GPUDatacenterReleased Apr 12, 2021

NVIDIA A10

24 GB GDDR6 · 384-bit · 9,216 cores

A single-slot, passively cooled Ampere data center GPU for mixed graphics and AI work: virtual desktops, RTX virtual workstations, media streaming and mainstream inference. Several cards fit in one server, and it supports the full NVIDIA vGPU software stack. Announced at GTC in April 2021.

General

Manufacturer: NVIDIA
Model: A10
Segment: Datacenter
GPU chip: GA102
Architecture: Ampere
Release date: Apr 12, 2021
Price: Search on Amazon

Memory

Memory size: 24 GB
Memory type: GDDR6
Bus width: 384-bit
Bandwidth: 600 GB/s
Memory clock: 1,563 MHz

Clocks & silicon

GPU clock: 885 MHz
Boost clock: 1,695 MHz
Shader units: 9,216 CUDA Cores
Tensor / AI cores: 288
RT cores: 72
Transistors: 28.3 billion
Technology (nm): 8 nm
Samsung 8N (8 nm)
TDP / TBP: 150 W

Interface & APIs

Bus interface: PCIe 4.0 x16
SLI / multi-GPU: None (no NVLink)
DirectX: 12 Ultimate (12_2)
OpenGL: 4.6
Vulkan: 1.3
Other APIs: CUDA (compute capability 8.6), OpenCL 3.0

AI support

Precisions: FP8FP16BF16INT8INT4TF32FP64
AI accelerator: 3rd Gen Tensor Cores (288) + 2nd Gen RT Cores (72)
Peak AI TOPS: 500 TOPS
Notes: Third-generation Tensor Cores; FP8 is not supported on Ampere. Peak throughput with structured sparsity: INT4 1,000 TOPS, INT8 500 TOPS, FP16 and BFLOAT16 250 TFLOPS, TF32 125 TFLOPS; dense rates are half. FP32 is 31.2 TFLOPS. One encoder and two decoders, including AV1 decode.