Commenters focused on Fujitsu’s new MONAKA CPU and its positioning for HPC and AI workloads, pointing to technical claims such as a 2nm process, ARMv9/SVE2 compatibility, wide memory (12-channel DDR5, ~844 GB/s), multi‑TFLOP throughput, and ISA extensions for inference (FP8). Some praised energy-efficiency and noted Fujitsu’s history in custom server silicon and high‑performance networking; others emphasized it’s clearly Arm‑based rather than a homegrown ISA and highlighted documentation that links MONAKA to FugakuNEXT and a next‑gen Monaka‑X. Several sources were cited for deeper specs and benchmarks, and a few commenters proposed the need for common AI throughput metrics (tokens/sec) rather than raw FLOPS or GHz.
Opinion divided sharply over manufacturing, market viability, and trust. Many questioned the “made‑in‑Japan/sovereign” framing, pointing out careful press wording and that TSMC or Rapidus production timelines complicate the claim. Some argued Fujitsu can regain niche HPC share with efficient CPUs, while others called the server/GPU market overcrowded and doubted Fujitsu’s sales, software ecosystem, and willingness to support markets outside Japan. Views on reputation varied too: a few recalled Fujitsu’s prior networking and reliable hardware, while others cited past product defects and the British Post Office scandal as reasons for mistrust.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.
BYD plans to develop a rapid 5-minute EV charging network in Canada using megawatt flash chargers. The project aims to significantly reduce charging times for electric vehicle owners. (driving.ca)
Zettascale is hiring ASIC and FPGA engineers to develop chips for artificial superintelligence and frontier AI models. The company's XPU Grasshopper chips focus on energy-efficient inference and training for advanced AI systems. (zscc.ai)
The RAM crisis is driven by major companies securing long-term supply contracts with manufacturers, limiting availability for consumers. As a result, consumer RAM supplies are severely constrained, impacting the broader market. (madshrimps.be)
Researchers developed accurate MATLAB models of AMD GPU matrix cores to address inconsistencies in hardware behavior and standard compliance. These models enable precise analysis of numerical features and application-level accuracy differences compared to NVIDIA tensor cores. (arxiv.org)
Fairphone reports that RAM now accounts for 60% of the materials cost in device manufacturing due to an AI-induced memory shortage. Smaller laptop and phone makers are redesigning products and testing for fake chips amid the supply constraints. (tomshardware.com)
Apple is exploring the possibility of returning to the server market with a new AI-focused system built around M8 chips and Nvidia networking hardware. The project aims to target companies running AI models independently, with a potential release date in 2029. (macrumors.com)
Today's best Hacker News stories, summarized and screenshotted, one email a day.