Workflow
一家GPU初创公司浮出水面,Raja Koduri创立,基于RISC-V
半导体芯闻·2025-08-07 10:33

Core Viewpoint - Oxmiq Labs, founded by legendary GPU architect Raja Koduri, aims to revolutionize GPU architecture and software by focusing on a software-first approach and developing a new GPU IP that allows for seamless execution of CUDA applications on non-Nvidia hardware without code modification [2][4][12]. Group 1: Company Overview - Oxmiq Labs is a new GPU software and IP startup with a team of over 500 years of combined experience in GPU and AI architecture [2]. - The company aspires to become the next generation of Arm, indicating a focus on flexible and scalable GPU solutions [2]. Group 2: Technology and Innovation - Oxmiq Labs is developing a GPU architecture that integrates RISC-V cores, near-memory computing, and advanced optical transmission technologies, aiming to balance flexibility and performance for multi-modal AI workloads [6][14]. - The OxCore GPU IP integrates scalar, vector, and tensor computing engines in a modular architecture, supporting near-memory and memory computing capabilities [6][8]. Group 3: Software Strategy - The company emphasizes a software-first strategy, with a comprehensive software stack designed to abstract the complexity of heterogeneous hardware and support AI and graphics workloads across various platforms [9][10]. - OXCapsule serves as the core of Oxmiq's software ecosystem, providing a unified runtime and scheduling layer for workload management [10]. Group 4: Product Offerings - Oxmiq offers a chiplet-based system-on-chip (SoC) builder, OxQuilt, allowing customers to create integrated computing clusters tailored to specific workload requirements [8]. - OXPython, a flagship component, enables Python-based CUDA applications to run seamlessly on non-Nvidia hardware, showcasing the company's commitment to breaking hardware silos [12][13]. Group 5: Funding and Market Position - Oxmiq Labs has secured $20 million in seed funding from notable tech investors, including MediaTek, and has achieved initial software revenue [14]. - The company aims to avoid the high capital demands typically associated with chip startups by prioritizing an IP licensing model [14].