NVIDIA A16 Enterprise 64GB 250W

GPU Architecture

NVIDIA Ampere

NVIDIA Third-Generation Tensor Cores

160 total Tensor Cores (40 cores per GPU, 4 GPUs)

NVIDIA CUDA Cores (shading units)

5120 total FP32 CUDA Cores (1280 cores per GPU, 4 GPUs)

NVIDIA RT Cores

40 total RT Cores (10 cores per GPU, 4 GPUs)

Double-Precision Performance (FP64)

Not applicable

Single-Precision Performance

FP32: 4x 4.5 TFLOPS<br>Tensor Float 32 (TF32): 4x 9 TFLOPS, 4x 18 TFLOPS*

Half-Precision Performance

FP16: 4x 17.9 TFLOPS, 4x 35.9 TFLOPS*

Bfloat16

Not applicable

Integer Performance

INT8: 4x 35.9 TOPS, 4x 71.8 TOPS*

GPU Memory

64GB GDDR6 (16 GB per GPU, 4 CPUs)

Memory Bandwidth

4x 200 GB/s

ECC

Yes

Interconnect Bandwidth

Not applicable

System Interface

PCIe Gen 4, x16 lanes

Form Factor

PCIe full height/length, double width (dual slot)

Multi-Instance GPU (MIG)

No support

Max Power Consumption

250 W

Thermal Solution

Passive

Graphics APIs

DirectX 12.07, Shader Model 5.17, OpenGL 4.68, Vulkan 1.18

Compute APIs

CUDA, DirectCompute, OpenCL, OpenACC