Nvidia's special graphics card A800 may be banned from sale, price rises to 110,000

In recent years, the popularity of AI and large model training has driven the demand for more powerful computing. NVIDIA’s A100 graphics card became highly sought after to meet this need. However, due to certain regulations, the A100 cannot be sold in some domestic markets. To address this gap, NVIDIA launched the A800 graphics card last year. This new card has been designed with reduced bandwidth to comply with export regulations, earning it the nickname “special edition.”

Performance Comparison: A800 vs A100

The NVIDIA A800 graphics card graphics card closely resembles the A100, but there are notable differences in performance. The A100 features a data transfer rate of 600GB per second, while the A800’s rate is 400GB per second. Although the A800 has a memory bandwidth of up to 2TB/s, it is lower than the A100’s performance. In terms of computing power, the A800 delivers 9.7TFlops FP64 double precision and 19.5TFlops FP32 single precision. The A800 comes in three versions:

1. 40GB HBM2 thread version with a PCIe interface
2. 80GB HBM2e video memory version
3. 80GB HBM2e SXM version

These versions are designed to meet the needs of different customers and use environments. Despite its reduced bandwidth, the A800 supports up to seven instances per card and can work with systems featuring up to eight cards in parallel.

Price Surge and Sales Ban Threat

Recently, reports have surfaced that the A800 may face a sales ban. This news has led to a dramatic price increase, with the cost of an A800 now reaching 110,000 yuan per unit. The price has risen by 20%-30% within just one week, compared to the previous price of 90,000 yuan. The price hike could be linked to NVIDIA’s limited production capacity. He Ning, Vice President of H3C Group, explained during an interview at the Shanghai MWC exhibition that the supply of high-end servers equipped with NVIDIA chips remains tight. The delivery cycle for these servers is approximately 30 weeks, leading to delays in hardware acquisition for AI models.

Impact on AI Industry and Domestic Software Development

The demand for powerful GPUs has intensified, especially with the launch of large models like ChatGPT. Additionally, domestic AI companies are making strides in developing their own large-model software, aiming to catch up with global competitors. However, the limited availability of NVIDIA GPUs and the threat of a sales ban on the A800 may slow the progress of these companies. The reduced access to essential hardware could delay the development of domestic AI models, potentially affecting their competitiveness in the global market.

Conclusion: Challenges for AI Development

While the A800 has proven to be an essential tool for companies needing substantial computational power, its potential ban and price increase may have significant consequences. NVIDIA’s production constraints, along with the surging demand for GPUs, will likely continue to impact AI research and development. For more information about the latest tech products and deals, visit Tiny Deals.