On-demand access to NVidia GPUs

NVIDIA HGX H100

NVIDIA HGX H200

NVIDIA GB200 NVL72

Instance

8X NVIDIA H100

8X NVIDIA H200

72X NVIDIA GB200

GPU/Memory

80 GB of VRAM/GPU

141 GB of VRAM/GPU

192GB

RAM

32x 64 GB DDR5

32x 96 GB DDR5

17280 GB

VCPUs

2x 32 Core

2x 48 Core

2592

Storage

8x 1.1 TB

4x 3.8 TB

276TB NVMe

Bandwidth

400 GB/s InfiniBand

400 GB/s RoCE

Up to 130TB/s

Access the best GPUs on the market

Our pricing is simple. Pay once, get free updates for life

NVIDIA HGX H100

Improve training performance by 4x

Experience 4x better training performance compared to A100 Tensor Core GPUs.

Decrease overall training time

H100 chips provide the ability to train a GPT-3 LLM workload in just minutes.

Get more output from your GPUs

Offload processing tasks and free up GPUs for training, experimentation, and more.

NVIDIA HGX H200

Experience unparalleled performance

Experience up to 1.9x higher performance, compared with H100 Tensor Core GPUs.

Get ultra-fast interconnect

Get 3200Gbps of NVIDIA Quantum-2 InfiniBand networking for low-latency connectivity between GPUs.

Optimize your GPU compute

NVIDIA BlueField-3 DPUs offload networking and storage tasks, increasing GPU utilization for model building.

NVIDIA GB200 NVL72

Enhance performance

Compared to H100 GPUs, see 4x higher training performance, along with 30x faster real-time trillion parameter LLM inference.

Increase compute access

Get access to 72 NVIDIA Blackwell GPUs, and 36 NVIDIA Grace CPUs — all through a single server.

Unlock the power of megaclusters

Connect tens of thousands of NVIDIA Blackwell systems through a single site that unlocks the power of 100K+ GPU megaclusters.

Get started now

Access hundreds of cutting-edge Nvidia H100 GPUs

For Developers →

For Enterprise →