Volcano engine releases large model training cloud platform: supports Vanka training and improves cluster performance by three times

On April 18, Volcano engine unveiled a series of cloud products such as self-developed DPU at its "momentum Conference", and launched a new version of machine learning platform: it supports Vanka-level large model training, microsecond delay network, and flexible computing can save 70% of computing costs. The performance of the GPU instance based on self-developed DPU is more than three times higher than that of the previous generation cluster. Tan Dai, president of the volcano engine, revealed that "more than 70% of the dozens of domestic enterprises in the field of large models are already on the volcano engine cloud."

Photo: Tan Dai, president of volcano engine

At the meeting, the volcano engine announced that the domestic business would be merged with byte jump. Based on the internal and external unified native cloud infrastructure, the idle computing resources of Douyin and other businesses can be quickly scheduled for volcano engine customers to use. Offline business resources can be scheduled at a minute-level 100,000 core CPU, and online business resources can also be reused by tides. Elastic computing preemptive instances offer a maximum discount of more than 80% compared with conventional products.