Large AI Training
Start with B200, H200 or H100 when model scale, memory and multi-GPU throughput are the main requirements.
NVIDIA cloud GPUs
Choose a GPU to Explore
Each card links to a dedicated page with complete specifications, capabilities, workload guidance, FAQs and provider considerations.
Enterprise AI and HPCNVIDIA H100
A high-end Hopper GPU for large-model training, fine-tuning, inference and accelerated computing.
- Architecture
- Hopper
- Memory
- 80 GB HBM3
Best forView GPU detailsLLM training and high-performance inference
High-memory AINVIDIA H200
A Hopper GPU with 141 GB HBM3e for larger models, long contexts and memory-intensive inference.
- Architecture
- Hopper
- Memory
- 141 GB HBM3e
Best forView GPU detailsLarge-model inference and memory-bound HPC
Blackwell AINVIDIA B200
A Blackwell-generation GPU designed for demanding generative AI training, inference and AI factory workloads.
- Architecture
- Blackwell
- Memory
- 180 GB HBM3e
Best forView GPU detailsLarge-scale training and FP4 inference
Mature AI and HPCNVIDIA A100
A widely available Ampere GPU for AI training, inference, analytics and scientific computing.
- Architecture
- Ampere
- Memory
- 80 GB HBM2e
Best forView GPU detailsAI, analytics and FP64-based HPC
AI and visual computingNVIDIA L40S
A versatile Ada Lovelace GPU for generative AI, rendering, virtual workstations and video workloads.
- Architecture
- Ada Lovelace
- Memory
- 48 GB GDDR6 ECC
Best forView GPU detailsInference, rendering and visual AI
Efficient inferenceNVIDIA L4
A power-efficient, low-profile GPU for AI inference, video processing and high-density deployments.
- Architecture
- Ada Lovelace
- Memory
- 24 GB
Best forView GPU detailsInference, video AI and virtual desktops
At-a-glance comparison
Compare Architecture, Memory and Workload Fit
| GPU | Architecture | Memory | Best for | Details |
|---|---|---|---|---|
| NVIDIA H100 | Hopper | 80 GB HBM3 | LLM training and high-performance inference | View page |
| NVIDIA H200 | Hopper | 141 GB HBM3e | Large-model inference and memory-bound HPC | View page |
| NVIDIA B200 | Blackwell | 180 GB HBM3e | Large-scale training and FP4 inference | View page |
| NVIDIA A100 | Ampere | 80 GB HBM2e | AI, analytics and FP64-based HPC | View page |
| NVIDIA L40S | Ada Lovelace | 48 GB GDDR6 ECC | Inference, rendering and visual AI | View page |
| NVIDIA L4 | Ada Lovelace | 24 GB | Inference, video AI and virtual desktops | View page |
Start by workload
Find a GPU for What You Need to Run
AI Inference
Compare H200 and H100 for large models, L40S for mixed AI workloads and L4 for efficient mainstream inference.
HPC and Analytics
A100 remains a practical option for mature CUDA, analytics and FP64 workloads, while H100 and H200 target higher-end deployments.
Rendering, Video and Vision
L40S is the stronger mixed AI and graphics option, while L4 prioritizes efficiency, video processing and dense deployment.
Ready to Compare Cloud GPU Providers?
After choosing a GPU, compare provider availability, deployment regions, complete instance configurations and rental pricing.
Compare cloud GPU providers