AMD has released details of its new EPYC 8005 processor series, based on the Zen 5 architecture. Featuring configurations of up to 84 cores and a 225W TDP, this line represents a significant evolution for servers, offering a balance between core density and energy efficiency. The series is set to bolster capabilities for on-premise deployments, supporting intensive workloads and addressing data sovereignty and TCO requirements.
AMD has unveiled further details on its EPYC 8005 "Sorano" processor series, succeeding the EPYC 8004 "Siena". With SKUs ranging from 8 to 84 cores, these new chips are crucial for companies planning on-premise AI workload deployments, offering greater control and TCO optimization. The availability of full specifications now allows for in-depth evaluation for self-hosted architectures.
Amprius Technologies, a specialist in high-energy-density silicon anode batteries, has announced a partnership with Matternet, a certified drone delivery company. The collaboration involves Amprius supplying its SiCore® lithium-ion cells for Matternet's autonomous delivery drones, aiming to significantly extend their operational range. This advancement is crucial for the efficiency of edge systems and distributed AI.
A Newegg bundle features an AM5 configuration with a 9950X3D2 processor, 64GB of GSkill RAM, 4TB of fast M.2 storage, and an MSI motherboard for $2,269. This high-end hardware combination can serve as a solid foundation for LLM development and inference in local environments, offering data control and a potential starting point for TCO evaluations in self-hosted contexts.
AEM, a materials specialist, has begun sampling anti-warpage film and PTFE materials, targeting the semiconductor and artificial intelligence sectors. This move highlights the importance of foundational materials for advanced chip manufacturing, which are critical for AI infrastructures, especially in on-premise deployment contexts where reliability and performance are paramount.
XPeng has introduced a mass-produced Robotaxi, integrating AI chips developed in-house. This move highlights the growing trend among automotive manufacturers to invest in proprietary silicon for artificial intelligence, aiming to optimize performance, energy efficiency, and control over autonomous driving systems. XPeng's approach underscores the importance of on-device AI processing for critical applications like autonomous vehicles.
Taiwan has announced new funding for research and development in humanoid robotics and materials innovation. These strategic investments aim to strengthen the island's technological leadership, with significant potential implications for the development of specialized hardware and artificial intelligence solutions for edge computing and on-premise deployments, crucial for data sovereignty and control over AI pipelines.
A comprehensive study explores real-time Diffusion Model optimization on the Apple M3 Ultra, featuring a 60-core GPU and 512 GB of unified memory. Researchers achieved 22.7 FPS for 512x512 img2img transformation by combining CoreML conversion and the SDXS-512 model. The research reveals that optimization strategies established for NVIDIA CUDA GPUs do not directly apply to Apple Silicon's architecture, highlighting a fundamentally different optimization landscape.
Liquid cooling, traditionally associated with high-end AI GPUs, is now expanding its adoption to crucial components such as memory cards and network cards. This evolution reflects the increasing power density and thermal requirements of modern AI infrastructures, with significant implications for on-premise deployments, TCO, and the sustainability of data centers dedicated to Large Language Models.
A comparative analysis benchmarked 21 GPUs, primarily consumer-grade, running a Text-to-Speech (TTS) model (OmniVoice) with a VRAM peak of approximately 5 GB. The tests, conducted on cloud rental platforms and compared against an RTX 3090, offer an estimate of relative performance. This informal study highlights the trade-offs between cost and capability for on-premise deployments of less demanding AI workloads, focusing on efficiency and hardware resource management.
SanDisk has highlighted that, currently, AI-dedicated data centers still lack a compelling economic case to fully replace hard disk drives (HDDs) with solid-state drives (SSDs). The statement underscores the challenges related to Total Cost of Ownership (TCO) and the varying performance and storage capacity requirements for AI workloads, suggesting that HDDs retain a crucial role in specific contexts.
Nvidia's upcoming Rubin platform is projected to significantly impact the LPDDR memory market, surpassing the combined demand from giants like Apple and Samsung by 2027. This forecast, based on market analysis, highlights the accelerating demand for AI-dedicated hardware and its profound implications for the entire technology supply chain, with significant consequences for on-premise deployment strategies and LLM workload TCO evaluation.
Recent Intel Xe graphics driver patches for Linux reveal the existence of multiple PCI IDs associated with the upcoming "Crescent Island" (CRI) accelerators. This discovery suggests a diversified offering of models, with implications for on-premise deployment strategies and hardware selection for AI workloads, impacting TCO and flexibility.
South Korean startup LetinAR is developing thumbnail-sized optical lenses, poised to become a key component for the upcoming era of AI-powered smart glasses. This innovation aims to overcome miniaturization and integration challenges, which are fundamental for deploying advanced AI capabilities in wearable devices.
An enthusiast has reverse-engineered a PlayStation 2 to integrate it into a portable device. The project, featuring a custom motherboard, combines modern functionalities with the console's original silicon, showcasing the complexity and dedication required for hardware customization.
The Nvidia GB300 processor is catalyzing significant growth in the AI server market, fueling demand for dedicated infrastructure. This expansion is further supported by the upcoming "Vera Rubin" phase, anticipated for the third quarter, which promises to bring new capabilities and availability to the artificial intelligence landscape.
Lotes has achieved record revenues, driven by the increasing demand for connectors in server and AI applications. The company is maintaining a competitive pricing strategy to expand its market share in a critical infrastructure segment for Large Language Models (LLM) deployments and artificial intelligence workloads.
The growing interest in running Large Language Models (LLMs) locally is driving the development of compact hardware. A recent reference to an updated "size chart" for Strix Halo mini PCs, projected for May 2026, highlights how dimensions and form factor are crucial for on-premise and edge deployments, influencing TCO, data management, and operational flexibility.
Nan Ya PCB, a key player in printed circuit board manufacturing, is increasing its production capacity. This move responds to the growing demand for advanced substrates, essential for next-generation AI chips. The expansion highlights the pressure on the AI hardware supply chain and its implications for on-premise and cloud deployment strategies, influencing the availability and TCO of dedicated artificial intelligence infrastructure.
Tata Electronics has announced an $11 billion investment to build a semiconductor manufacturing facility in Dholera, India, in collaboration with ASML. This project aims to strengthen India's autonomy in the chip sector, which is crucial for the global technology ecosystem and for the availability of hardware for on-premise deployments of AI and LLM workloads.