NVIDIA A30 GPU

Tensor Core Acceleration

Third-generation Tensor Cores deliver up to 165 TFLOPS TF32 performance and up to 330 TFLOPS with sparsity, enabling high-speed matrix operations for deep learning workloads.

Multi-Precision Compute

Supports TF32, BF16, FP16, INT8, and INT4 precisions, optimizing performance for AI training and inference through automatic mixed precision.

High-Bandwidth Memory

24GB HBM2 memory with 933 GB/s bandwidth efficiently handles large datasets, ideal for memory-intensive AI models, simulations, and rendering workloads.

Multi-Instance GPU (MIG)

Allows partitioning of a single NVIDIA A30 GPU into up to four isolated instances (6GB, 12GB, or 24GB), enabling secure multi-tenant workload isolation.

PCIe Gen4 Connectivity

PCIe 4.0 provides up to 64 GB/s bidirectional bandwidth, enabling scalable multi-GPU deployments without requiring NVLink.

Optimized Power Efficiency

165W TDP delivers high performance with lower power consumption, offering up to 5× inference speedup over previous-generation GPUs.

AI Framework Integration

Native support for CUDA, cuDNN, TensorRT, and Triton Inference Server accelerates TensorFlow, PyTorch, and ONNX Runtime workloads.

HPC Precision Support

Provides up to 10.3 TFLOPS FP64 Tensor Core performance for scientific simulations, CFD, and other double-precision HPC workloads.

Architecture & Platform

• GPU Architecture: NVIDIA Ampere data-center architecture
• Compute APIs Supported: CUDA, OpenCL, DirectCompute, OpenACC
• NVLink Support: Third-generation NVLink, up to ~200 GB/s for multi-GPU scaling

Performance (Peak Compute)

• FP64: 5.2 TFLOPS
• FP64 Tensor Core: 10.3 TFLOPS
• FP32: 10.3 TFLOPS
• TF32 Tensor Core: Up to 82 TFLOPS (165 TFLOPS with sparsity)
• FP16 / BF16 Tensor Core: Up to 165 TFLOPS (330 TFLOPS with sparsity)
• INT8 / INT4 Tensor Core: Up to 330 / 661 TOPS (1321 TOPS with sparsity)

Memory

• GPU Memory: 24 GB HBM2
• Memory Bandwidth: 933 GB/s
• ECC Memory: Supported

Interconnect & Expansion

• System Interface: PCIe Gen4 ×16 (64 GB/s)
• NVLink Bandwidth: ~200 GB/s bidirectional (optional)

Multi-Instance GPU (MIG)

• Supports up to 4 independent GPU instances for workload isolation and QoS control

Form Factor & Power

• Form Factor: Dual-slot, full-height full-length (FHFL) PCIe card
• Maximum Power (TDP): ~165 W
• Thermal Solution: Passive cooling (server airflow dependent)

Use Cases

• AI inference and training
• Machine learning and data analytics
• High-performance computing (HPC) workloads
• Enterprise and cloud data center deployments

Ampere Architecture Power

NVIDIA A30 GPU leverages Ampere architecture with third-generation Tensor Cores delivering up to 165 TFLOPS TF32 performance for accelerated AI training and inference.

High-Capacity HBM2 Memory

24 GB HBM2 memory with 933 GB/s bandwidth efficiently handles large datasets for NLP, data analytics, and complex scientific simulations.

Multi-Instance GPU Support

Supports partitioning into up to four isolated GPU instances (6GB, 12GB, or 24GB), enabling secure multi-tenant workloads and predictable performance.

FP64 Tensor Precision

Delivers up to 10.3 TFLOPS FP64 Tensor Core performance, accelerating HPC scientific computing and double-precision simulation workloads.

Optimized Power Efficiency

165W TDP design balances high performance with energy efficiency, making NVIDIA A30 ideal for mainstream data center deployments.

PCIe Gen4 & NVLink Connectivity

64 GB/s PCIe Gen4 interface with optional ~200 GB/s NVLink enables fast data transfers and scalable multi-GPU configurations.

AI Precision Flexibility

Supports TF32, BF16, FP16, INT8, and INT4 precisions, optimizing performance for mixed-precision training and inference workloads.

Enterprise Workload Versatility

Designed for AI inference at scale, data analytics, and HPC applications across virtualized and cloud-based data center environments.