The "NVIDIA A" line of GPUs, are designed for servers that use the NVIDIA Ampere architecture, now includes a power-efficient model with a TDP of 40 to 60W.
In addition, because of its slim design (with a thickness of one slot), it enables low profiles, allowing it to be put in compact servers and workstations where it was previously difficult to integrate a GPU.
Peak FP32 performance is 4.5TF, TF32 Tensor Core performance is 9TF / 18TF, BFLOAT16 Tensor Core performance is 18TF / 36TF, Peak FP16 Tensor Core performance is 18TF / 36TF, Peak INT8 Tensor Core performance is 36TOPS / 72TOPS, and Peak INT4 Tensor Core performance is 4.5TF. Is 72TOPS / 144TOPS, and AI inference performance can be up to 20 times faster than on CPU-only servers, according to the company.
The basic specs are as follows: RT core 10 cores, GPU memory GDDR6 16GB, memory bandwidth 200GB / sec, interface PCI-Express4.0 (x8), media engine video encoder x1, video decoder x2, and media engine video encoder x1 and decoder x2.