DeepSeek all-in-one machine, crazily launching new products
DeepSeek all-in-one machine, crazily launching new products
DeepSeek's' fire 'burns to all-in-one machines.
The computing craze sparked by DeepSeek is still ongoing.
China Telecom recently announced the launch of the DeepSeek version of the Xirang intelligent computing all-in-one machine, which is based on Huawei Ascend chips at the hardware level and offers various specifications such as 8-card, 16 card, and 32 card models. Previously, JD Cloud also released the DeepSeek big model all-in-one machine, which supports domestic AI acceleration chips such as Huawei Ascend, Haiguang, Cambricon, Moore Thread, and TianTian Zhixin.
At present, DeepSeek big model all-in-one machines are divided into inference all-in-one machines and training push all-in-one machines. The DeepSeek inference all-in-one machine is equipped with models of different sizes such as DeepSeeker R1 32B, 70B, and Full Blood 671B, with prices ranging from tens of thousands to millions. It is mainly aimed at enterprise users who are sensitive to data security and privacy. The price of the training and pushing all-in-one machine is even higher, with the price of the all-in-one machine used for pre training and fine-tuning of the DeepSeeker R1 32B model reaching millions.
Some cloud technology experts believe that the key is whether the large model all-in-one machine can generate economic value. The main purpose of deploying large model all-in-one machines is localization, which solves data privacy issues. Previously, the economic value of large model all-in-one machines was not high, and high investment affected their implementation. In the future, applications still need to have greater development
Cloud Service 'National Team'
The DeepSeek version of China Telecom's "Tianyi Cloud" Xirang intelligent computing all-in-one machine is based on domestically produced Ascend chips and provides various specifications and models such as Ascend 8-card, 16 card, 32 card, etc; The DeepSeek version of the mobile cloud intelligent computing all-in-one machine is based on domestic CPU/GPU and operating system. The all-in-one machine adopts domestic chips, matched with self-developed inference engine, and perfectly adapts to the DeepSeek big model, achieving 100% localization of the technical link.
China Mobile's "Mobile Cloud" announced the launch of the DeepSeek version, an intelligent computing all-in-one machine, which helps users to solve autonomous and controllable DeepSeek model applications in one stop. The DeepSeek version of the mobile cloud intelligent computing all-in-one machine is built with domestic CPU/GPU and operating system as the core, providing a complete set of information and innovation solutions. The characteristics of "one cloud, multiple cores, full stack autonomy, security and trustworthiness" not only ensure enterprise data security and business compliance, but also lay a solid foundation for national information security.
China Unicom's "Unicom Cloud" announced the launch of the DeepSeek all-in-one machine+GPU cloud server "software and hardware integration" solution, accelerating the intelligent transformation of industry customers. Based on the advanced computing scheduling platform "Xingluo" of China Unicom Cloud, it is deeply adapted to domestic computing chips to achieve 100% independent and controllable technology links. For the deep integration of DeepSeek's entire series of large models, we provide solid computing power support for enterprise digital transformation with domestic computing power, ensuring data and system security.
It is reported that the layout of the aforementioned manufacturers aims to meet the needs of privatization and localization deployment of central state-owned enterprises, universities, and other institutions.
Ascend DeepSeek all-in-one machine
Ascend has launched a rich range of all-in-one products, including servers, inference cards, and acceleration modules, which are fully compatible with the DeepSeek V3/R1 full blooded/distilled versions of the entire series of models. They cover intelligent dialogue, intelligent programming, document analysis, development boards, and other full scenario applications, and can meet the AI needs of various users in different scenarios.
At the same time, Ascend provides broad innovation space for partners with its open technology architecture and abundant development resources. As of February 12th, 13 companies have built their own DeepSeek all-in-one products based on Ascend products, meeting the diverse needs of segmented markets, further expanding the application boundaries of AI technology in various industries, and jointly promoting the vigorous development of the artificial intelligence industry.
Lenovo DeepSeek all-in-one solution
Lenovo Group and Muxi Corporation jointly released an all-in-one solution based on the DeepSeek big model. This solution is based on the core architecture of "Lenovo server/workstation+Muxi training and push integrated GPU+autonomous algorithm", combined with Lenovo AI force intelligent agent development platform, to launch a dual product form of intelligent agent integrated machine and training and push integrated server, taking the lead in achieving full chain coverage from training of billions of parameter large models to the landing of scene based reasoning.
The DeepSeek intelligent agent all-in-one machine for agile deployment uses ThinkStationPX workstation as the carrier and is equipped with Muxi Xisi N260 GPU.
Wave DeepSeek "reasoning all-in-one machine"
The Inspur Innovation Technology Server NF5468H7 deeply integrates the entire DeepSeek series of models with domestic AI acceleration cards to create an "out of the box" intelligent computing solution. Fully compatible with DeepSeek V3 (universal base), R1 (inference optimization), Janus Pro (multimodal) and other series of models, supporting rapid adaptation to scenarios such as intelligent customer service and dialogue, text analysis, industrial quality inspection, medical image recognition, etc.
Equipped with domestic AI computing power products, the single card FP16 computing power reaches the international mainstream level, supports FP32/FP16 hybrid precision acceleration, and can complete real-time inference of billions of parameter models on a single machine.
New H3C DeepSeek all-in-one machine
New H3C Group, relying on its full stack technology ecosystem and industry practice in the AIGC field, has launched the industry's first DeepSeek big model all-in-one solution with deep coupling of "intelligent computing algorithm governance".
Infrastructure layer: Provide pre integrated intelligent computing power clusters to achieve unified orchestration of network, storage, and security. Algorithm service layer: Pre installed DeepSeek basic large model and industry optimized version, supporting distillation and fine-tuning of subsequent models. Application enablement layer: Built in H3C AlStore intelligent asset platform, with over 1000 open source datasets, models, and images available for free, allowing for one click payment for commercial resources. Shorten the time for users to obtain AI assets such as hundreds of GB images and models from a few days to a few hours.
JD Cloud DeepSeek Large Model All in One Machine
The DeepSeek big model all-in-one machine released by JD Cloud has the core advantages of "fully compatible with domestic computing power" and "out of the box", with built-in dual engines of intelligent agents and knowledge bases. It comes pre installed with over 100 industry solution templates and thousands of plugins, supporting rapid construction of local knowledge bases. It achieves deep integration between models and business scenarios through a three-layer architecture of "model instructions+knowledge base injection+business system integration", improving inference efficiency.
This all-in-one machine is compatible with domestic chips such as Huawei Ascend and Cambricon, balancing independent controllability and cost optimization, and is suitable for high compliance industries such as finance and healthcare.
The business model of all-in-one machines
In fact, even before DeepSeek became popular, all-in-one machines had become a major trend in the commercial exploration of large models.
Last year, a representative from a server manufacturer stated in an interview that the usage habits of all-in-one machines and traditional hardware users are more compatible, and customers have a relatively strong willingness to accept all-in-one machines. About 30% of users are interested in customized application development. There is a gap between the current application status and market trends. How can enterprises, especially traditional enterprises, quickly implement big model applications? Based on this background, we have developed a big model all-in-one machine, "the person said.
Several AI computing professionals have expressed their recognition of the business model of all-in-one machines, believing that they have high integration, can be quickly deployed privately, and are relatively inexpensive.
____________ Article source: Semiconductor industry vertical and horizontal
