Nvidia GPUs Datasheet

NVIDIA GPUs - Technical Overview & Product Index

NVIDIA Data Center & Enterprise GPUs Technical Datasheet

Complete technical specification matrix, architecture capabilities, and master part number index for NVIDIA Tensor Core Data Center Accelerators and Enterprise Graphics Processing Units.

1. Product Overview & Key Highlights

Next-Generation AI & HPC Acceleration: Powered by NVIDIA Hopper, Ampere, and Ada Lovelace architectures designed to scale Large Language Models (LLMs), deep learning training, and complex scientific simulations.
High-Bandwidth Memory (HBM3 / HBM2e): Industry-leading memory bandwidth (up to 3.35 TB/s) to eliminate data bottlenecks in memory-intensive AI training and real-time inferencing.
Multi-Instance GPU (MIG) Technology: Enables physical partitioning of single GPUs into up to seven fully isolated instances, optimizing hardware utilization for multi-tenant enterprise workloads.
Fourth-Gen NVLink & NVSwitch Integration: Ultra-fast GPU-to-GPU interconnects delivering up to 900 GB/s bidirectional throughput per GPU for seamless scale-up cluster performance.
Enterprise Workstation & Virtualization Support: Full compatibility with NVIDIA vGPU software, CUDA, TensorRT, and Omniverse for CAD, rendering, and enterprise AI workflows.

2. Architecture Comparison (Hopper vs. Ampere vs. Ada Lovelace)

Architecture	Target Workloads	Tensor Cores	Interconnect Technology	Memory Type	Key Capabilities
Hopper (e.g., H100, H200)	Generative AI, Large Language Models (LLMs) & Exascale HPC	4th Gen Tensor Cores w/ Transformer Engine	NVLink 4 (900 GB/s) / PCIe Gen5	HBM3 / HBM3e (Up to 141GB)	DPX Instructions, FP8 Precision, 2nd Gen MIG
Ampere (e.g., A100, A40, A30)	AI Training/Inference, Data Analytics & High-Performance Compute	3rd Gen Tensor Cores w/ TF32 Support	NVLink 3 (600 GB/s) / PCIe Gen4	HBM2e / GDDR6 (Up to 80GB)	Multi-Instance GPU (MIG), Structural Sparsity
Ada Lovelace (e.g., L40S, L4)	AI Inference, Graphics, Virtual Workstations & Video Transcoding	4th Gen Tensor Cores + 3rd Gen RT Cores	PCIe Gen4 x16	GDDR6 with ECC (Up to 48GB)	AV1 Encoding, Shader Execution Reordering (SER)

3. Data Center GPU Models & Hardware Specifications Datasheet

Part Number / SKU	Architecture	Frame Buffer Memory	Memory Bandwidth	Form Factor & Interface	Thermal / Power (TDP)
900-21010-0000-000 (H100 SXM5)	Hopper	80GB HBM3	3.35 TB/s	SXM5 Module / NVLink	700W (Passive/Liquid)
900-11010-0000-000 (H100 PCIe)	Hopper	80GB HBM3	2.0 TB/s	Dual-Slot FHFL PCIe Gen5	350W (Passive)
900-21001-0000-000 (A100 SXM4 80GB)	Ampere	80GB HBM2e	2.0 TB/s	SXM4 Module / NVLink	400W / 500W
900-11001-0000-000 (A100 PCIe 80GB)	Ampere	80GB HBM2e	1.93 TB/s	Dual-Slot FHFL PCIe Gen4	300W (Passive)
900-1G133-0000-000 (L40S)	Ada Lovelace	48GB GDDR6 w/ ECC	864 GB/s	Dual-Slot FHFL PCIe Gen4	350W (Passive)
900-1G183-0000-000 (L4)	Ada Lovelace	24GB GDDR6 w/ ECC	300 GB/s	Low-Profile Single-Slot PCIe Gen4	75W (Passive)

4. Master Part Number Index

Part Number	Product Category	Description Summary
900-21010-0000-000	Data Center GPU	NVIDIA H100 Tensor Core GPU, SXM5 Form Factor, 80GB HBM3 Memory
900-11010-0000-000	Data Center GPU	NVIDIA H100 Tensor Core GPU, PCIe Gen5 Interface, 80GB HBM3 Memory
900-21001-0000-000	Data Center GPU	NVIDIA A100 Tensor Core GPU, SXM4 Form Factor, 80GB HBM2e Memory
900-11001-0000-000	Data Center GPU	NVIDIA A100 Tensor Core GPU, PCIe Gen4 Interface, 80GB HBM2e Memory
900-1G133-0000-000	Data Center GPU	NVIDIA L40S Universal Accelerator GPU, PCIe Gen4, 48GB GDDR6 Memory
900-1G183-0000-000	Inference GPU	NVIDIA L4 Tensor Core GPU, Low-Profile PCIe Gen4, 24GB GDDR6, 75W TDP
900-1G133-0020-000	Enterprise GPU	NVIDIA A40 Tensor Core GPU, PCIe Gen4, 48GB GDDR6 Memory, Active/Passive
NVLK-BRIDGE-2S-2U	Interconnect Accessory	NVIDIA 2-Slot / 2U NVLink Bridge for PCIe Ampere/Ada GPUs