• Fractile Computes in Memory for AI Inference

    Sierpinski carpet

    Startup Fractile claims its data-center NPU will be 100× faster than an Nvidia H100 at 1/10th the cost. The secretive company indicates it’s employing in-memory computation to improve MAC-unit utilization and power efficiency. continue reading

  • Please Welcome New XPU-Family Additions

    xpu.pub logo

    I’m pleased to announce two new members of the XPU.pub family. To provide headlines beyond XPU.pub posts and a discussion forum, I’m introducing XPU.news. It’s a site organized like Hacker News and focused on processors. Please register, submit links, and discuss your ideas. (Your browser will probably throw a security warning the first time you… continue reading

  • Tenstorrent Connects to Developers Through Wormhole

    Tenstorrent Wormhole n300

    Tenstorrent’s new Wormhole AI accelerator (NPU) is a development platform like its predecessor. Among upgrades to the earlier Grayskull, Wormhole adds Ethernet for chip-to-chip interconnect. It should helps Tenstorrent develop an ecosystem for a next-generation NPU. continue reading

  • Reliability Plagues Intel as Raptor Defects Emerge

    It's all about the Pentiums music video screencap

    CPU instability indicates Intel’s 13th and 14th Generation Core (Raptor Lake) processors have defects affecting reliability. How the company addresses the problem will tarnish or burnish its reputation. continue reading

  • What Do You Want to See from XPU.pub?

    xpu.pub logo

    Thanks for reading this and my other notes. Extra gratitude to those who have signed up for my newsletter. Looking at the list, I think I know half of you! Here’s a hint at what I’m planning to do next and a request for you to send me ideas. continue reading

  • AMD Zen 5 Raises IPC, Speeds up Math and AI Processing

    enable images to see person experiencing Zen

    AMD’s new Zen 5 CPU boasts a 16% IPC increase over Zen 4, with significant performance gains in math-intensive tasks owing to doubled AVX throughput. A tighter physical design allows for similar die size despite increased cache and wider execution units. The new microarchitecture widens both the front end and back end and will be… continue reading

Popular

Sponsored Content and Plugs

Other Sites You Might Like

Read More

Altera AMD Ampere Apple Arm auto Broadcom Ceva Computex CPU data center DPU edge AI embedded Epyc FPGA Google GPU Imagination Immortalis Intel interconnect Marvell MCU MediaTek memory Meta Microchip Microsoft MLPerf NPU (AI accelerator) Nvidia NXP PC process tech Qualcomm RISC-V SambaNova Semidynamics smartphone SoftBank software Syntiant Tenstorrent Untether

error: Selecting disabled if not logged in