Skip to main content
Uptime99.99%
BillingPer-second
Locations8

Rent GPUs On-Demand

Rent NVIDIA RTX 4090, A100 80GB, and V100 32GB GPUs by the hour for AI training, inference, and generative media. No long-term contracts, no minimum spend. Launch your first GPU instance in minutes.

Launch ConsoleView Full Pricing

Platform

Built for AI Workloads

Everything you need to train models, run inference, and iterate on GPU-intensive projects without managing hardware.

Persistent Storage

Attach NVMe volumes that persist across sessions. Stop an instance, restart it later, and pick up exactly where you left off.

Fast Spin-Up

Go from zero to a running GPU instance in under a minute. Pre-built images for PyTorch, vLLM, and ComfyUI are ready to launch.

Datacenter-Grade GPUs

All GPUs are hosted in Tier 3 facilities with redundant power, redundant ISP connections, and backup parts on-site.

Reservations

Lock in capacity with 1-month or 3-month reservations at reduced rates. Convert from hourly to reserved at any time.

Container Mode

Run Docker containers directly on GPU hardware. Use familiar CLI tools, pull any image, and deploy multi-container workloads.

Full Virtual Machines

Get a dedicated VM with root access, your own kernel, and full IOMMU-enforced GPU isolation. No shared resources, no noisy neighbors.

GPUs by Workload

Find the Right GPU for Your AI Workload

Pick the GPU that fits your workload. Each model has a dedicated page with specs, real benchmark numbers, and live pricing.

Consumer-Class GPUs

24 GB GDDR6X · Ada Lovelace

Rent RTX 4090

Cost-efficient AI training, fine-tuning, and diffusion workloads.

View specs and benchmarks

32 GB GDDR7 · Blackwell

Rent RTX 5090

~2× the LLM throughput of the 4090 at a slightly lower cost per million tokens.

View specs and benchmarks

96 GB GDDR7 · Blackwell

Rent RTX PRO 6000

Single-GPU host for ~100B-parameter models — strong vs H100 on cost.

View specs and benchmarks

Datacenter & Specialty GPUs

48 GB GDDR6 ECC · Ada Lovelace

NVIDIA L40S

Multi-tenant inference, virtual workstations, and vGPU-ready workloads. Strong choice for vision and generative media at lower hourly cost than HBM-class datacenter GPUs.

View specs and pricing

288 GB HBM3e · CDNA 4

AMD Instinct MI350X

Largest VRAM on the roster — 288 GB HBM3e at 8 TB/s. Ideal for very-large models, MoE inference, HPC workloads, and deployments that need to avoid NVIDIA lock-in.

View specs and pricing

80 GB HBM2e · Ampere

NVIDIA A100 80GB

The proven datacenter workhorse — 2 039 GB/s HBM2e, up to 7 MIG partitions, and NVLink 3.0 from $1.05/hr. Built for large-model training and high-throughput inference.

View specs and pricing

32 GB HBM2 · Volta

NVIDIA V100 32GB

Cost-leader for memory-bound workloads — 900 GB/s HBM2 and strong FP64 from $0.29/hr. Sustainable second-life silicon for fine-tuning, HPC, and batch inference.

View specs and pricing

Pricing

GPU Rentals at a Fraction of Hyperscaler Prices

Run AI training and deep learning workloads without the high costs of cloud platforms. Transparent, pay-as-you-go pricing.

RTX 4090

$0.39/hr

24 GB GDDR6X VRAM (Ada Lovelace)
16 384 CUDA cores + 512 Tensor Cores
1.0 TB/s memory bandwidth
14 vCPU cores · 47 GB system RAM
300 GB NVMe SSD
1 IPv4 · 10 Gbps network
Rent now

A100 80GB

$1.05/hr

80 GB HBM2e VRAM (Ampere SXM4)
6 912 CUDA cores + 432 Tensor Cores
2.0 TB/s memory bandwidth
30 vCPU cores · 210 GB system RAM
860 GB NVMe SSD
NVLink · 1 IPv4 · 10 Gbps network
Rent now

V100 32GB

$0.25/hr

32 GB HBM2 ECC VRAM (Volta SXM3)
5 120 CUDA cores + 640 Tensor Cores
900 GB/s memory bandwidth
8 vCPU cores · 85 GB system RAM
860 GB NVMe SSD
NVLink 2.0 · 1 IPv4 · 10 Gbps network
Rent now

Global Availability

Launch GPUs Where You Need Them

Choose from datacenters across the US, UK, Italy, and Central Asia. Minimize latency, meet data residency requirements, and keep GPU availability even when your preferred region is full.

GPU Rental FAQ

Common Questions About GPU Rentals

Yes. CloudRift is built for AI: rent NVIDIA RTX 4090, A100 80GB, and V100 32GB GPUs by the hour for LLM training, inference, fine-tuning, and generative media. RTX 5090, RTX PRO 6000, L40S, and AMD MI350X are also in the catalog — availability varies, so check the console for live stock. Pre-built images for PyTorch and vLLM let you start a job in under a minute.
On-demand hourly pricing starts at $0.29 for the V100 32GB, $0.39 for the RTX 4090, and $1.05 for the A100 80GB. Other catalog GPUs are listed at $0.60 for the RTX 5090, $0.63 for the L40S, $1.29 for the RTX PRO 6000, and $3.65 for the AMD MI350X, subject to availability. Reserved 1-month and 3-month commitments cut the rate further. See live pricing in the table above.
The RTX 4090 (24 GB) is the cost-leader for fine-tuning and diffusion. The RTX 5090 (32 GB) delivers about 2× the LLM throughput. The RTX PRO 6000 (96 GB) is a single-card host for ~100B-parameter models and competes directly with the H100 on cost. Compare full specs and benchmarks on the dedicated GPU pages.
Yes. Multi-GPU nodes are available with up to 8 GPUs per instance for data-parallel or pipeline-parallel training. Available configurations vary by datacenter — see the global availability map below.
No. On-demand hourly billing has no minimum commitment. If you want a discount, opt into a 1-month or 3-month reservation; you can convert from hourly at any time.
Open the console, create an instance, pick a GPU and a region, choose VM or container deployment, and launch. Most instances are running in under a minute.
Scale with us

Need Dedicated GPU Infrastructure?

On-demand rentals are the starting point. When you are ready to scale, talk to us about sovereign AI deployments with dedicated infrastructure, custom SLAs, and enterprise support.

Request DemoLearn More