GPUDatacenterReleased Apr 12, 2021

NVIDIA A30

24 GB HBM2 · 3072-bit · 3,584 cores

A dual-slot Ampere data center GPU for mainstream enterprise servers, built on a cut-down GA100 die with 24 GB of HBM2. It offers the full Ampere Tensor Core feature set, including FP64 acceleration and Multi-Instance GPU partitioning, within a 165 W budget.

General

Manufacturer: NVIDIA
Model: A30
Segment: Datacenter
GPU chip: GA100
Architecture: Ampere
Release date: Apr 12, 2021
Price: Search on Amazon

Memory

Memory size: 24 GB
Memory type: HBM2
Bus width: 3072-bit
Bandwidth: 933 GB/s
Memory clock: —

Clocks & silicon

GPU clock: —
Boost clock: —
Shader units: 3,584 CUDA Cores
Tensor / AI cores: 224
RT cores: —
Transistors: 54.2 billion
Technology (nm): 7 nm
TSMC N7
TDP / TBP: 165 W

Interface & APIs

Bus interface: PCIe 4.0 x16
SLI / multi-GPU: NVLink 3.0 (200 GB/s)
DirectX: N/A (compute-only)
OpenGL: N/A (compute-only)
Vulkan: N/A (compute-only)
Other APIs: CUDA (compute capability 8.0)

AI support

Precisions: FP8FP16BF16INT8INT4TF32FP64
AI accelerator: Third-generation Tensor Cores (224)
Peak AI TOPS: 661 TOPS
Notes: Third-generation Tensor Cores; FP8 is not supported on Ampere. Peak throughput with structured sparsity: INT4 1,321 TOPS, INT8 661 TOPS, FP16 and BFLOAT16 330 TFLOPS, TF32 165 TFLOPS; dense rates are half. FP64 Tensor Core is 10.3 TFLOPS, FP32 10.3 TFLOPS and FP64 5.2 TFLOPS. Multi-Instance GPU splits the card into up to four 6 GB instances.