The Intel QuickAssist (QAT) driver for the Linux 7.1 kernel introduces support for Zstandard (Zstd) compression and decompression offloading. This integration extends hardware acceleration to QuickAssist Gen 4, Gen 5, and Gen 6 for compression, while limiting decompression to Gen 6. The update is crucial for optimizing performance and TCO in on-premise deployments, freeing up CPU resources and improving data throughput.
The Asus Zenbook A16 introduces the Snapdragon X2 Elite Extreme, a chip promising significant on-device AI performance. However, the review suggests the chip's effectiveness is contingent on overall system integration, a critical factor for those evaluating Large Language Model (LLM) deployments on local or edge hardware, where the balance between silicio power and system design determines TCO and data sovereignty.
The community of developers and tech professionals is inquiring about the real capabilities and optimal use cases of devices featuring the M5 Max chip with 128GB of unified memory for running Large Language Models (LLMs) locally. The goal is to gather honest feedback on performance, satisfactions, and limitations compared to cloud-based frontier models.
The UALink Consortium, comprising tech giants, has released the 2.0 specifications for its GPU interconnect standards, positioning itself as an alternative to Nvidia's NVLink and NVSwitch. Its modular approach, separating the physical layer from protocols, aims to accelerate development. However, the market arrival of silicio based on version 1.0 is still months away, highlighting the complexity and lead times for critical AI infrastructure.
The introduction of Apple's M1 Silicio chips in late 2020 marked a technological turning point, lauded for its innovations. However, Apple's "walled garden" model, characterized by total platform control and reliance on its proprietary silicio, has raised questions about its expansion beyond the company's ecosystem. This approach limits deployment options for enterprises seeking flexibility and control, particularly for AI/LLM workloads.
Wonderful Hi-Tech, led by Chairman Ming-Lieh Chang, is strategically investing in AI servers and the satellite sector. This move aims to capitalize on emerging market opportunities, positioning the company in key areas for the next phase of technological and infrastructural expansion.
Intel is revitalizing its advanced chip packaging business, reactivating a key plant in New Mexico with billions in investments, including funds from the US CHIPS Act. This strategic move aims to solidify its position in the AI market by combining multiple chiplets into a single custom component, placing it in direct competition with giants like TSMC to meet the growing demand for computing power.
Anthropic has revealed an annual run rate of $30 billion and plans to deploy 3.5 GW of new Google AI accelerators. Broadcom has been commissioned by Google to produce these next-generation AI and datacenter networking chips, underscoring the crucial role of custom silicio in large-scale AI infrastructures.
Nvidia marks a strategic shift with the development of its "Vera" CPU, moving away from reliance on external solutions. This move aims to strengthen hardware integration for AI workloads, with significant implications for on-premise deployments seeking optimization, control, and data sovereignty.
Nvidia introduces Vera, its first CPU, marking a strategic evolution towards greater hardware integration. This move aims to optimize AI and HPC system performance, offering new perspectives for on-premise deployments seeking control and efficiency. The initiative could redefine the balance between CPUs and GPUs, impacting TCO and data sovereignty.
A bipartisan legislative proposal in the United States aims to block the export of DUV (Deep Ultraviolet) chipmaking and etching tools to prominent Chinese companies, including Huawei and SMIC. This initiative, focused on lithography equipment, highlights growing geopolitical tensions and their repercussions on the global semiconductor supply chain, with potential effects on on-premise AI deployments.
At NVIDIA GTCX 2026, MSI showcased a range of hardware solutions designed for demanding AI workloads. The offerings include desktop workstations like EdgeXpert and XpertStation WS300, alongside multi-GPU servers featuring advanced air and liquid cooling systems. These proposals highlight MSI's commitment to providing robust infrastructure for on-premise deployments and Large Language Model inference.
After a year of preparations, the Linux kernel is set to remove support for i486-class CPUs. This decision, anticipated with the release of Linux 7.1, marks a significant step in the operating system's evolution, with implications for legacy hardware and on-premise deployment strategies.
Tiny Corp, known for its Tinygrad framework and the development of a "sovereign" AMD driver stack, has opened pre-orders for its Exabox system. Priced at an estimated $10 million, the system promises massive AI compute power, targeting on-premise deployments for companies seeking control and data sovereignty. Deliveries are expected next year.
Intel is heavily investing in advanced chip packaging, a technology proving crucial for the expansion of artificial intelligence. This strategy could generate billions, positioning the company at the forefront of hardware innovation for AI workloads, with significant implications for on-premise deployments and data sovereignty.
AMD is preparing to launch the Ryzen 9 9950X3D2 Dual Edition, a flagship desktop processor featuring a dual-cache architecture. Initial listings from retailers in Canada and the UK indicate a price point of approximately $1,000. This high-performance chip could offer an interesting solution for intensive workloads, including LLM inference scenarios on self-hosted infrastructures.
The UK has confirmed the integration of the DragonFire laser weapon system onto Royal Navy destroyers by 2027. Capable of neutralizing high-speed drones at a cost of just $13 per shot, this technology marks a significant step in air defense evolution, offering an economical and precise alternative to traditional missiles. Its adoption reflects a trend towards high-efficiency and operational control solutions.
AMD and Valve have introduced significant updates for Kaveri and Kabini APUs in the upcoming Linux kernel 7.1. These efforts aim to optimize the user experience, highlighting the importance of continuous driver support and open-source collaboration for hardware stability and performance in self-hosted environments.
Advantech has revealed the specifications for Intel's new Wildcat Lake CPUs, targeting the low-budget segment. The Core 7 350, Core 5 320, and Core 3 305 models were spotted in the datasheet for the MIO-5356 Single Board Computer, indicating their potential use in embedded solutions and edge-based AI workloads where TCO and energy efficiency are paramount.
The upcoming Mesa 26.1 release introduces a feature that simplifies the simulation of a GPU reset using the LLVMpipe software driver. This seemingly minor addition offers a significant advantage to compositor and application developers. It allows them to more efficiently test how their code behaves in GPU recovery scenarios, thereby contributing to improved software robustness and reliability in critical environments.