Skip to main content
high

A10 GPU

The A10 balances AI inference and graphics performance in a power-efficient package, making it suitable for medium-scale AI workloads and virtualized environments. It offers solid compute and memory capacity without the higher cost of top-tier GPUs.

VRAM 24GB
CUDA Cores 9,216
Tensor Cores 288
TDP 150W
Process 8nm
From
$0.15/hr
across 6 providers
A10 GPU

Cloud Pricing

Cheapest on Vast.ai 91% below avg
ProviderConfigPrice / hrUpdatedSource
1×
$0.15/hr
5/8/2026
1×
$0.59/hr
5/10/2026
2×
$0.60/hr
5/10/2026
1×
$0.75/hr
5/13/2026
1×
$0.83/hr
5/13/2026
1×
$1.01/hr
5/13/2026
4×
$1.42/hr
5/13/2026
1×
$1.97/hr36mo
5/10/2026
1×
$2.00/hr
5/13/2026
2×
$2.01/hr36mo
5/10/2026
8×
$2.04/hr
5/13/2026
1×
$2.66/hr12mo
5/10/2026
2×
$2.71/hr12mo
5/10/2026
1×
$3.20/hr
5/10/2026
2×
$3.26/hr
5/10/2026
Direct from providerVia marketplace

Prices updated daily. Last check: May 13, 2026

Performance

FP16
125 TFLOPS
FP32
31.2 TFLOPS
BF16
125 TFLOPS
INT8
250 TOPS
Bandwidth
600 GB/s

Strengths & Limitations

Strengths

  • 24GB GDDR6 memory provides substantial capacity for large models and datasets
  • Single-slot form factor enables higher density server configurations
  • 150W TDP offers efficient power consumption relative to performance
  • 288 Tensor cores with multiple precision support (TF32, BFLOAT16, FP16, INT8, INT4)
  • 600GB/s memory bandwidth supports memory-intensive applications
  • NVIDIA vGPU software compatibility enables virtualized GPU sharing
  • PCIe Gen4 support provides 64GB/s interconnect bandwidth

Limitations

  • 150W power draw may require adequate cooling in dense deployments
  • Ampere architecture lacks some newer features found in Ada Lovelace and Hopper generations
  • Graphics-focused positioning may be overkill for pure AI inference workloads
  • Lower compute density compared to dedicated AI accelerators like H100 or newer GB300 series
  • Released in 2021, representing previous-generation technology compared to current offerings

Key Features

NVIDIA Ampere Architecture
288 Tensor Cores with multi-precision support
NVIDIA vGPU software support
PCIe Gen4 connectivity
TF32 Tensor operations
BFLOAT16 precision support
INT4 and INT8 inference optimization
Single-slot FHFL form factor

About A10

The NVIDIA A10 is a dual-purpose GPU built on the Ampere architecture, designed to handle both AI inference workloads and professional graphics applications. Released in 2021, the A10 sits in the high-performance tier of NVIDIA's data center GPU lineup, positioning itself as a versatile solution for hybrid workloads that require both compute and visualization capabilities. With its compact single-slot design and 150W TDP, the A10 offers efficient deployment in space-constrained server environments. The A10 features 24GB of GDDR6 memory with 600GB/s of memory bandwidth, 9,216 CUDA cores, and 288 Tensor cores. It delivers 31.2 TFLOPS of FP32 performance and 125 TFLOPS of FP16 performance through its Tensor cores. The GPU supports multiple precision formats including TF32, BFLOAT16, FP16, INT8, and INT4, with INT8 performance reaching 250 TOPS. Built on an 8nm manufacturing process, the A10 includes PCIe Gen4 connectivity providing 64GB/s of bandwidth. In cloud deployments, the A10 is commonly used for graphics-rich virtual desktop infrastructure (VDI), AI inference workloads, and professional visualization applications. Its support for NVIDIA vGPU software makes it suitable for multi-tenant environments where GPU resources need to be shared across multiple users or applications, particularly in scenarios combining 3D design workflows with AI-accelerated applications.

Common Use Cases

The A10 is well-suited for hybrid workloads that combine professional graphics and AI inference, making it ideal for virtual desktop infrastructure serving CAD applications, architectural visualization, and content creation workflows. Its 24GB memory capacity and Tensor core capabilities support AI inference for computer vision, natural language processing, and recommendation systems at moderate scale. The GPU's support for NVIDIA vGPU software makes it particularly valuable in multi-tenant cloud environments where GPU resources are shared across multiple users running graphics-intensive applications or AI workloads that don't require the full compute power of larger data center GPUs.

Full Specifications

Hardware

Manufacturer
NVIDIA
Architecture
Ampere
CUDA Cores
9,216
Tensor Cores
288
RT Cores
72
Process Node
8nm
TDP
150W

Memory & Performance

VRAM
24GB
Memory Interface
384-bit
Memory Bandwidth
600 GB/s
FP32
31.2 TFLOPS
FP16
125 TFLOPS
BF16
125 TFLOPS
INT8
250 TOPS
Release
2021

Frequently Asked Questions

How much does an A10 cost per hour in the cloud?

A10 pricing varies by provider, region, and commitment level. Check the pricing table above for current rates across all providers.

What is the A10 best used for?

The A10 excels at hybrid workloads combining professional graphics and AI inference, including virtual desktop infrastructure for CAD applications, architectural visualization, moderate-scale AI inference, and multi-tenant environments requiring GPU virtualization through NVIDIA vGPU software.

How does the A10 compare to newer GPUs for AI workloads?

While the A10's 288 Tensor cores and 24GB memory handle AI inference well, newer architectures like Ada Lovelace and Hopper offer improved efficiency and features. The A10's strength lies in its dual-purpose design for graphics and AI, whereas dedicated AI accelerators like H100 or GB300 series provide higher compute density for pure AI workloads.