia, while continuing to invest in cutting-edge hardware.
Intel has released new Linux kernel graphics driver patches. The goal is to enable Adaptive Sync SDP (Secondary Data Packet) handling for Panel Replay and Auxless Adaptive Link Power Management (ALPM) modes.
Brooklyn-based Looking Glass is launching Musubi, an AI-powered holographic frame. After nearly a decade of developing 3D screens, the company aims to bring photos and videos to life with a three-dimensional visual experience.
Broadcom has announced the Taurus BCM83640, a new 3nm optical digital signal processor (DSP) designed for next-generation AI networks. This 400G per lane PAM4 chip promises to improve the density and speed of communications in artificial intelligence applications.
Cortical Labs aims to solve the AI energy crisis with biological computers. The 'body-in-the-box' CL1 technology promises lower energy consumption than a calculator, opening new frontiers in high-efficiency computing.
MediaTek introduces the new Genio platform, a 3nm system-on-chip (SoC) designed for high-end AIoT applications requiring advanced image processing. The platform aims to meet the growing demands for computing power and energy efficiency in the sector.
The collaboration between Qualcomm and Wayve aims to simplify the integration of physical AI in vehicles. The goal is to provide production-ready advanced driver assistance systems (ADAS), combining Wayve's AI driving layer with Qualcomm's Snapdragon Ride system-on-chips. This partnership aims to reduce development costs, complexity, and project risks for manufacturers.
Demand for Mac minis in China temporarily surged due to the popularity of OpenClaw AI agents, with costs exceeding CNY10,000. This phenomenon highlights the growing interest in affordable hardware solutions for artificial intelligence applications.
Advanced Semiconductor Engineering (ASE) has broken ground on a new site in Southern Taiwan dedicated to advanced packaging for artificial intelligence applications. Completion is scheduled for the second quarter of 2028. The investment aims to meet the growing demand for high-performance packaging solutions in the AI sector.
Nvidia is reportedly evaluating Samsung's HBM4 packaging capabilities for its future Rubin GPUs. This assessment is crucial to ensure that the next generation of high-performance GPUs can fully leverage advanced HBM memory, which is essential for artificial intelligence and high-performance computing workloads.
Applied Materials partners with Micron and SK Hynix on a US$5 billion research and development hub focused on memory for artificial intelligence applications. The initiative aims to improve the performance and efficiency of memory used in AI workloads.
According to DIGITIMES, Nvidia is streamlining its supply chain for PCBs (Printed Circuit Boards) used in the CoWoP (Chip-on-Wafer-on-Substrate) advanced packaging process. The company will focus on three main partners to address increasing demand and production capacity challenges.
A proper cooling system is essential to maintain optimal PC performance, especially with intensive workloads such as large language model inference. The article analyzes how to balance air pressure inside the case to maximize cooling efficiency and minimize dust buildup.
The partnership between ABB and NVIDIA aims to bridge the gap between digital models and real production environments through RobotStudio HyperReality. Physical simulation based on NVIDIA Omniverse promises to reduce deployment costs by up to 40% and accelerate time-to-market by 50%. Foxconn and Workr are already testing the solution.
Newegg offers a bundle including 32GB of Corsair Vengeance DDR5 RAM, the new AMD 9850X3D processor, and an X870E-E motherboard, at a significant discount. An opportunity to upgrade the heart of your AM5 PC.
RADV Radeon Vulkan driver developers on Valve's Linux graphics team are evaluating the idea of greater use of per-game/app profiles within this open-source driver and for Mesa drivers at large. The consideration now is extending the current DriConf capabilities to allow for more per-game optimizations.
The RPCS3 emulator achieves remarkable performance, with Minecraft exceeding 1500 FPS. This result highlights the progress in Playstation 3 emulation, with one frame rendered every 0.00064 seconds.
SK Hynix has announced the development of the 1c LPDDR6 DRAM, specifically designed for devices leveraging artificial intelligence. This new generation of memory promises to improve performance and energy efficiency in AI devices, opening new possibilities for advanced applications.
According to a report from Korea, Nvidia is reportedly planning to restart production of the RTX 3060 GPU, relying on Samsung. The decision is said to be motivated by restrictions imposed by China.
Ayar Labs CEO Mark Wade highlights the challenges in scaling silicio photonics production. The company aims to overcome these obstacles to meet the growing demand for high-bandwidth solutions in data centers and artificial intelligence.