Bidders flag that Nvidia A100 Tensor Core GPU is at EOL

Several bidders for India’s AI Mission have flagged that the Nvidia A100 Tensor Core GPU is at its end-of-life (EOL). Nvidia announced the EOL in January 2024 and will discontinue all A100 products, including the PCIe and SXM A100 GPUs. Despite this, E2E Networks, the company who quoted A100, claims A100s continue to be available at major cloud service providers (CSPs).

Implications of EOL GPUs

• Availability and Support: Sunil Gupta, chief executive of Yotta Data services, notes that while EOL GPUs can be used, there may be no technical support or replacements available if problems arise, such as performance issues or card failure.
• Suitability for Workloads: EOL GPUs are suitable for development or User Acceptance Testing (UAT) environments. However, they should be avoided for mission-critical production workloads requiring 24/7 uptime.
• Cost-Effectiveness: Anuj Bairathi, chief executive of Cyfuture, suggests that older GPUs, like the A100, can still process AI workloads and may offer a better price than newer generations. Vinay Chhabra, managing director of AceCloud, adds that using legacy GPUs alongside newer models can save costs and maintain good performance, especially during supply shortages.
• Nvidia’s Stance: Nvidia has indicated that A100 will still be available for a while through system vendors and partners, who can deliver A100 for much longer. Software support for the product is expected to continue for many years after the last time buy.
  Alternative Products

When GPUs like the A100 and H100 reach EOL, Nvidia typically recommends alternative products with improved performance, power efficiency, and features. Fusion Worldwide reported that Nvidia issued an EOL notice impacting A100 80G, with a final forecast date of February 27.

CT Bureau