On-Demand GPU Compute

Powerful GPU compute, ready when your workload is.

Deploy dedicated NVIDIA GPU compute for AI inference, model training, rendering, data science, and accelerated applications. Pay only for the compute time you use, with prepaid billing and full server control.

View GPU plans How hourly billing works

Hourly

Usage billing

Prepaid

GPU Credit

Hibernate

Release compute

Persistent

System storage

GPU workload fabric

One accelerated server. Multiple production paths.

On demand

GPU engine

NVIDIA RTX A6000

48GB VRAM

Current hourly rate

0.77 GPUC/hr

AI inference

Private models and generation.

AI APIs

Accelerated endpoints for apps.

Training

Fine-tuning and data processing.

Rendering

Graphics, video, and visual jobs.

Persistent data

Keep models and outputs.

Hibernate when idle

Release compute, retain data.

GPU Cloud Plans

Choose the GPU your workload actually needs.

Start with affordable GPU compute for development and inference, then move into larger VRAM and stronger accelerators for training, production AI, and demanding compute workloads.

Prepaid GPU billing

Fund once. Runtime draws from GPU Credit.

Usage rates are shown in GPU Credits (GPUC) per hour. Pay for credits in your account currency.

Browse plans

GPU Cloud

GPU Start

A flexible 16GB GPU environment for AI experimentation, smaller models, development workloads, rendering, and GPU-accelerated applications.

NVIDIA RTX A4000

16GB VRAM

0.24 GPUC /hr

Current hourly rate Out of stock

Opens after live price and capacity validation.

4 vCPU

21GB RAM

100GB persistent root storageKept during hibernation. Cleared when the VM is deleted.

Recommended

GPU Creator

48GB of GPU memory for generative AI, professional rendering, larger inference workloads, visual applications, and demanding development environments.

NVIDIA RTX A6000

48GB VRAM

0.77 GPUC /hr

Current hourly rate Available

Deploy GPU

28 vCPU

58GB RAM

100GB persistent root storageKept during hibernation. Cleared when the VM is deleted.

GPU Cloud

GPU Pro

High-performance GPU compute for production inference, accelerated graphics, rendering pipelines, and demanding GPU applications.

NVIDIA L40

48GB VRAM

1.53 GPUC /hr

Current hourly rate Available

Deploy GPU

28 vCPU

58GB RAM

100GB persistent root storageKept during hibernation. Cleared when the VM is deleted.

725GB temporary workspaceCleared on hibernate or delete.

GPU Cloud

GPU AI

80GB of accelerator memory for model training, fine-tuning, large inference workloads, RAG pipelines, data science, and compute-intensive machine learning.

NVIDIA A100

80GB VRAM

2.06 GPUC /hr

Current hourly rate Out of stock

Opens after live price and capacity validation.

28 vCPU

120GB RAM

100GB persistent root storageKept during hibernation. Cleared when the VM is deleted.

750GB temporary workspaceCleared on hibernate or delete.

GPU Cloud

GPU Enterprise

High-end GPU acceleration for demanding model training, transformer workloads, large-scale inference, scientific computing, and production AI systems.

NVIDIA H100

80GB VRAM

3.80 GPUC /hr

Current hourly rate Out of stock

Opens after live price and capacity validation.

28 vCPU

180GB RAM

100GB persistent root storageKept during hibernation. Cleared when the VM is deleted.

750GB temporary workspaceCleared on hibernate or delete.

Prepaid hourly usage

Pay for the compute time you use.

GPU Credit keeps spending bounded. Your balance covers compute and any persistent resources that remain allocated.

Add GPU Credit

Choose an amount and complete payment before deploying.

Choose a GPU

Match VRAM and accelerator class to your workload.

Run your workload

Usage is metered server-side while billable resources remain allocated.

Hibernate compute

Release the accelerator when your compute session is finished.

Power Off is not a billing stop. Hibernate your GPU server when you are finished with compute to release the accelerator and stop GPU compute charges. Persistent storage and retained IP charges may continue.

Workloads

Match the accelerator to the job.

Start with the memory your workload requires, then account for model size, concurrency, training data, and software overhead. No single GPU fits every model.

AI Inference

Serve models and APIs with the VRAM headroom your traffic needs.

Model Training

Run training jobs on high-memory accelerators sized for the model.

Fine-Tuning

Adapt existing models without maintaining permanent GPU capacity.

Generative AI

Build image, video, and multimodal workflows on dedicated GPU compute.

Rendering

Accelerate visual production, graphics pipelines, and render workloads.

Computer Vision

Process visual data for detection, analysis, and production services.

Data Science

Run GPU-accelerated notebooks, analytics, and machine-learning experiments.

GPU Application Hosting

Host applications and APIs that depend on dedicated acceleration.

GPU comparison

Five GPU tiers, compared directly.

Hourly rates and availability follow the same live pricing and capacity data used to decide whether deployment is allowed.

Temporary storage is not persistent.

It is cleared on hibernate or delete. Save important files to persistent storage first.

Swipe table left or right

Plan	GPU	VRAM	vCPU	RAM	Root	Temporary	Hourly price	Recommended workload	Availability
GPU Start	NVIDIA RTX A4000	16GB	4	21GB	100GB	None	0.24 GPUC/hrCurrent hourly rate	AI development	Out of stock
GPU Creator	NVIDIA RTX A6000	48GB	28	58GB	100GB	None	0.77 GPUC/hrCurrent hourly rate	Generative AI	Available
GPU Pro	NVIDIA L40	48GB	28	58GB	100GB	725GB	1.53 GPUC/hrCurrent hourly rate	Production inference	Available
GPU AI	NVIDIA A100	80GB	28	120GB	100GB	750GB	2.06 GPUC/hrCurrent hourly rate	LLM inference	Out of stock
GPU Enterprise	NVIDIA H100	80GB	28	180GB	100GB	750GB	3.80 GPUC/hrCurrent hourly rate	Large language models	Out of stock

Persistent storage keeps the work that matters.

Your persistent system disk remains available through hibernation, but it is removed when the VM is deleted. Storage can continue to incur charges while compute is released.

System files and saved project data remain on persistent storage.
Storage billing is separate from GPU compute billing.
Move long-term data to a separate persistent volume or snapshot before deleting the VM.

Temporary workspace is built for speed, not retention.

GPU Pro, GPU AI, and GPU Enterprise include temporary high-speed storage. It is erased during hibernation or deletion.

Move checkpoints, datasets, outputs, and anything you need later onto persistent storage before hibernating.

Compute control

Power Off pauses the VM. Hibernate releases the GPU.

These controls serve different jobs. Use hibernation when the goal is to stop GPU compute charges.

Power Off

Stops execution inside the virtual machine. The GPU allocation may remain billable, so Power Off must not be treated as a billing stop.

Hibernate

Releases GPU, CPU, and RAM after hibernation is confirmed. GPU compute charges stop, while persistent storage and a retained IP may continue billing.

GPU Cloud FAQ

The details that matter before you deploy.

Straight answers about pricing, storage, capacity, and hibernation.

How does prepaid GPU billing work?

Add GPU Credit before deployment. Server-side metering deducts billable compute and persistent-resource usage from that balance.

Does Power Off stop GPU charges?

Not necessarily. Power Off can leave compute allocated. Hibernate and wait for confirmation when you want to release the accelerator and stop GPU compute charges.

What survives hibernation?

Persistent system storage and attached persistent volumes remain. Temporary high-speed workspace is erased, so move important data before hibernating.

Can I always resume the same GPU immediately?

Resume depends on current capacity for the same GPU tier. If capacity is unavailable, your server remains hibernated and persistent data stays in place.

Which GPU should I choose?

Start with the VRAM required by your model or application, then allow headroom for context, batches, concurrent requests, training data, and framework overhead.

GPU compute on your terms

Choose the GPU. Fund the runtime. Keep control.

Start with the accelerator your workload needs, then hibernate when the compute session is finished.

View GPU plans Add GPU Credit