Sharon AI Launches Cloud Media Tech with On-Demand GPU Service
Menu
NVIDIA A40
The NVIDIA A40 GPU, built on the NVIDIA Ampere architecture, is designed to accelerate demanding visual computing workloads in data centers. Combining second-generation RT cores, third-generation Tensor cores, and 48GB of GDDR6 memory, the A40 delivers high-performance capabilities for rendering, AI training, virtual workstations, and data science workflows. With support for NVIDIA vGPU, NVLink, and enhanced security, the A40 is a versatile solution for modern enterprise environments.
Basic Product Information
Product Name
NVIDIA A40 GPU
Architecture
NVIDIA Ampere
Memory
48GB GDDR6 with ECC
Compute Power
Up to 299.4 TFLOPS (FP16 Tensor Core)
Release Year
2022
Use Cases
Virtual workstations, 3D rendering, AI training, data science, visual computing
Key Advantages
48GB GDDR6 Memory
Large memory capacity for data-intensive workloads such as AI training and 3D design.
Second-Generation RT Cores
Up to 2X faster ray-tracing performance for rendering and visualization tasks.
Third-Generation Tensor Cores
Enhanced AI performance with support for TF32 and structural sparsity for faster training.
Virtualization Ready
Supports NVIDIA vGPU for multiple user access and scalable workloads.
NVLink Support
Scalable up to 96GB memory with NVLink for larger datasets and higher performance.
Specifications
Performance Specifications
CUDA Cores
10,752
RT Cores
84 (Second-Generation)
Tensor Cores
336 (Third-Generation)
Peak FP32 Performance
37.4 TFLOPS
NVDEC | NVDEC (Includes AV1 decode)
1x | 2x
Tensor Core Performance:
FP16
149.7 | 299.4 TFLOPS (with sparsity)
TF32
74.8 | 149.6 TFLOPS (with sparsity)
BFLOAT16
149.7 | 299.4 TFLOPS (with sparsity)
INT8
299.3 | 598.6 TOPS (with sparsity)
INT4
598.7 | 1,197.4 TOPS (with sparsity)
Memory and Bandwidth
GPU Memory
48GB GDDR6 with ECC
Memory Bandwidth
696GB/s
Thermal and Power
Max Power Consumption
300W
Cooling
Passive cooling
Power Connector
8-pin CPU
Board Specifications
Form Factor
Dual-slot (4.4” H x 10.5” L)
Interconnect Interface:
NVIDIA NVLink® (112.5GB/s bidirectional)
PCIe Gen4 x16 (64GB/s bi-directional)
Display Outputs
3x DisplayPort 1.4 (disabled by default, can be enabled via software)
Display Ports
4 x DisplayPort 1.4a
Dimensions
4.4” (H) x 10.5” (L) - dual slot
NVENC/NVDEC
1x NVENC / 2x NVDEC (includes AV1 decode support)
Supported Technologies
Virtual GPU (vGPU)
Supports NVIDIA vPC/vApps, NVIDIA RTX Virtual Workstation, and NVIDIA Virtual Compute Server for multi-user virtualization.
Secure Boot
Hardware root of trust for enhanced security.
NVLink Support
Yes, scalable up to 96GB memory.
Server Compatibility
Supported by a wide range of servers from worldwide OEMs, with backward compatibility for PCIe Gen3 systems.
Additional Features
01 Second-Generation RT Cores
Deliver up to 2X faster ray tracing and denoising performance, making the A40 ideal for rendering, architectural design, and virtual prototyping.
02 Third-Generation Tensor Cores
Accelerate AI model training and inference with TF32, BFLOAT16, and INT8 support.
03 vGPU Support
Enables efficient sharing of GPU resources across multiple virtual machines for scalable enterprise deployments.
04 NVLink
Scalable memory and performance by linking two A40 GPUs, providing up to 96GB of memory for larger datasets.