Artificial Intelligence thread
- Thread starter 9dashline
- Start date
I think we might be approaching the embodied AI revolution in smart factories with all the recent humanoid robots launched, and their focus on AI.
Have we heard of Huawei's Ascend 910B? China Mobile built an AI cluster in Ningxia. The cluster includes a mixed variety of AI chips: A800, H800, Ascend 910B, and other domestic GPUs. It is to offer AI computation on the market for big model trainings.
随着ChatGPT等AI通用大模型的爆发，智算已成为算力新的发展高峰，直接影响数字经济发展速度，决定社会智能发展高度。8月20日，记者从2023中国算力大会第二届西部数谷算力产业大会上获悉，宁夏主动把握产业变革机遇，建成西部首个面向市场的千卡级AI训练集群——中国移动（宁夏中卫）数据中心，可提供算力服务运营，支撑行业大模型产业发展。
该数据中心采用A800、H800、华为910B及其他国产GPU服务器，以IB互联方式组网，形成算力、存储、管理等全链条能力，实现毫秒级算存数据传输，满足无阻塞大模型训练需求。集群测试性能比肩美国人工智能研究公司OpenAI的训练集群，是国内首个通过“信通院可信人工智能生成内容（AIGC）”认证的平台，可为客户提供裸金属、容器化服务。
据介绍，宁夏移动建设运营智算业务具有数据中心和基础网络双优势。该数据中心是目前西北唯一工信部双认证、宁夏规模最大的新型绿色数据中心，采用创新型制冷技术，全年PUE控制在1.25以内。互联网出口总带宽12.6T，直连全国26个主要城市，占宁夏区域整体承载能力的63%。其重点打造了低时延、高品质、业界领先的12580传输时延圈，至北京时延7.87毫秒，至上海时延14.09毫秒，至韶关时延15.73毫秒。
国家工业信息安全发展研究中心发布的《宁夏算力网络承载能力评估报告》显示，宁夏移动网络承载能力可满足95%以上的场景需求。中国移动（宁夏中卫）数据中心在硬件、软件、技术、运营等方面已初步成熟完善，可为不同类型客户提供从算力出租、语料采集到模型训练的个性化、定制化服务。
“目前，平台资源已全部售罄，市场需求远大于建设规模。我们将建立智算业务运营体系及客户服务体系，引入合作伙伴，成立服务保障专业团队，全力支撑智算业务发展，助力企业快速转型。”宁夏移动相关负责人表示。
下一步，宁夏移动将积极培育AIGC创新大生态，启动实施行业融合赋能行动，全产业链协同，培育推广一批影响面大、带动性强、示范效应突出、安全保障能力强的应用场景，推动人工智能技术在融合应用中迭代升级，加快推进各行业数字化转型、网络化协同、智能化改造。
随着ChatGPT等AI通用大模型的爆发，智算已成为算力新的发展高峰，直接影响数字经济发展速度，决定社会智能发展高度。8月20日，记者从2023中国算力大会第二届西部数谷算力产业大会上获悉，宁夏主动把握产业变革机遇，建成西部首个面向市场的千卡级AI训练集群——中国移动（宁夏中卫）数据中心，可提供算力服务运营，支撑行业大模型产业发展。
该数据中心采用A800、H800、华为910B及其他国产GPU服务器，以IB互联方式组网，形成算力、存储、管理等全链条能力，实现毫秒级算存数据传输，满足无阻塞大模型训练需求。集群测试性能比肩美国人工智能研究公司OpenAI的训练集群，是国内首个通过“信通院可信人工智能生成内容（AIGC）”认证的平台，可为客户提供裸金属、容器化服务。
据介绍，宁夏移动建设运营智算业务具有数据中心和基础网络双优势。该数据中心是目前西北唯一工信部双认证、宁夏规模最大的新型绿色数据中心，采用创新型制冷技术，全年PUE控制在1.25以内。互联网出口总带宽12.6T，直连全国26个主要城市，占宁夏区域整体承载能力的63%。其重点打造了低时延、高品质、业界领先的12580传输时延圈，至北京时延7.87毫秒，至上海时延14.09毫秒，至韶关时延15.73毫秒。
国家工业信息安全发展研究中心发布的《宁夏算力网络承载能力评估报告》显示，宁夏移动网络承载能力可满足95%以上的场景需求。中国移动（宁夏中卫）数据中心在硬件、软件、技术、运营等方面已初步成熟完善，可为不同类型客户提供从算力出租、语料采集到模型训练的个性化、定制化服务。
“目前，平台资源已全部售罄，市场需求远大于建设规模。我们将建立智算业务运营体系及客户服务体系，引入合作伙伴，成立服务保障专业团队，全力支撑智算业务发展，助力企业快速转型。”宁夏移动相关负责人表示。
下一步，宁夏移动将积极培育AIGC创新大生态，启动实施行业融合赋能行动，全产业链协同，培育推广一批影响面大、带动性强、示范效应突出、安全保障能力强的应用场景，推动人工智能技术在融合应用中迭代升级，加快推进各行业数字化转型、网络化协同、智能化改造。
910B should be the improved Ascend chip with 320 TFLOPS FP16 over the original 256 TFLOPS, from what I've seen mentioned. Though it would be nice to see what new AI chips Huawei unveils this year.
So this is the same chip as we generally refer as "Ascend 910", not something newer after TSMC stopped fabbing for Huawei?
Chinese AI startup DPTechnology secures over RMB 700 million($96 million) in a new funding round
China’s AI startup DPTechnology(深势科技) recently completed a new funding round, raising over RMB700 million(about $96 million).
Participating in this financing round are Zhongyuan Capital, MSA Capital, Loyal Valley Capital, Evergreen Scitech Delta, and several other industrial funds .
The funds raised from this round will be primarily used to continue attracting top talents in the industry, deepening the full-stack capabilities from algorithmic innovation to industrial software and solutions, and speeding the industrialization and commercialization of related technologies in sectors such as biomedicine, new energy, chemicals, and semiconductors.
Founded in 2018 and headquartered in Beijing, DPTechnology has established R&D centers in Shanghai and Shenzhen. Its research and technology team is led by academicians from the Chinese Academy of Sciences, bringing together over a hundred scientists and engineers from various fields including mathematics, physics, chemistry, biology, materials, and computer science, according to the company.
DPTechnology is the pioneer of the AI for Science scientific research paradigm. It is dedicated to using artificial intelligence and molecular simulation algorithms, combined with advanced computing methods, to solve important scientific problems and create a new generation of micro-scale industrial design and simulation platforms for human civilization’s most fundamental biomedical, energy, materials, and information science research.
Powered by AI for Science new paradigm, DPTechnology combined multi-scale modeling, machine learning with high-performance computing for scientific research. The company has introduced a series of computation engines and pre-trained models that address fundamental scientific challenges such as micro- and mesoscale simulation, molecular property prediction, and experimental characterization optimization.
Over the past year, DPTechnology and its partners have released a series of significant technological achievements, upgrading a range of AI for Science infrastructures, and propelling its conception into the era of pre-trained models.
China’s AI startup DPTechnology(深势科技) recently completed a new funding round, raising over RMB700 million(about $96 million).
Participating in this financing round are Zhongyuan Capital, MSA Capital, Loyal Valley Capital, Evergreen Scitech Delta, and several other industrial funds .
The funds raised from this round will be primarily used to continue attracting top talents in the industry, deepening the full-stack capabilities from algorithmic innovation to industrial software and solutions, and speeding the industrialization and commercialization of related technologies in sectors such as biomedicine, new energy, chemicals, and semiconductors.
Founded in 2018 and headquartered in Beijing, DPTechnology has established R&D centers in Shanghai and Shenzhen. Its research and technology team is led by academicians from the Chinese Academy of Sciences, bringing together over a hundred scientists and engineers from various fields including mathematics, physics, chemistry, biology, materials, and computer science, according to the company.
DPTechnology is the pioneer of the AI for Science scientific research paradigm. It is dedicated to using artificial intelligence and molecular simulation algorithms, combined with advanced computing methods, to solve important scientific problems and create a new generation of micro-scale industrial design and simulation platforms for human civilization’s most fundamental biomedical, energy, materials, and information science research.
Powered by AI for Science new paradigm, DPTechnology combined multi-scale modeling, machine learning with high-performance computing for scientific research. The company has introduced a series of computation engines and pre-trained models that address fundamental scientific challenges such as micro- and mesoscale simulation, molecular property prediction, and experimental characterization optimization.
Over the past year, DPTechnology and its partners have released a series of significant technological achievements, upgrading a range of AI for Science infrastructures, and propelling its conception into the era of pre-trained models.
Not looking good, I work in finance, maybe I should start a career as a chef? Because most of the finance job in Poland is outsourcing so a monkey can do the job too, so it's ideal for automation
There is robotic chef that is being deployed. I seriously doubt it is possible for AI to replace humans except in very basic jobs and functions within the next 5 years. However, call center is quite vulnerable as AI probably could soon replace many functions of customer service agents.
I do wonder if that 910b is actually referring to another new variant that's more powerful. But you are probably right it is just the 320tflops version