13 configurations available nowfrom $449/mo

Dedicated GPU server rental. Delivered in under 10 minutes.

Rent whole NVIDIA Blackwell, Hopper and AMD Instinct GPUs for one flat monthly price — never shared, never metered. Root access, NVMe and bandwidth included.

Not sure which GPU? Size your model or check GPU requirements by model.

Configure your server

NVIDIA H200

In stock

141 GB HBM3e · 24 vCPU · 240 GB RAM · 2 TB NVMe · 25 Gbps

Fits models up to 70B in FP8, 32B in BF16Check your model

$3,090/mo

≈ $4.23 per GPU-hour

Delivered in under 10 minutes after payment.

  • Whole GPUsNever sliced or shared
  • One flat priceNo setup or egress fees
  • 99.9% availability SLACredits up to 100%
  • Cancel any monthOne-month minimum
  • Pay with cryptoUp to +50% top-up bonus
  • Engineers 24/7Sizing & setup help

GPU pricing

Dedicated GPU server prices, per month.

Flat monthly prices for whole, dedicated GPUs, with the per-GPU-hour equivalent so you can compare with hourly clouds. Every server includes:

  • Root access
  • Local NVMe
  • IPv4 + IPv6
  • DDoS protection
  • 20–50 TB, no egress fees
  • 24/7 engineers

USD · billed monthly · excl. VAT

8-GPU servers

  • 8× HGX B300 New

    NVIDIA Blackwell UltraBare metal
    8× 288 GB HBM3e 8 TB/s per GPU

    CPU: 2× Xeon 6776P · 128 coresMemory: 4 TB DDR5

    Storage: 8× 3.84 TB NVMeNetwork: 8× 800G ConnectX-8

    Up to 671Bin FP8 · same in BF16
    Limited stock

    $39,900/mo

    ≈ $6.83/GPU-hr

  • 8× HGX B200

    NVIDIA BlackwellBare metal
    8× 180 GB HBM3e 8 TB/s per GPU

    CPU: 2× Xeon 8570 · 112 coresMemory: 2 TB DDR5

    Storage: 8× 3.84 TB NVMeNetwork: 8× 400G ConnectX-7

    Up to 671Bin FP8 · 405B in BF16
    In stock

    $34,900/mo

    ≈ $5.98/GPU-hr

  • 8× HGX H200

    NVIDIA HopperBare metal
    8× 141 GB HBM3e 4.8 TB/s per GPU

    CPU: 2× Xeon 8480+ · 112 coresMemory: 2 TB DDR5

    Storage: 8× 3.84 TB NVMeNetwork: 8× 400G ConnectX-7

    Up to 671Bin FP8 · 405B in BF16
    In stock

    $23,900/mo

    ≈ $4.09/GPU-hr

  • 8× HGX H100

    NVIDIA HopperBare metal
    8× 80 GB HBM3 3.35 TB/s per GPU

    CPU: 2× Xeon 8480+ · 112 coresMemory: 2 TB DDR5

    Storage: 8× 3.84 TB NVMeNetwork: 8× 400G ConnectX-7

    Up to 405Bin FP8 · 235B in BF16
    In stock

    $16,900/mo

    ≈ $2.89/GPU-hr

  • 8× Instinct MI355X

    AMD CDNA 4Bare metal
    8× 288 GB HBM3e 8 TB/s per GPU

    CPU: 2× EPYC 9575F · 128 coresMemory: 3 TB DDR5

    Storage: 8× 3.84 TB NVMeNetwork: 8× 400G RoCE

    Up to 671Bin FP8 · same in BF16
    Limited stock

    $19,900/mo

    ≈ $3.41/GPU-hr

  • 8× RTX PRO 6000 Blackwell

    NVIDIA BlackwellBare metal
    8× 96 GB GDDR7 1.6 TB/s per GPU

    CPU: 2× EPYC 9555 · 128 coresMemory: 1.5 TB DDR5

    Storage: 4× 3.84 TB NVMeNetwork: 2× 100 GbE

    Up to 405Bin FP8 · 235B in BF16
    In stock

    $10,490/mo

    ≈ $1.80/GPU-hr

Single GPU

  • B300 New

    NVIDIA Blackwell UltraWhole-GPU VM
    288 GB HBM3e 8 TB/s per GPU

    CPU: 28 vCPUMemory: 384 GB

    Storage: 3.5 TB NVMeNetwork: 25 Gbps

    Up to 120Bin FP8 · 70B in BF16
    Limited stock

    $5,190/mo

    ≈ $7.11/GPU-hr

  • B200

    NVIDIA BlackwellWhole-GPU VM
    180 GB HBM3e 8 TB/s per GPU

    CPU: 26 vCPUMemory: 256 GB

    Storage: 2.75 TB NVMeNetwork: 25 Gbps

    Up to 120Bin FP8 · 70B in BF16
    In stock

    $4,590/mo

    ≈ $6.29/GPU-hr

  • H200 Popular

    NVIDIA HopperWhole-GPU VM
    141 GB HBM3e 4.8 TB/s per GPU

    CPU: 24 vCPUMemory: 240 GB

    Storage: 2 TB NVMeNetwork: 25 Gbps

    Up to 70Bin FP8 · 32B in BF16
    In stock

    $3,090/mo

    ≈ $4.23/GPU-hr

  • H100 SXM

    NVIDIA HopperWhole-GPU VM
    80 GB HBM3 3.35 TB/s per GPU

    CPU: 24 vCPUMemory: 225 GB

    Storage: 2 TB NVMeNetwork: 25 Gbps

    Up to 32Bin FP8 · same in BF16
    In stock

    $2,190/mo

    ≈ $3.00/GPU-hr

  • RTX PRO 6000 Blackwell

    NVIDIA BlackwellWhole-GPU VM
    96 GB GDDR7 1.6 TB/s per GPU

    CPU: 16 vCPUMemory: 188 GB

    Storage: 1.92 TB NVMeNetwork: 10 Gbps

    Up to 70Bin FP8 · 32B in BF16
    In stock

    $1,390/mo

    ≈ $1.90/GPU-hr

  • L40S

    NVIDIA Ada LovelaceWhole-GPU VM
    48 GB GDDR6 864 GB/s per GPU

    CPU: 16 vCPUMemory: 128 GB

    Storage: 1 TB NVMeNetwork: 10 Gbps

    Up to 32Bin FP8 · 14B in BF16
    In stock

    $890/mo

    ≈ $1.22/GPU-hr

  • GeForce RTX 5090

    NVIDIA BlackwellWhole-GPU VM
    32 GB GDDR7 1.8 TB/s per GPU

    CPU: 16 vCPUMemory: 64 GB

    Storage: 1 TB NVMeNetwork: 10 Gbps

    Up to 14Bin FP8 · 8B in BF16
    In stock

    $449/mo

    ≈ $0.62/GPU-hr

Rack-scale systems

  • Vera Rubin NVL72 Next gen

    NVIDIA RubinRack-scale
    72× 288 GB HBM4 19.2 TB/s per GPU

    CPU: 36× NVIDIA VeraMemory: Up to 54 TB LPDDR5X

    Storage: Storage sized per orderNetwork: NVLink 6 · ConnectX-9

    Frontier scale72-GPU NVLink domain
    Reservations open

    Price on request

    Next-gen allocation

  • GB300 NVL72

    NVIDIA Blackwell UltraRack-scale
    72× 288 GB HBM3e 8 TB/s per GPU

    CPU: 36× NVIDIA GraceMemory: 17 TB LPDDR5X

    Storage: NVMe per tray, sized per orderNetwork: NVLink 5 · 72× 800G

    Frontier scale72-GPU NVLink domain
    On request

    $7,990/GPU·mo

    From · 72 GPUs · 12–36 mo

Several servers or a 6- to 12-month term? Once your first server is active, ask for term or multi-server pricing by ticket from your dashboard. More than 8 GPUs in one NVLink domain: rack-scale systems.

How it works

Rent a GPU server in three steps.

  1. 1

    Choose a server

    Specs and monthly price shown before you pay, and your server delivered in under 10 minutes.

  2. 2

    Pay from your balance

    Top up in BTC, ETH, XMR, USDT or SOL, with up to 50% bonus credit. No setup fee.

  3. 3

    Get root access

    IP and SSH details in your dashboard. Renew, upgrade or cancel every month.

Why OverGPU

Dedicated GPU. Full month. Full control.

Hourly clouds meter every minute and slice what they can. We rent whole GPUs on simple monthly terms, with every spec published and the price fixed.

  • 15GPU configurations, from RTX 5090 to HGX B300
  • 100%dedicated GPUs, never sliced or shared
  • $0setup or egress fees, bandwidth included
  • 24/7support from engineers who run GPU clusters
  • Whole GPUs, never shared

    No vGPU slicing, no MIG partitions, no noisy neighbours. All the memory and compute, all month long.

  • One flat monthly price

    No hourly meter, no setup fee, no surprise egress bill. NVMe, IPv4 and bandwidth are in the price you see.

  • The latest silicon

    Blackwell Ultra B300, B200, H200 and AMD MI355X — the hardware frontier labs train on, without a one-year contract.

  • Full root, your stack

    Ubuntu with drivers, CUDA or ROCm, Docker and the NVIDIA Container Toolkit ready to go. Keep your environment all month.

  • Crypto, with a bonus

    Top up in Bitcoin, Ethereum, Monero, USDT (TRC-20) or Solana and get up to 50% bonus credit. Card payments are coming soon.

  • Engineers, not ticket bots

    Help with sizing, drivers, NCCL and networking is part of the service, around the clock.

Use cases

The right GPU server for your workload.

Pick what you run and get three configurations sized for it, each with its monthly price and live stock.

Pre-train and post-train on 8-GPU nodes

NVLink-connected HGX servers with 400–800G networking. Keep the same machines, data and checkpoints for the whole run: no preemption, no re-queueing.

GPU servers for LLM training
  • 8× HGX B300 Best fit

    Largest memory: 2.3 TB of HBM3e per node

    8× 288 GB HBM3e Limited

    $39,900/mo

  • 8× HGX B200

    1.4 TB of HBM3e and NVLink 5 for 70B+ training

    8× 180 GB HBM3e In stock

    $34,900/mo

  • 8× HGX H200

    Proven Hopper stack with 1.1 TB of HBM3e

    8× 141 GB HBM3e In stock

    $23,900/mo

Fine-tune without watching the meter

From LoRA to full fine-tunes of 70B-class models. Datasets, checkpoints and your environment stay on the server all month.

GPU servers for fine-tuning
  • H200 Best fit

    141 GB: QLoRA on 70B, LoRA on 30B-class models

    141 GB HBM3e In stock

    $3,090/mo

  • H100 SXM

    The value pick for LoRA and QLoRA

    80 GB HBM3 In stock

    $2,190/mo

  • B200

    180 GB: LoRA on 70B models on a single GPU

    180 GB HBM3e In stock

    $4,590/mo

Serve models 24/7 at a flat cost

Run vLLM, SGLang or TensorRT-LLM around the clock. At high utilisation, one monthly price beats paying by the hour.

GPU servers for inference
  • H200 Best fit

    Large KV cache for long contexts

    141 GB HBM3e In stock

    $3,090/mo

  • B200

    180 GB at 8 TB/s for big models and long contexts

    180 GB HBM3e In stock

    $4,590/mo

  • RTX PRO 6000 Blackwell

    96 GB for cost-efficient serving

    96 GB GDDR7 In stock

    $1,390/mo

Image and video generation, unthrottled

ComfyUI, diffusion and video models with plenty of VRAM, and no queue behind other tenants.

GPU servers for generative media
  • RTX PRO 6000 Blackwell Best fit

    96 GB for video models and big batches

    96 GB GDDR7 In stock

    $1,390/mo

  • GeForce RTX 5090

    32 GB Blackwell card for image pipelines

    32 GB GDDR7 In stock

    $449/mo

  • L40S

    48 GB for SDXL and Flux at scale

    48 GB GDDR6 In stock

    $890/mo

Render capacity booked by the month

Blackwell RT cores and 96 GB of GDDR7 for Blender, Redshift, Octane and Unreal, sized for crunch weeks.

GPU servers for rendering & vfx
  • 8× RTX PRO 6000 Blackwell Best fit

    Eight 96 GB GPUs to render frames in parallel

    8× 96 GB GDDR7 In stock

    $10,490/mo

  • RTX PRO 6000 Blackwell

    One 96 GB GPU for look-dev

    96 GB GDDR7 In stock

    $1,390/mo

  • L40S

    Ada RT cores at an entry price

    48 GB GDDR6 In stock

    $890/mo

Fixed budgets for labs and science

Simulations, genomics and scientific computing on monthly terms that match grants and semesters.

GPU servers for research & hpc
  • 8× HGX H100 Best fit

    8× H100 for multi-GPU research jobs

    8× 80 GB HBM3 In stock

    $16,900/mo

  • 8× Instinct MI355X

    288 GB of HBM3e per GPU with ROCm

    8× 288 GB HBM3e Limited

    $19,900/mo

  • H100 SXM

    A single H100 for teaching and prototyping

    80 GB HBM3 In stock

    $2,190/mo

Payments

Rent GPUs with crypto. Get up to 50% bonus credit.

Top up your balance in US dollars with Bitcoin, Ethereum, Monero, USDT (TRC-20) or Solana; your servers and renewals are paid from it automatically. Card payments are coming soon. How paying with crypto works.

Crypto top-ups

Credited once the network confirms.

  • BitcoinBTC
  • EthereumETH
  • MoneroXMR
  • Tether USDTUSDT · TRC-20
  • SolanaSOL
  • CardComing soon
  • Exchange rate locked when you check out
  • From $100 per top-up, credited in US dollars
  • Servers and renewals paid from your balance

Top-up bonus

+5% for every $100, up to +50%.

  • Top up $100+5%$105 credit
  • Top up $250+10%$275 credit
  • Top up $500+25%$625 credit
  • Top up $1,000 or more+50%$1,500 credit
  • Bonus added with your payment
  • Usable for any server or renewal
  • Paid credit is spent first; bonus credit has no cash value
Monthly invoiceExample
NVIDIA H200 · 1× 141 GB
$3,090.00
Setup fee
$0.00
Bandwidth · 20 TB
Included
IPv4 + IPv6
Included
DDoS protection
Included
Egress fees
$0.00
Stopped-server charges
$0.00
Total per month
$3,090.00

Paid from your balance. Every invoice is in your dashboard.

Infrastructure

GPU hosting built like infrastructure, not a marketplace.

No peer-to-peer hosts, no mystery hardware. Our GPU servers run in professional data centers, are monitored around the clock, and every component is published before you order. More on our infrastructure.

  • 99.9% availability SLA

    Power, network and hardware, with service credits if we miss it.

  • Monitored 24/7

    GPU health, temperatures and links watched; failed parts replaced first.

  • Protected by default

    DDoS mitigation on every public IP, root access by SSH key or password.

  • Clean hand-offs

    Disks securely erased between customers. Your data never meets the next tenant.

Spec sheet · 8-GPU server NVIDIA HGX B300 Limited stock
GPUs
8× NVIDIA B300 SXM
GPU memory
2,304 GB HBM3e 288 GB per GPU · 8 TB/s
Interconnect
NVLink 5 1.8 TB/s per GPU
CPU
2× Intel Xeon 6776P 128 cores
System memory
4 TB DDR5
Storage
8× 3.84 TB NVMe + 2× 1.92 TB boot
Networking
8× 800G ConnectX-8 + 2× 100 GbE
Included
IPv4 + IPv6 /64 · 50 TB/mo

$39,900/mo

FAQ

GPU server rental, answered.

Billing, payments, access and guarantees — the short version. Customers get answers from our engineers 24/7, by ticket.

What is OverGPU?

OverGPU rents dedicated NVIDIA and AMD GPU servers for one flat monthly price: whole GPUs, never shared, from a single GeForce RTX 5090 at $449 a month to an 8-GPU HGX B300 server at $39,900 a month, plus rack-scale GB300 NVL72 systems. You pay from a USD balance topped up in Bitcoin, Ethereum, Monero, USDT (TRC-20) or Solana, and servers are delivered in under 10 minutes after payment.

How does monthly billing work?

You pay each month in advance from your prepaid balance. The price is flat: no hourly meter, no setup fee and no charge for a stopped server. Plans renew automatically from your balance and we remind you by email before renewals. Cancel anytime from your dashboard and your server runs until the end of the paid month.

Are the GPUs really dedicated to me?

Yes. Every GPU you rent is a whole physical GPU assigned only to you: no vGPU slicing, no MIG partitions, no time-sharing. 8-GPU servers are bare-metal machines. Single-GPU plans run in a dedicated virtual machine with the full GPU passed through, so 100% of its memory and compute are yours.

Which payment methods do you accept?

Crypto: Bitcoin (BTC), Ethereum (ETH), Monero (XMR), Tether USDT on TRON (TRC-20) and Solana (SOL). You top up a prepaid balance in US dollars, from $100, and every full $100 in one top-up adds 5% bonus credit, up to 50% from $1,000. Card payments are coming soon. Every server payment and renewal comes with an invoice in your dashboard.

How fast will my server be ready?

Single-GPU and 8-GPU servers are delivered in under 10 minutes after payment, in stock or in limited stock.

Do I get full root access?

Yes, full root over SSH. We provide Ubuntu 22.04 and 24.04 images with NVIDIA drivers, CUDA, Docker and the NVIDIA Container Toolkit preinstalled, ROCm images for AMD, and Debian or Rocky Linux on request. You can install anything allowed by our acceptable use policy.

What is included in the price? Are there egress fees?

Every plan includes the GPUs, the listed CPU, RAM and NVMe storage, a public IPv4 address with an IPv6 range, DDoS protection, OS images and 24/7 support. Bandwidth is included with no egress fees: 20 TB per month on single-GPU plans and 50 TB on 8-GPU servers, under fair use.

Can I upgrade or change GPU model later?

Yes, subject to stock: request it from your server page in the dashboard. An upgrade is charged pro rata from your balance once it is in place, and a move to a smaller plan applies at your next renewal. We can help you move data between servers.

What is the minimum term? How do I cancel?

The minimum term is one month. You can cancel at any time from your dashboard before your next renewal date, with no cancellation fee. Your data is securely erased after the end of the paid period. Need a 6 or 12-month term or several servers? Ask us for term pricing.

What uptime do you guarantee?

We commit to 99.9% monthly availability for power, network and hardware on every server, with service credits if we fall short. Failed hardware is replaced as a priority and incidents are published on our status page.

Do you require identity verification?

Most orders only need an account and a funded balance. For large orders, or when fraud and sanctions checks require it, we may ask for verification. If we do, we tell you exactly why and what we need, and we decide quickly.

Get started

Get your GPU server this month.

One H100 or a fleet of HGX B300 nodes: choose the configuration, create your account in the same form and pay the first month from your balance. Your server is delivered in under 10 minutes.

  1. 1

    Choose and set up

    Configuration, system, software and login, in one form.

  2. 2

    Pay from your balance

    Top up in crypto if needed, with up to 50% bonus credit.

  3. 3

    Get root access

    In under 10 minutes after payment.

  • Delivered in under 10 minutes
  • Whole GPUs, never shared
  • Cancel any month, no fee
  • Up to +50% bonus credit on top-ups

Pay only for the servers you order

Deploy a GPU server

Choose a configuration, create your free account in the same form and pay the first month from your balance.

  • Support by ticket, 24/7From your dashboard, for customers with an active server.
  • Crypto top-ups with bonusBTC, ETH, XMR, USDT (TRC-20) and SOL, up to +50%.
  • Delivered in under 10 minutesFrom payment to root access.

Log in to OverGPU

Manage your GPU servers, balance and invoices.

New to OverGPU? Create an account