科技牛带动指数突破3800点后，工信部发声支持GPU，国产GPU六龙争霸！ 1.摩尔线程——逐光飞龙 全功能 GPU，MTT S80 跑 3A 大作，驱动月更；S2000 智算集群推动 OISA 标准，Pre-IPO 估值 255 亿。 2.景嘉微——霸权陆龙 全市场唯一一个cpu，gpu（dpu)双龙头，也是唯一一个形成生态的通用芯片公司，产品包括海光7000 CPU和深算一号/二号GPU（DCU协处理器），广泛应用于服务器、金融、互联网及AI领域，非常具有 有稀缺性。 "黄埔军校"，JM9 接近 GTX1050，党政/金融国产替代占比40%，JM11 瞄准通用计算，堪称国芯基石。 虽然国产GPU短期很难赶上英伟达，AMD，英特尔等，毕竟发展这么多年，多少人想进去分一杯羹， 国外也就只有英伟达跟AMD，连英特尔想做都进不去，国内初创的几家，10个亿几百人两三年的量级 研发肯定还是有巨大差距的，但是就像家电、汽车、手机等行业一样，10年或者20年后还是有可能的， 我觉得人工智能很像当初互联网行业，会深刻改变人类的生活。会有一大批的高增长的企业。而且时间 会非常的长。在微软刚上市时买入，或者也可以等互联网泡沫破 ...

Core Insights - The global computing power landscape is evolving into two parallel paths, with domestic and international routes differing, and opportunities for domestic chips in inference scenarios are expected to emerge this year [1][4] - Two AI chip companies, Muxi Integrated Circuit (Shanghai) Co., Ltd. and Moore Threads Intelligent Technology (Beijing) Co., Ltd., are nearing their IPOs, with Muxi's counseling work completed and Moore Threads' status changing to acceptance [1][2] - The demand for AI chips in China is increasing, with several companies adapting to the popular DeepSeek-R1 model, leading to a significant shift in market dynamics [3][4] Company Summaries - Muxi and Moore Threads were founded in 2020, while Suiruan Technology and Birun Technology were established in 2018 and 2019, respectively; the latter two companies have slower IPO progress compared to the former [2] - Moore Threads has the highest valuation among the four companies at 25.5 billion yuan, followed by Suiruan Technology at 16 billion yuan, Birun Technology at 15.5 billion yuan, and Muxi at 10 billion yuan [2] - Founders of these companies have backgrounds in major overseas chip firms, with Moore Threads' CEO previously serving as NVIDIA's China General Manager and Muxi's founders coming from AMD [2] Market Dynamics - The demand for domestic AI chips has surged, with IDC reporting that 34.6% of the Chinese data center accelerator market was comprised of domestic solutions last year, and this figure is expected to exceed 40% in the first half of this year [3] - The DeepSeek application has driven significant demand for domestic computing power, with many companies adapting their products to this model [3][4] - The global computing power foundation is shifting, with 98% of large model training still relying on NVIDIA, but there is potential for a portion of pre-training to transition to non-NVIDIA cards by the end of the year [4]

Core Viewpoint - The article discusses the performance of domestic GPUs, highlighting the competitive landscape and the impact of sanctions on production capabilities. It emphasizes that while the BR100 from Birun is the most powerful, it cannot be mass-produced due to sanctions, leaving Huawei's 910C as the strongest available option [1][2]. Performance Comparison - The performance ranking of GPUs indicates that the Huawei 910C leads, followed by the Haiguang BW100, with the latter achieving approximately 87% of its potential performance. The Cambrian 590's performance is uncertain, potentially reaching 80% [2]. - The Haiguang BW100 has a deep computing capability of over 400T, which is about 50% of the H800's performance, while the 910C reaches approximately 60% of the H800's performance [1][2]. Memory and Bandwidth - Most domestic GPUs utilize HBM2e memory due to sanctions preventing the use of HBM3e. Sufficient memory is crucial for initiating inference tasks, as insufficient memory can hinder model startup [3]. - Huawei leads significantly in memory bandwidth, which is a critical factor for performance [4]. Pricing Insights - Current market prices for GPUs are as follows: Haiguang BW100 at approximately 100,000 yuan, Huawei 910B at around 70,000 yuan, and 910C at about 180,000 yuan. The Cambrian 590 has seen a price drop from 80,000-85,000 yuan to between 60,000-70,000 yuan [2]. Technical Specifications - A table summarizes key specifications of various GPUs, including memory type, bandwidth, capacity, and arithmetic intensity. For instance, the Huawei 910C features HBM2e with a bandwidth of 3.2 TB/s and a memory capacity of 64 GB [5].