IT Home, on-site report, November 12: At the opening ceremony of the Baidu World 2026 conference today, Shen Jue, Executive Vice President of Baidu Group, unveiled the Baidu Tianchi 1024 Card Super Node, a computing cluster set to be launched on the market next year.

Baidu Tianchi's 1024-card super-node features industry-leading optical engine technology, tripling inter-card bandwidth for high-speed data connectivity. Compared to the previous generation, training performance is improved by 2.5 times and inference performance by 3 times.

Baidu Tianchi's 1024 Card Super Node, paired with the G3.5 KV Cache GPU Memory Acceleration Cabinet, enhances cache reuse capabilities and doubles long-context inference performance. In multi-turn dialogue scenarios with high cache hit rates, TTFT (Time to First Token) is reduced by 90%, and overall throughput increases sevenfold.
A single Tianchi cluster with 1,024 GPUs can support the training of a large model with 5 trillion parameters and features excellent horizontal scalability, enabling expansion to clusters of up to ten thousand GPUs.
IT Home noted that Baidu previously announced at last year’s Baidu World Conference that the Tianchi exa-node with thousands of GPUs would launch in 2029; now, the timeline has been accelerated by one year. Additionally, the Kunlun芯 N-series will also be released in 2028.



