The article presents a new method for mapping code onto Coarse-Grain Reconfigurable Arrays (CGRAs) by decoupling time and space dimensions. This approach significantly enhances compilation speed, achieving a remarkable average speedup of approximately 100,000 times for large CGRAs.
The article presents an analysis of NVIDIA's Blackwell architecture, highlighting its architectural advancements such as 5th generation tensor cores and tensor memory. It introduces an open-source microbenchmark suite aimed at optimizing workloads for modern GPU architectures.
The article discusses various technical advancements in the semiconductor industry, including developments in 3D interconnected networks, neuromorphic accelerators, and memristor-based devices. It highlights the importance of open-source chiplets for high-performance computing and artificial intelligence applications.
Avicena Tech Corp has unveiled its LightBundle optical links, which now achieve data rates of 4Gbps per lane while maintaining low transmitter currents of 100µA per LED. This technology demonstrates a remarkably low bit-error rate of 1x10–12 without the need for forward error correction.
Lenovo's client device business has regained its footing, achieving 25% of global PC shipments and solid profits, largely due to strategic acquisitions. This resurgence mirrors the peak performance seen during the COVID-19 pandemic.
The article discusses the growing interest in glass interposer substrates and hybrid bonding techniques in semiconductor packaging. It highlights a recent publication that explores these technologies in detail, authored by industry experts.
The article introduces Bombyx, a compiler toolchain that translates OpenCilk programs into a format suitable for FPGA hardware acceleration. It emphasizes a shift from OpenCilk's implicit task model to a more efficient explicit continuation-passing model, aligning better with FPGA architectures.
The article discusses the challenges faced by traditional RDMA engines in handling memory page faults, which can hinder performance in high-speed interconnection networks. It highlights the limitations of current techniques like pinning address spaces, which complicate programming and do not fully prevent page faults.
The article discusses the need for Europe's Chips Act 2.0 to adapt to the emerging computing paradigms. It emphasizes the importance of strategic planning for Europe's semiconductor industry to ensure its competitiveness.
Dr. Huaguang Wang from Indium Corporation will present on enhancing the reliability of large BGA assemblies for AI servers and high-performance computing at EPTC 2025 in Singapore. His insights are expected to address critical challenges in the reliability of semiconductor packaging for these advanced applications.
Researchers from ETH Zurich and the University of Bologna have published a roadmap for open-source chiplet-based RISC-V systems aimed at high-performance computing (HPC) and artificial intelligence (AI). The initiative seeks to bridge the performance gap with proprietary designs, starting with a project named Occamy.
The Texas Advanced Computing Center's (TACC) new 'Horizon' supercomputer is set to follow the architecture established by the previous 'Vista' supercomputer. This development is part of TACC's ongoing efforts to enhance computational capabilities for academic science.
The article discusses key architectural considerations that facilitate scalable, low-latency, and power-efficient data movement in high-bandwidth AI and HPC systems. This advancement in fabric IP is crucial for enhancing performance in demanding computational environments.
The article discusses how AI training and inference cluster architectures are significantly boosting datacenter infrastructure spending while also positively influencing HPC architectures. This is primarily due to increased funding for AI initiatives and the necessity to upgrade existing HPC systems for traditional tasks like simulation and modeling.
Nvidia's Vera Rubin AI/HPC platform is expected to launch in Q3 2026, which is crucial for achieving the company's ambitious $500 billion GPU sales goal. This comes despite the setback of losing access to the Chinese market.
A new pilot line is being established to optimize the production process for large organic interposers. This initiative aims to enhance volume production capabilities in the semiconductor sector.
D-Wave is experiencing a significant increase in quantum computer sales and has announced new achievements in quantum supremacy. This positions the company as a leader in the emerging quantum computing market.
Hewlett Packard Enterprise is showcasing its latest advancements in large-scale computing at the SC25 high-performance computing conference. This event highlights HPE's strategic vision for the future of supercomputing.
Nvidia is focusing on integrating quantum computing with AI to maintain its leadership in the AI market while also venturing into the quantum computing sector. CEO Jensen Huang envisions Nvidia as a crucial infrastructure provider and accelerator in this emerging field.
The article discusses the latest rankings of supercomputers in the Top500 list, highlighting the emergence of modest HPC centers over the past two years. It emphasizes the focus on new machines introduced in the rankings during the June and November updates.