Lambda Reserved Cloud Clusters

NVIDIA A100 clusters in the cloud with
1,600 Gb/s networking

TRUSTED BY FORTUNE 500 COMPANIES
microsoft-1
mit-gray-lockup
amazon-2
google-1
raytheon-1
sony-1

The fastest GPUs on the market, NVIDIA H100s are coming to Lambda Cloud

Sign-up for more info

Our on-demand NVIDIA H100 GPUs are expected early 2023 and our NVIDIA H100 bare metal clusters in Lambda Cloud will be available for reserved contracts shortly thereafter.

PRICING

The best prices for NVIDIA A100 clusters

Save 25-57% compared to AWS on reserved NVIDIA 8x A100 bare metal instances and 90% on egress. With Lambda, you can transfer a 1TB model for under $10. 

  GPU GPU Memory Network Bandwidth (Gbps) 1-Year Price Savings on GPU
AWS p4d.24xlarge NVIDIA 8x A100 40 GB 400 $19.22/hour  
Lambda Reserved Cloud Cluster NVIDIA 8x A100 40 GB 1600 $8.36/hour 57%
NETWORKING SPEED
network-speed-comparison-chart

The fastest network for distributed training

Train large transformer models at top speeds with 1600 Gbps of RDMA inter-node bandwidth. Our networking is 4x faster than AWS and 16x faster than GCP.

pre-configured-ml-logos

Pre-configured for machine learning

Start training your models immediately with pre-configured software, shared storage, and networking for deep learning. All you have to do is choose your GPU nodes and CPU nodes.

LAMBDA SUPPORT

Deep learning expertise

Access a dedicated support team of linux engineers.
Our premium support upgrade includes software support, a two-hour onboarding, and a technical account manager.

  Basic Premium
Price (% of Contract) Included 5% of contract value
Response time Next business day 8 hours
Software support   All 3rd party software installed by Lambda.

Kubernetes not included at this time.
Onboarding (total of 2-hours)  
Priority chat and email routing  
Share Technical Account Manager (TAM)  
FLEXIBLE CONTRACTS

Dedicated GPUs with flexible contract lengths

Reserve a bare-metal cloud cluster for as few as 6 months with options to reserve a cloud cluster for 1-3 years. Deploy your cluster and start training in as few as 12 days or less. 

  VRAM per GPU (GB) vCPUs RAM (GB) Storage (TB) Network Bandwidth (Gbps) GPUDirect RDMA NVLink 6 months 1 year 3 year
8x NVIDIA A100 (40 GB) 40 GB 128 2048 GB 27 TB NVMe SSD 1600 $8.80/hour $8.36/hour $8.10/hour
8x NVIDIA A100 (80 GB) 80 GB 256 2048 GB Minimum 7 TB NVMe SSD 800 - 1600 $11.43/hour  $10.86/hour  $10.52/hour 

All clusters come with a minimum of 4 servers.
Egress fees are $0.009/GB.

Optional add-ons

  Cores vCPUs RAM (GB) Storage (GB) Network Bandwidth (Gbps) Price
CPU VM Instances 32 64 256 GB 500 GB SSD Up to 10 Gbps $1.41/hour
Block storage (sold in 10TB blocks)            $0.04/GB/month