Nvidias GPU module shipments will double this year, exceeding 4 million units

Source: Tiantian IC

    The latest AI supply chain tracking report from Morgan Stanley, a US based foreign investment, shows that Nvidia has shifted some H100 orders to H200 and B100, with an expected sales volume of 400000 units in H100 in 2024.

    Da Mo pointed out that the shipment volume of Nvidia GPU modules in 2023 is about 1.8 million, and it is expected to double in 2024, possibly exceeding 4 million. With the launch of new production lines in Mexico, Nvidia‘s GPU carrier shipments will also significantly increase in 2024.

    In addition, Damo stated that Nvidia China‘s customized version of H20 produced as many as 200000 to 300000 chips per month in the first half of the year. However, due to poor performance compared to H100 chips and concerns that the United States may tighten restrictions again, Chinese companies are unwilling to purchase the downgraded H20 and are testing domestic alternatives.

    According to the previously exposed parameter information, Nvidia H20 is of the same series as H100 and H200, both using Nvidia Hopper architecture, but with an increased graphics memory capacity of 96GB HBM3 and a GPU graphics memory bandwidth of 4.0TB/s. In terms of computing power, the FP8 of this product has a computing power of 296 TFLOPS and FP16 has a computing power of 148 TFLOPS, which is only 1/13 of the strongest AI chip H200 today.