近日，北京智源人工智能研究院副院长、总工程师林咏华表示，当前中国AI芯片的大模型集群训练性能，只有个别接近英伟达A100/A800，大多数不到50%。此外，中国AI芯片公司有40余家，但中国AI芯片整体的市场占有率加起来不超过10%，各家AI芯片软件各异、生态十分零碎割裂。
根据报道，中国AI芯片的大模型训练性能与国外相比存在明显差距。以英伟达A100/A800为例，中国AI芯片的大模型集群训练性能目前只有个别接近，大多数仍达不到其水平。此外，中国AI芯片公司数量众多，但市场份额加起来不超过10%。这些问题导致了中国AI芯片的生态差距巨大。
为了解决这个问题，中国AI芯片公司需要加强合作，提高软件生态的统一性和完整性。同时，政府也需要加大对AI芯片行业的支持力度，促进AI芯片产业的快速发展。
新闻翻译：
Chinese AI chip giant performance falling behind foreign counterparts
Keywords: Chinese AI chip, large model training performance, NVIDIA A100/A800, ecosystem gap
News content:
Recently, Dr. Lin Yonghua, the副院长 and general engineer of Beijing Academy of Artificial Intelligence, stated that the current large model cluster training performance of Chinese AI chips is only slightly better than that of NVIDIA A100/A800, with most of them falling short of the benchmark. Moreover, there are more than 40 Chinese AI chip companies, but the combined market share of Chinese AI chips does not exceed 10%, and each company operates its own software ecosystem, which is fragmented and inconsistent.
According to the report, the training performance of Chinese AI chips’ large models is significantly behind that of foreign counterparts. Taking NVIDIA A100/A800 as an example, the training performance of Chinese AI chips’ large model clusters is currently only slightly better than that of NVIDIA’s, while most of them still fall short of the benchmark. In addition, there are more than 40 Chinese AI chip companies, but the combined market share of Chinese AI chips does not exceed 10%. These issues have led to a huge ecosystem gap in the Chinese AI chip industry.
To solve this problem, Chinese AI chip companies need to strengthen cooperation, improve the unity and integrity of the software ecosystem. Meanwhile, the government also needs to increase its support for the AI chip industry, and promote the rapid development of the AI chip industry in China.
【来源】https://www.chinastarmarket.cn/detail/1530280
Views: 1