On October 25th, the preview version of Lenovo AI host P7 was officially launched. This device features localized operation and high performance. With a total weight of only 300 grams, its compact size allows it to run continuously 7×24 hours at the desktop.
In terms of hardware configuration, this host is equipped with domestic dual-core chips, with two chips having distinct functions: one is responsible for AI computing inference, and the other handles system scheduling and task execution. The dual-core cooperation maximizes the overall computing power. In terms of performance, it can run large models with 122B parameters locally at maximum, supports 256K ultra-long context, and the inference speed can reach 50 Tokens per second when the network is disconnected. Meanwhile, the overall power consumption at full load is less than 30W, and the operating noise is below 35 dBA, so even when placed next to a desk for long periods, it does not interfere with daily work.
In terms of the software ecosystem and feature design, the device comes with a rich local model library, including general large models, image generation, document recognition, and speech-to-text, etc. There is no need to rely on cloud interfaces every time. Additionally, the new device features Lenovo Tianxi AI, which works out of the box, and users can also configure the Agent by themselves. In Auto mode, Tianxi AI can intelligently select the most suitable and cost-effective solution between local and cloud models based on task complexity, while ensuring good task performance and keeping usage costs reasonable.
To address the challenges faced by current users in cloud-based AI calls, such as high costs, data privacy leaks, and task interruptions due to unstable networks, local AI hosts keep all data processed locally, effectively ensuring information security. Currently, Lenovo’s AI Host P7 has launched a “preview program” before its official release. The first batch of users are invited to try out the product in real programming and office scenarios and provide feedback. By deeply participating in the co-creation of the product, their experience can be optimized to the best possible level.