⚡ 本页包含 AI 生成的分析内容，仅供参考
本文介绍了NVIDIA GB10 SoC，它是一款用于AI加速的双芯片系统，采用台积电3nm工艺，集成了20个ARM v9.2 CPU核心和Blackwell iGPU，支持第五代Tensor Cores和第四代RT Cores，实现了31 TFLOPS FP32和1 PFLOPS FP4的算力，为DGX Spark工作站提供强大算力。
Nvidia, Westford, MA, 2Nvidia, Santa Clara, CA, 3Nvidia, Austin, TX 1 Abstract This paper details the GB10 SoC that powers the recently launched DGX™ Spark workstation. A dual-die solution, fabricated in TSMC’s 3nm process, co-packages a CPU die featuring 20 ARM v9.2 cores alongside a Blackwell iGPU die tuned for desktop performance. The GPU features 5th-generation Tensor Cores, 4th-generation RT cores, and 24MB of L2 cache, achieving 31TFLOPS of FP32 and 1PFLOPS of FP4. Introduction NVIDIA recently introduced the DGX Spark workstation (a.k.a. Project DIGITS) powered by the innovative GB10 SoC (Fig. 17.1.1), addressing the growing need for accessible AI development platforms that bridge datacenter capabilities with desktop accessibility. The GB10 SoC enables AI developers, researchers, data scientists, and students to prototype, fine-tune, and run large AI models locally while maintaining seamless workload portability
A. Skende1, G. Rosseel2, N. Pinckney3