NVIDIA A100

Key Features of the NVIDIA A100

The A100 features 6,912 CUDA cores and 432 tensor cores to dramatically increase the speed of machine learning and artificial intelligence tasks. The GPU supports up to 80GB of HBM2E memory with over 2TB/s memory bandwidth, making it capable of processing the largest models and datasets.

Performance and power

The processor operates at 1410 MHz and is capable of achieving up to 19.5 TFLOPS in FP32 computation, as well as 156 TFLOPS in Tensor Float 32 (TF32) computation. With support for Multi-Instance GPU (MIG) technology, the A100 can be partitioned into up to seven separate GPU instances, allowing flexible resource allocation depending on the task at hand.

Technologies and features

• Third-generation Tensor Cores: delivers up to 20x faster performance than the previous generation, supports new data formats such as TF32 and BF16, and improved FP64 support for high-performance computing.
• NVLink and NVSwitch: new versions of these technologies double the communication bandwidth between GPUs to 600 GB/s, accelerating data transfer under intensive workloads.

Applications for the NVIDIA A100

The A100 is ideal for machine learning, high-performance computing and data analytics applications. It is used in state-of-the-art computing centers, delivering high performance for applications ranging from scientific research to commercial solutions.