Tesla P4 @ Inference 2560cores Memory: 8GB 5.5TFlops/FP32, 22TOPS/INT8, GPU104 INT8: offering an 8-bit vector dot product with 32-bit accumulate, 4x 8-bit vector dot product
Tesla P40 @ Inference 3840cores Memory: 24GB 12TFlops/FP32, 47TOPS/INT8, GPU102 INT8: offering an 8-bit vector dot product with 32-bit accumulate, 4x 8-bit vector dot product
Tesla P100 @ Training 3584cores 4.7TF/double 9.3TF/Single-preicision 18.7T/Half-preision Memory: 16GB, 732GB/s; 12GB, 549GB/s APIS: CUDA, DirectCompute, OpenCL, OpenACC
TitanXP 3840Cores Memory: 12GB, Memory Speed: 11.4Gbps, Mmeory Band: 547.7GB/s SinglePrecision: 384-bit, 12TFlops
GPU: Pascal Architecture,
Tesla V100(SXM2) Memory: 16GB L2: 6MB Half: 30TF Single Precision: 15TFlops Double Precision: 7.5TF TensorPerformace: 120TFlops, 4x4FP16 matrix, GPU: Volta