NVIDIA accelerators at scale.
H200, H100, and L4 GPUs on demand. Purpose-built for ML training, LLM inference, 3D rendering, and scientific computing. NVLink and InfiniBand for multi-node scaling.
Latest GPUs
NVIDIA H200
900 GB/s
NVLink 4.0
400 Gbps
InfiniBand
MIG support
Isolation
Machine families
Purpose-built configurations for every workload profile — from web serving to GPU-accelerated ML training.
H200 Instances
Latest-generation NVIDIA H200 Tensor Core GPUs with 141 GB HBM3e memory. Ideal for the largest LLM training runs and real-time generative AI inference.
GPUs
1 – 8 × H200
GPU Memory
Up to 1.1 TB HBM3e
NVLink
5.0
Network
400G InfiniBand
Engineered for AI.
Purpose-built GPU infrastructure from single-GPU inference to multi-node training.
Tensor Core acceleration
4th and 5th gen Tensor Cores with FP8, FP16, BF16, and TF32 for mixed-precision training and inference.
NVLink & InfiniBand
Up to 900 GB/s GPU-to-GPU bandwidth. 400G InfiniBand for multi-node scaling across thousands of GPUs.
Multi-Instance GPU (MIG)
Partition a single GPU into up to 7 isolated instances. Run multiple inference models on one GPU securely.
High-bandwidth storage
Local NVMe SSDs and parallel file systems for training data. No I/O bottleneck for large dataset workloads.
Confidential GPU computing
Hardware-based GPU memory encryption. Train on sensitive data without exposing it to the cloud operator.
Spot GPU instances
Access GPU capacity at up to 70% discount for fault-tolerant training jobs with automatic checkpointing.
Getting started
Launch your first instance in three steps. CLI, console, or API — your choice.
ur compute instances create gpu-training \
--machine-type=a3-highgpu-8g \
--accelerator=type=nvidia-h100,count=8 \
--image-family=deep-learning-vm \
--zone=eu-west1-bAccelerate every workload.
From foundation model training to real-time inference — GPU instances for every scale.
Train foundation models at scale
Multi-node H100/H200 clusters with NVLink and InfiniBand. Train 70B+ parameter models with FSDP, DeepSpeed, or Megatron-LM. Automatic checkpointing for spot preemption.
View tutorialSuggested configuration
8× A3-highgpu-8g · 64× H100 · InfiniBand
Estimate your costs
Create detailed configurations to see exactly how much your architecture will cost. Pay for what you use, down to the second.
Configuration 1
GPU Platform
Compute Shape
Storage & Network
Pricing Strategy
Cost details
GPU prices are hourly but calculated as monthly estimates here. Spot clusters are subject to preemption.
Works seamlessly with
Frequently asked questions
Ready to accelerate?
Launch your first GPU instance in minutes. New accounts receive GPU credits.