Before the official release of the third quarter report, global GPU giant Nvidia was constantly making new moves.
At the recent 2024 Supercomputing Conference, Nvidia unveiled two new AI hardware, namely the Hopper H200 NVL and GB200 NVL4 supercomputers.
It is reported that the PCIe based Hopper H200 NVL has been fully launched, and these cards can connect up to 4 GPUs through the NVLINK domain, providing up to 7 times faster bandwidth than standard PCIe solutions. Nvidia states that the H200 NVL solution can adapt to any data center and offers a range of flexible server configurations optimized for hybrid HPC and AI workloads. The memory capacity of H200 NVL is 1.5 times that of the previous H100 NVL, the bandwidth is also 1.2 times, and it has 1.7 times the AI inference performance. In HPC applications, its performance is also 30% higher.
The Nvidia GB200 NVL4 is a brand new module that is a larger extension of the existing GB200 Grace Blackwell Superchip AI solution. The GB200 NVL4 module is configured with two Blackwell GB200 GPUs on a larger motherboard, which means it has two Grace CPUs and four Blackwell B200 GPUs.
Compared to the previous generation GH200 NVL4 system, which includes 4 Grace CPUs and 4 Hopper GPUs, the new GB200 NVL4 has 2.2 times the simulation performance, 1.8 times the AI training performance, and 1.8 times the AI inference performance.
It is reported that the GB200 NVL4 supercomputer will be launched in the second half of 2025.
In addition to launching two new AI hardware this time. On the eve of the release of the third quarter report, another unfavorable news also attracted market attention. According to media reports, Nvidia's new generation of Blackwell artificial intelligence (AI) graphics processors are facing delayed delivery and are experiencing issues with overheating of supporting servers, causing some customers to worry that they do not have enough time to get their new data centers up and running.
Insiders revealed that overheating occurs when it is installed in server racks that can accommodate up to 72 chips. Multiple sources, including Nvidia employees, customers, and suppliers, have stated that Nvidia has repeatedly requested suppliers to adjust rack designs in order to address this issue; However, despite multiple attempts, the issue still persists. Affected by the above issues, some Nvidia customers are concerned that there may not be enough time to advance the construction and operation of new data centers. Nvidia responded to the outside world by saying, "Engineering iterations are normal and expected
The Blackwell artificial intelligence chip was released in March this year, and had previously stated that it would be shipped in the second quarter, but it was later postponed. NVIDIA's official website claims that it will "break down the barriers of generative AI and accelerated computing" and bring "breakthrough progress". Currently, investors are closely monitoring the shipment status of Blackwell chips.
Despite the mix of positive and negative news, the market is still most concerned about Nvidia's third quarter performance, which is scheduled to be disclosed after the US stock market closes on Wednesday this week. This is also Nvidia's first financial report since its stock was included in the Dow Jones Industrial Average.
According to Visible Alpha's statistics, the market expects Nvidia's third quarter revenue to reach $33.28 billion, an increase of 84% compared to the same period last year. Net profit will increase by 88.9% year-on-year to $17.45 billion, and net earnings per share will jump from 37 cents in the same period last year to 70 cents.
The data center remains its most lucrative industry, with Nvidia achieving a record high revenue of $26.3 billion in the second quarter. Analysts expect it to break through the new high again in the new quarter, reaching a record high of $29.53 billion. There is also its most promising new product: Black well. Nvidia stated that with the increase in production since the first quarter, it has brought in billions of dollars in revenue, and the company expects this data to reach 5-6 billion dollars in the future.