GPUDatacenterReleased Nov 9, 2021

NVIDIA A100X

80 GB HBM2e · 5120-bit · 6,912 cores

The A100X is one of NVIDIA's converged accelerators, announced at GTC in November 2021. It puts an A100 80GB Tensor Core GPU and a BlueField-2 DPU on one dual-slot card, joined by an integrated PCIe Gen4 switch so that data from the DPU's two 100 Gb/s Ethernet or InfiniBand ports can reach GPU memory without passing through the host. It targets AI and 5G workloads where networking, storage and security processing are offloaded to the DPU's Arm cores.

General

Manufacturer: NVIDIA
Model: A100X
Segment: Datacenter
GPU chip: GA100
Architecture: Ampere
Release date: Nov 9, 2021
Price: Search on Amazon

Memory

Memory size: 80 GB
Memory type: HBM2e
Bus width: 5120-bit
Bandwidth: 2,039 GB/s
Memory clock: —

Clocks & silicon

GPU clock: —
Boost clock: —
Shader units: 6,912 CUDA Cores
Tensor / AI cores: 432
RT cores: —
Transistors: 54.2 billion
Technology (nm): 7 nm
TSMC N7
TDP / TBP: 300 W

Interface & APIs

Bus interface: PCIe 4.0
SLI / multi-GPU: NVLink bridge (3rd gen)
DirectX: —
OpenGL: —
Vulkan: —
Other APIs: —

AI support

Precisions: FP8FP16BF16INT8INT4TF32FP64
AI accelerator: Tensor Cores (3rd gen, 432)
Peak AI TOPS: —
Notes: Third-generation Tensor Cores support FP64, TF32, FP16, BF16, INT8 and INT4 with 2:4 structured sparsity; FP8 is not supported. Multi-Instance GPU can split the GPU into up to seven isolated instances.