杭州深度求索公司（DeepSeek）正式推出了其最新版本的人工智能模型——DeepSeek V3.1。这一版本 的模型采用了UE8M0FP8Scale参数精度，特别适配即将发布的下一代国产芯片结构，在技术和性能上取 得了重大突破，有望为国产AI芯片的发展注入强大动力。 据深度求索公司介绍，UE8M0FP8是一种专门为国产芯片架构优化的浮点数格式。相比传统的FP16或 FP32格式，FP8能够在保持相对较高数值精度的同时，显著减少内存占用和计算开销，尤其适合大规模 AI推理与训练。而"UE8M0"这一定制化命名，体现了该技术针对国产芯片特性所做的深度优化。通过 这种适配，DeepSeek V3.1在推理效率上实现了质的飞跃，相较于此前版本提升了40%，能够在更短时 间内输出结果，大大提高了AI应用的响应速度。 DeepSeek V3.1在数学推理和代码生成等关键性能指标上表现卓越。在数学推理任务中，该模型的正确 率高达92%，展现出强大的逻辑运算和问题解决能力。在代码生成方面，DeepSeek V3.1更是超越了行 业标杆GPT-435%，在Aider多语言编程基准测试中取得了71.6%的高分，且完成一次编程任 ...

Core Viewpoint - The article discusses the recent regulations issued by the U.S. Department of Commerce, which impose restrictions on the use of Huawei's Ascend AI chips globally, highlighting the implications for companies using these advanced computing chips [1][3]. Summary by Sections U.S. Regulations on Huawei Chips - The U.S. Department of Commerce has stated that using Huawei's Ascend chips anywhere in the world violates U.S. export control regulations [3]. - Specific models mentioned include the Huawei Ascend 910B, 910C, and 910D, which may lead to penalties for companies that utilize them [3]. Classification of High-Performance Chips - The regulations categorize advanced high-performance chips into three classes based on their total processing performance (TPP) and performance density: 1. Chips with TPP greater than or equal to 4800 TOPS, or TPP greater than or equal to 1600 TOPS with a performance density of 5.92 or higher [4]. 2. Chips with TPP between 2400 TOPS and 4800 TOPS, and performance density between 1.6 and 5.92, or TPP above 1600 TOPS with performance density between 3.2 and 5.92 [4]. 3. High Bandwidth Memory (HBM) components with memory bandwidth density greater than 2 GB/s per square millimeter [5]. Consequences of Non-Compliance - The regulations indicate that violations could result in severe penalties, including up to 20 years of imprisonment [6]. - Experts have commented that these guidelines are quite stringent, effectively forcing companies to choose between Huawei's H chips and NVIDIA's N chips [6].