GPUDatacenterReleased —

NVIDIA A10M

20 GB GDDR6 · 7,168 cores

The NVIDIA A10M is an OEM variant of the Ampere-generation A10 data center accelerator, supplied through server vendors such as HPE rather than sold at retail. Like the A10 it is a passively cooled 150 W PCIe card without display outputs, designed for virtual desktop infrastructure, cloud graphics and inference where one accelerator serves many users. It carries 20 GB of GDDR6, less than the 24 GB of the standard A10.

General

Manufacturer: NVIDIA
Model: A10M
Segment: Datacenter
GPU chip: GA102
Architecture: Ampere
Release date: —
Price: Search on Amazon

Memory

Memory size: 20 GB
Memory type: GDDR6
Bus width: —
Bandwidth: —
Memory clock: —

Clocks & silicon

GPU clock: —
Boost clock: —
Shader units: 7,168 CUDA Cores
Tensor / AI cores: —
RT cores: —
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: —
DirectX: 12 Ultimate (12_2)
OpenGL: 4.6
Vulkan: 1.3
Other APIs: CUDA 8.6, OpenCL 3.0

AI support

Precisions: FP8FP16BF16INT8INT4TF32FP64
AI accelerator: 3rd Gen Tensor Cores + 2nd Gen RT Cores
Peak AI TOPS: —
Notes: Third-generation Tensor Cores; FP8 is not supported on Ampere. Supported Tensor Core precisions are TF32, BFLOAT16, FP16, INT8 and INT4, with 2:4 structured sparsity doubling peak throughput over the dense rates. FP64 runs on the CUDA cores at a reduced rate typical of the GA102 die.