📁 Hardware

Rumors suggest Nvidia is developing the RTX 5050 with 9GB of GDDR7 VRAM on a 96-bit bus, alongside an RTX 5060 with a cut-down GB205 GPU. The new cards could target entry-level market segments, offering a more accessible option for consumers.

Despite initial issues with open-source drivers, Intel's GMA500 driver, created to support PowerVR SGX graphics hardware (code-named Poulsbo), continues to receive updates in the Linux kernel, almost twenty years after its introduction. An example of persistence in the open-source world.

An international team has created a high-resolution 3D atlas of ant morphology, called Antscan. Using a particle accelerator, researchers digitized 792 ant species, making detailed 3D models of exoskeletons, muscles, and internal organs accessible online. This resource opens new perspectives for scientific research, education, and engineering.

Micron has announced SOCAMM2, a new 256GB memory module designed for AI servers. The new technology allows scaling memory up to 2TB per CPU, enhancing the performance of artificial intelligence applications. This solution is particularly relevant for workloads requiring high memory capacity and bandwidth.

Broadcom will soon deploy multiple gigawatts worth of custom accelerators at Meta, OpenAI, and Anthropic. The company argues this shows that AI companies and hyperscalers can’t successfully develop and deploy their own silicio any time soon.

Broadcom and TSMC are collaborating on chips for artificial intelligence applications, leveraging 3.5D integration. This strategic move could position Broadcom as a direct competitor to Nvidia in the high-performance ASIC (Application-Specific Integrated Circuit) market, offering new options for AI workloads.

Intel is preparing support for the Xe3P graphics architecture in the Mesa OpenGL "Iris" and Vulkan "ANV" open-source drivers. This follows the enablement of Xe3P in the mainline Linux kernel for upcoming Nova Lake integrated graphics and the Crescent Island AI inference accelerator.

The Linux Foundation introduced the OCUDU Ecosystem Foundation at Mobile World Congress (MWC). This initiative aims to advance open-source AI-RAN (Radio Access Network) innovation for 5G and early 6G network solutions. Early performance tests on AMD EPYC and Intel Xeon server platforms indicate an initial lead for AMD EPYC.

The new Nvidia driver 595.71 appears to introduce overclocking limitations on some GeForce graphics cards, particularly the RTX 40 and 50 series. Not all GPUs are affected, but the driver release seems problematic for those aiming to maximize hardware performance.

Nvidia has entered into multi-year agreements with Lumentum and Coherent, marking a significant step forward in the commercialization of silicio photonics. These collaborations could optimize high-speed interconnects, crucial for future accelerated computing architectures and artificial intelligence workloads.

The US Department of Defense is reverse-engineering Iranian-made Shahed drones. The goal is to better understand asymmetric warfare tactics and develop countermeasures, particularly through the use of AI-controlled drone swarms within the LUCAS project.

Nvidia and MediaTek are investing in new optics technologies for AI data centers. These investments aim to improve the performance and energy efficiency of the computing infrastructures required for training and inference of artificial intelligence models. The evolution of optical interconnects is crucial to manage the increasing bandwidth demand in modern data centers.

Apple has announced the new M5 Pro and M5 Max chips, based on a new Fusion architecture. The new processors aim to improve performance in the field of artificial intelligence and machine learning, integrating specific optimizations for these workloads.

Seagate has begun shipping 44TB hard drives based on the Mozaic 4+ platform with HAMR (Heat-Assisted Magnetic Recording) technology, targeting data centers. The new platform expands capacity up to 10 platters per drive.

Linux benchmarks on Intel's new Xe3 B390 GPUs (Panther Lake architecture) show improvements in OpenGL, Vulkan, and OpenCL performance compared to previous generations. Performance analysis using Intel Rendering Toolkit and OpenVINO for AI workloads, comparing results with Lunar Lake and Meteor Lake.

Rumors suggest the Nvidia RTX 5070 mobile GPU might get a VRAM boost, from 8GB to 12GB. This news comes amid an ongoing global memory crisis, raising questions about Nvidia's strategies for the mobile sector and the availability of its Blackwell chips.

Apple announced the new Fusion Architecture with the M5 Pro and M5 Max SoCs, featuring a next-generation GPU. This architecture promises significant improvements in graphics performance, opening new possibilities for professional applications and gaming.

Intel has sent a "drm-xe-next" pull request to DRM-Next for the Linux 7.1 kernel, containing improvements to the Xe graphics driver. These updates are intended for integration into the DRM-Next branch and aim to enhance the performance and stability of Intel's graphics drivers.

KDDI and AVITA are collaborating to develop AI humanoids for customer service, combining physical interaction with artificial intelligence. The initiative aims to address operational gaps due to workforce reduction, integrating advanced avatars with high-capacity communication infrastructures. Initial trials are planned for autumn 2026.

Microsoft introduces Shader Execution Reordering (SER) in the latest DirectX SDK, enhancing ray tracing efficiency. Intel Arc B-series GPUs demonstrate up to 90% performance uplift. This optimization is part of DirectX 12 Ultimate.