Nvidia Tesla A100 40GB HBM2 5120-Bit PCI-Express 4.0 x16 1x 8-Pin Graphics Card - 699-21001-0200-400

NVIDIA Tesla A100 40GB HBM2 PCIe 4.0 x16 GPU — 699-21001-0200-400

The NVIDIA Tesla A100 40GB HBM2 (699-21001-0200-400) is an alternate SKU variant of NVIDIA's Ampere-architecture A100 PCIe data center GPU accelerator. The 699-21001-0200-400 part number typically denotes an OEM or channel-specific configuration — verify compatibility with your server platform before ordering. Featuring 40GB of HBM2 memory, 1,555 GB/s memory bandwidth, and a PCIe 4.0 x16 interface, it delivers the same transformative AI training, deep learning inference, and HPC performance as the reference A100 PCIe.

Key Specifications

• Part Number / MPN: 699-21001-0200-400
• Manufacturer: NVIDIA
• Product Line: Tesla A100
• Architecture: Ampere (GA100)
• Memory: 40GB HBM2
• Memory Bandwidth: 1,555 GB/s
• Memory Bus Width: 5120-bit
• Interface: PCI-Express 4.0 x16
• Power Connector: 1x 8-pin PCIe
• TDP: 250W
• Cooling: Passive (requires server airflow)
• Form Factor: Full-height, full-length (FHFL) PCIe
• SKU Type: OEM/channel variant — verify server compatibility

Compute Performance

• FP64 (Double Precision): 9.7 TFLOPS
• FP64 Tensor Core: 19.5 TFLOPS
• FP32 (Single Precision): 19.5 TFLOPS
• TF32 Tensor Core: 156 TFLOPS (312 TFLOPS with sparsity)
• BFLOAT16 Tensor Core: 312 TFLOPS (624 TFLOPS with sparsity)
• FP16 Tensor Core: 312 TFLOPS (624 TFLOPS with sparsity)
• INT8 Tensor Core: 624 TOPS (1,248 TOPS with sparsity)
• Memory Bandwidth: 1,555 GB/s

Memory

• Capacity: 40GB HBM2
• Bandwidth: 1,555 GB/s
• Bus Width: 5120-bit
• ECC: Yes — hardware ECC for data integrity

Features

• NVIDIA Ampere architecture with 3rd-gen Tensor Cores
• Structural sparsity support — up to 2x throughput on sparse AI models
• Multi-Instance GPU (MIG) — partition into up to 7 isolated GPU instances
• NVLink 3.0 support (via NVLink Bridge for multi-GPU PCIe configurations)
• PCIe 4.0 x16 — compatible with PCIe 3.0 servers at reduced bandwidth
• Passive cooling — designed for high-airflow data center environments
• Supports CUDA 11+, cuDNN, TensorRT, NCCL, and NVIDIA AI Enterprise

Physical & Environmental

• Form Factor: Full-height, full-length (FHFL) dual-slot PCIe
• TDP: 250W
• Power Connector: 1x 8-pin PCIe
• Cooling: Passive (server airflow required)
• Operating Temperature: 0°C to 35°C (inlet air)

Compatibility

• Compatible with PCIe 4.0 x16 and PCIe 3.0 x16 server slots
• OEM/channel variant — verify specific server platform qualification before ordering
• Requires 250W slot/system power budget per GPU
• Compatible with NVIDIA AI Enterprise, CUDA 11.x/12.x, PyTorch, TensorFlow, JAX, MXNet
• Ideal for AI/ML training, LLM inference, HPC simulation, genomics, financial modeling, and scientific computing