Nvidia confirms that its Ampere Architecture will power Geforce series graphics cards

In a report from Market Watch has confirmed that Nvidia plans to release its Ampere graphics architecture onto both the consumer and professional markets, succeeding both Volta and Turing. 

When speaking to Market Watch, Nvidia’s CEO, Jensen Huang, confirmed that Ampere will be a common architecture across both lineups, but that while “there’s great overlap in the architecture, but not in the configuration”. This highlights that each graphics cards lineup will offer unique features which will specify these GPUs for their intended markets. 

Nvidia’s first launch for Ampere is targetted at the AI requirements of modern data centers, cloud computing providers and researchers. Consumer/gaming graphics cards will release at a later date. 

Market Watch has confirmed that Nvidia’s Tesla A100 silicon uses a 7nm manufacturing process and offers users 54 billion transistors. These GPUs can be partitioned by users to allow the GPU to be split between seven different tasks with dedicated resource pools. This can allow a single Tesla A100 to act as seven separate graphics processors with no resource overlap. Nvidia has also stated that its new designs offer a 20x boost in AI performance over today’s Volta series graphics/accelerator cards. 

Nvidia plans to sell its Tesla A100 graphics chips in sets of eight as part of their latest DGX A100 systems, which can handle 56 separate tasks and offer up to 5 petaflops of AI performance. Nvidia’s first DGX A100 systems will ship with pricing starting at $200,000, and early units were shipped to customers earlier this month. 

Nvidia’s GA100 GPU design will offer many of the same strengths as Nvidia’s Tesla V100 silicon, providing support for fas FP64 compute as well as standard FP32 calculations. FP64 isn’t useful for gaming applications, making fast FP64 support a feature which will likely be cut from Nvidia’s consumer-grade Ampere graphics hardware.