Skip to main content
NVIDIA H100

The Flagship AI Training GPU

80GB HBM3 memory. 3.35 TB/s bandwidth. Transformer Engine for 4x faster LLM training. The H100 is the gold standard for training large language models.

80GB HBM3 Memory
3.35 TB/s Memory Bandwidth
4x Faster LLM Training
900 GB/s NVLink Bandwidth

Built for Large-Scale AI

The H100 accelerates every stage of the ML pipeline.

80GB HBM3 Memory

High-bandwidth memory delivers 3.35 TB/s bandwidth for training large language models and foundation models.

NVLink Interconnect

900 GB/s GPU-to-GPU bandwidth enables efficient multi-GPU training for models that don't fit on a single GPU.

Transformer Engine

Hardware-accelerated transformer computations with automatic mixed precision for faster training.

4x Faster Training

Up to 4x faster large language model training compared to A100, reducing time and cost.

Pre-configured Environment

PyTorch, TensorFlow, CUDA 12, and popular ML frameworks ready to use. Start training immediately.

Expert Support

Our ML infrastructure team helps with environment setup, optimization, and debugging.

Technical Specifications

H100 SXM specifications for reference.

Specification Value
GPU Memory 80GB HBM3
Memory Bandwidth 3.35 TB/s
FP8 Performance 3,958 TFLOPS
FP16 Performance 1,979 TFLOPS
FP32 Performance 989 TFLOPS
NVLink Bandwidth 900 GB/s
TDP 700W (SXM)
Form Factor SXM5 / PCIe

What H100 Excels At

Optimized for the most demanding AI workloads.

Large Language Model Training

Train GPT-style models, LLaMA, and other LLMs efficiently with H100's transformer engine and massive memory bandwidth.

LLM Fine-tuning

Fine-tune foundation models on your data. H100's memory capacity handles large batch sizes and long context lengths.

Distributed Training

Scale to multi-GPU and multi-node training with NVLink for efficient gradient synchronization.

High-Throughput Inference

Deploy large models for production inference with high throughput and low latency.

On-Demand & Reserved Pricing

Flexible pricing for every workload. Reserved instances available for long-term projects.

1x H100 SXM

$2.49 /hr
or $1,495 /mo reserved
  • 1× NVIDIA H100 SXM
  • 80GB HBM3
  • 24 vCPU
  • 240GB RAM
  • 1TB NVMe
  • Pre-configured ML Environment
Get Started

8x H100 SXM

$19.99 /hr
or $11,995 /mo reserved
  • 8× NVIDIA H100 SXM
  • 640GB HBM3
  • 192 vCPU
  • 1.9TB RAM
  • 8TB NVMe
  • Pre-configured ML Environment
Get Started

Need a custom configuration or cluster? Contact us for a quote.

Frequently Asked Questions

What's the difference between H100 SXM and PCIe? +

H100 SXM has higher power (700W vs 350W) and NVLink support for multi-GPU configurations. SXM offers better performance for large-scale training. PCIe is suitable for single-GPU workloads and inference.

How does H100 compare to A100 for training? +

H100 offers up to 4x faster training for large language models thanks to the Transformer Engine, higher memory bandwidth (3.35 TB/s vs 2 TB/s), and improved FP8 support. For LLM workloads, H100 provides significantly better performance per dollar.

What frameworks are pre-installed? +

PyTorch 2.x, TensorFlow 2.x, CUDA 12.x, cuDNN, NCCL, and Hugging Face Transformers are pre-installed. We also provide Docker images with popular ML frameworks.

Can I use H100 for inference? +

Absolutely. H100 excels at high-throughput inference, especially for large language models. The Transformer Engine accelerates both training and inference workloads.

Is reserved pricing available? +

Yes. Reserved instances (1-month, 3-month, annual) offer 20-40% discounts compared to on-demand pricing. Contact our team for a quote.

Ready to Train?

Get Your H100 Instance Today

Talk to our team about your AI workload. We'll help you configure the right setup.