EN EN FR FR
Blog Winston AI @CustomerPanel
UnderHost
RTX 5090 & RTX 4070 TI SUPER • NVIDIA CUDA • BARE METAL GPU

GPU
Dedicated Servers

Enterprise bare metal GPU servers for AI training, LLM inference, rendering, and deep learning. Choose RTX 5090 performance or RTX 4070 Ti SUPER value builds with dedicated CPU, RAM, NVMe storage, and optional AI stack setup.

RTX 5090Top GPU Option
1GbpsNetwork Port
96GBDDR5 System RAM
3.84TBNVMe SSD
50TBBandwidth
24hProvisioning
DEPLOY GPU SERVER VS CLOUD PROVIDERS
POWERED BY NVIDIA NVIDIA RTX NVIDIA CUDA TensorFlow PyTorch JupyterLab Docker Ubuntu 22.04
SERVER CONFIGURATIONS

Choose Your Platform

Choose RTX 5090 performance builds for heavier AI workloads or RTX 4070 Ti SUPER value builds for efficient CUDA acceleration. Intel RTX 4070 Ti inventory is currently out of stock.

Intel INTEL PLATFORM

Dual Intel Xeon E5-2680 v4 + Tesla T4 GPU

Dual Intel Xeon Processor NVIDIA Tesla T4 GPU
CPU 2× Intel Xeon E5-2680 v4
Cores / Threads 56 Logical Cores @ 3.3 GHz
RAM 64GB DDR4
Storage 1TB SSD Storage
GPU 1× NVIDIA Tesla T4 - 16GB GDDR6
Compute 2,560 CUDA cores - 8.1 TFLOPS FP32
AI / Tensor 320 Tensor Cores - Turing inference GPU
Bandwidth 100TB / Unmetered Bandwidth
Network 1Gbps Port
IPv4 5 IPv4 Included
IPv6 /64 IPv6 Subnet Included
CPU Score 27,801
Category Budget
Location Switzerland
$599.95 /month
Intel INTEL PLATFORM

Intel® Core™ i9-7900X - RTX 4070 Ti SUPER

Intel Core i9 Processor NVIDIA RTX 4070 Ti SUPER GPU
CPU 10 Cores / 20 Threads @ 4.3 GHz
RAM 64GB DDR4
Storage 1TB NVMe SSD
GPU 1× NVIDIA GeForce RTX 4070 Ti SUPER (16GB GDDR6X)
Compute 8,448 CUDA cores - up to ~44 TFLOPS FP32
Memory 16GB GDDR6X - 256-bit bus - 672 GB/s bandwidth
AI / Tensor 4th Gen Tensor Cores - up to ~706 AI TOPS
Network 1Gbps Dedicated Port
IPv4 + IPv6 1 IPv4 + IPv6 Included
OS Ubuntu 22.04 LTS Pre-installed
AI Stack CUDA, cuDNN, TensorFlow, PyTorch Ready
CPU Score 21,086
Access Full Root Access
$599.95 /month
AMD Version AMD Version OUT OF STOCK
Amd AMD PLATFORM

AMD Ryzen™ 9 3950X - RTX 4070 Ti SUPER

AMD Ryzen 9 Processor NVIDIA RTX 4070 Ti SUPER GPU
CPU 16 Cores / 32 Threads @ 4.7 GHz
RAM 64GB DDR4
Storage 1TB NVMe SSD
GPU 1× NVIDIA GeForce RTX 4070 Ti SUPER (16GB GDDR6X)
Compute 8,448 CUDA cores - up to ~44 TFLOPS FP32
Memory 16GB GDDR6X - 256-bit bus - 672 GB/s bandwidth
AI / Tensor 4th Gen Tensor Cores - up to ~706 AI TOPS
Network 1Gbps Dedicated Port
IPv4 + IPv6 1 IPv4 + IPv6 Included
OS Ubuntu 22.04 LTS Pre-installed
AI Stack CUDA, cuDNN, TensorFlow, PyTorch Ready
CPU Score 39,251
Access Full Root Access
$599.95 /month
Intel Version Intel Version - OOS ORDER NOW
Amd AMD PLATFORM MORE CORES

AMD Ryzen 9950X + RTX 5090 GPU

AMD Ryzen 9950X Processor NVIDIA GeForce RTX 5090 GPU
CPU AMD Ryzen 9950X
Cores / Threads 16 Cores / 32 Threads @ 4.3 GHz
RAM 96GB DDR5
Storage 3.84TB NVMe SSD
GPU 1× NVIDIA GeForce RTX 5090 GPU
Compute 21,760 CUDA cores - ~104.8 TFLOPS FP32
Memory 32GB GDDR7 - 512-bit bus - 1,792 GB/s bandwidth
AI / Tensor 5th Gen Tensor Cores - up to 3,352 AI TOPS
Bandwidth 50TB Bandwidth
Network 1Gbps Port
IPv4 1 IPv4 Included
Remote Management IPMI / Remote Management Included
CPU Score 66,082
Category Performance
Location Utah, USA
$1,198.95 /month
PERFORMANCE

Built For Demanding AI Workloads

Every hardware and infrastructure decision is optimised for AI, LLM training, and inference throughput.

NVIDIA RTX GPU Options

Choose RTX 5090 performance builds or RTX 4070 Ti SUPER value builds, each with dedicated NVIDIA acceleration for AI, rendering, inference, and CUDA workloads.

GPU Compute Power

Dedicated RTX hardware accelerates model training, inference, rendering, and parallel compute without shared cloud GPU contention.

Up to 96GB DDR5 RAM

High-capacity system memory keeps your data pipeline moving and helps large datasets, preprocessing jobs, and application services run beside GPU workloads.

Up to 3.84TB NVMe SSD

Fast NVMe storage keeps datasets, checkpoints, model weights, and render assets close to the GPU with fewer I/O bottlenecks.

Ubuntu 22.04 + Full Stack

Pre-configured Ubuntu 22.04 LTS with NVIDIA drivers, CUDA, cuDNN, TensorFlow, PyTorch, JAX, JupyterLab, Docker, and the NVIDIA Container Toolkit - start immediately.

Dedicated Bare Metal

100% dedicated resources - no hypervisor, no shared hosts. Choose AMD Ryzen 9950X, AMD Ryzen 9 3950X, or Intel platform availability. All cores, GPU, RAM, and storage are yours exclusively.

HARDWARE SPECS

GPU Hardware Deep Dive

A practical view of the GPU, memory, storage, and network capacity available across the current UnderHost GPU lineup.

gpu0 - COMPUTE
CUDA Cores
8,448
Tensor Cores
264 (4th gen)
RT Cores
66 (3rd gen)
Base Clock
2,340 MHz
Boost Clock
2,610 MHz
FP32 Perf
20 TFLOPS
gpu0 - MEMORY
VRAM
16GB GDDR6X
Memory Bandwidth
672 GB/s
Interface
256-bit
Memory Clock
21 Gbps
PCIe Interface
PCIe 4.0 x16
Power (TBP)
285W
server - SYSTEM
Config
RTX 5090 / RTX 4070 Ti S
GPU Class
NVIDIA RTX
Compute
GPU Accelerated
System RAM
Up to 96GB DDR5
Storage
Up to 3.84TB NVMe
Network
Up to 1Gbps
WHY DEDICATED RTX GPU

The Right GPU for AI Infrastructure

UnderHost GPU servers are built for customers who need predictable bare metal performance, direct root access, and RTX acceleration without shared cloud GPU contention or surprise egress billing.

RTX 5090 or RTX 4070 Ti SUPER

Pick the GPU tier that fits your workload, from value CUDA acceleration to higher-end RTX 5090 performance.

Dedicated CUDA Resources

No shared GPU scheduler. Your selected GPU, CPU, RAM, and NVMe storage stay assigned to your server.

Fast Local Storage

NVMe storage keeps checkpoints, datasets, and model weights close to the compute layer.

Server-Ready Builds

Hardware is selected and assembled for sustained GPU workloads in a managed datacenter environment.

Bare Metal Isolation

Full root access and dedicated hardware make the platform easier to tune, secure, and troubleshoot.

Dedicated GPU acceleration
NVIDIA GPU ACCELERATION

Run CUDA-ready workloads on dedicated NVIDIA GPU hardware, from RTX 5090 performance builds to Tesla T4 inference servers.

CUDA
Ready
RTX/Tesla
Options
Bare Metal
Dedicated
AI SOFTWARE STACK

Pre-Configured. Ready to Train.

Every server ships with a complete AI/ML stack - no setup overhead, no dependency hell. Open your JupyterLab and start training within minutes.

Ubuntu 22.04 LTS

Long-term support base OS - stable, secure, and the recommended platform for NVIDIA GPU drivers and CUDA on bare metal.

CUDA Toolkit

NVIDIA CUDA Toolkit + cuDNN pre-installed - the foundation for all GPU-accelerated compute on NVIDIA hardware.

TensorFlow

Google's open-source ML framework - pre-configured with GPU support, CUDA acceleration, and cuDNN for deep learning pipelines.

PyTorch

Meta's dynamic ML framework. Pre-installed with GPU support - the leading choice for LLM development, research, and production inference.

JAX

Google's NumPy-compatible framework optimised for high-performance numerical computing and automatic differentiation.

JupyterLab

Interactive browser-based IDE for data science and ML experiments - installed and ready with GPU kernel support.

Docker + NVIDIA Toolkit

NVIDIA Container Toolkit enables GPU-accelerated Docker containers - isolate workloads, reproduce environments.

Data Science Tools

Pre-loaded with common Python data science ecosystem: NumPy, Pandas, Scikit-learn, Matplotlib, HuggingFace Transformers.

IDEAL FOR

What Will You Build?

LLM Training & Fine-Tuning

Train or fine-tune large language models. 32GB VRAM fits 7B–13B parameter models in full FP16 precision.

Neural Network Training

CNNs, RNNs, Transformers, GANs - any architecture benefits from dedicated CUDA + Tensor Cores.

AI Inference & Serving

Deploy models with low-latency GPU inference. Run multiple API endpoints simultaneously.

3D Rendering & Visualization

Ray tracing with 66 dedicated RT cores. Scientific visualization, architectural rendering.

Financial Modeling

High-throughput Monte Carlo simulations and quantitative analysis on dedicated hardware.

Video Processing & Transcoding

NVENC/NVDEC hardware acceleration for real-time video transcoding and processing pipelines.

ubuntu@gpu-server:~$

$ nvidia-smi --query-gpu=name,memory.total,driver_version --format=csv

name, memory.total [MiB], driver_version

NVIDIA GeForce RTX 5090, dedicated GPU, latest

NVIDIA GeForce RTX 4070 Ti SUPER, dedicated GPU, latest


$ python3 -c "import torch; print('GPU acceleration ready')"

GPU acceleration ready


$ python3 -c "import tensorflow as tf; print(tf.config.list_physical_devices('GPU'))"

[PhysicalDevice(name='/physical_device:GPU:0', ...)]


$ jupyter lab --ip=0.0.0.0 --no-browser

[JupyterLab] Serving at http://0.0.0.0:8888/lab

[JupyterLab] GPU kernel available ✓

$

COMPARISON

UnderHost vs Cloud Providers

Compare our dedicated GPU servers against major cloud GPU instances. Same or better performance at a fraction of the cost.

UnderHost UnderHost
AWS AWS G5
Azure Azure NCv3
GCP GCP N1
GPU Model
RTX 5090 / RTX 4070 Ti SUPER / Tesla T4
1× NVIDIA A10G
1× NVIDIA T4
1× NVIDIA T4
Total GPU Memory
Up to RTX 5090 class
24GB GDDR6
16GB GDDR6
16GB GDDR6
FP32 Performance
Up to RTX 5090 performance
31.2 TFLOPS
8.1 TFLOPS
8.1 TFLOPS
Monthly Cost
from $599.95 flat
$1,200+ est.
$1,100+ est.
$1,000+ est.
CPU Resources
up to 16 dedicated cores
4 vCPUs (shared)
4 vCPUs (shared)
4 vCPUs (shared)
RAM
up to 96GB DDR5
16GB
16GB
16GB
Storage
up to 3.84TB NVMe SSD
125GB SSD
100GB SSD
100GB SSD
Bandwidth
up to 1Gbps / 50TB
Metered ($/GB)
Metered ($/GB)
Metered ($/GB)
Dedicated Hardware
✓ Full bare metal
✗ Shared host
✗ Shared host
✗ Shared host
AI Stack
✓ Pre-installed
✗ Manual setup
✗ Manual setup
✗ Manual setup

Save up to 2× vs equivalent cloud GPU instances

Flat monthly rate. No per-hour billing surprises. No egress fees. 100% dedicated hardware.

$599.95 /month flat
ADD-ONS

Enhance Your GPU Server

Network Upgrades

1Gbps Shared (included) FREE
1Gbps Dedicated 100TB $25.00/mo
1Gbps Dedicated Unmetered $119.95/mo

Backups & Management

Full Server Management $165.00/mo
2TB Remote Backups from $29.95/mo
Custom GPU Build Quote on request

All Plans Include

Ubuntu 22.04 + AI Stack FREE
CUDA + cuDNN + Drivers FREE
Full Root Access FREE
24/7 Standard Support FREE
1 IPv4 + IPv6 FREE
CUSTOM CONFIG

Need a Custom
GPU Configuration?

We can build GPU servers with different models, higher RAM, additional GPUs, or custom storage configurations. Whether you need an H100, A100, or a multi-GPU cluster - we can design the right solution.

Provide your specs - GPU model, RAM, storage, bandwidth requirements - and we will quote within 24 hours.

REQUEST CUSTOM GPU SERVER
GPU DEDICATED SERVERS FAQ

Got questions?We've got answers!

Everything you need to know about GPU dedicated server hosting for AI and ML workloads.

What AI/ML frameworks are pre-installed?

Every server ships with Ubuntu 22.04 LTS, NVIDIA drivers, CUDA Toolkit, cuDNN, TensorFlow, PyTorch, JAX, Keras, JupyterLab, Docker, and the NVIDIA Container Toolkit - all pre-configured and GPU-enabled. Additional frameworks installed on request.

Can I run multiple users or concurrent jobs?

Yes. Configure multiple user accounts and run concurrent training jobs. We recommend Docker containers or Slurm for resource isolation between projects or team members. Both are supported on our Ubuntu base.

What performance can I expect?

Performance depends on the selected plan. RTX 5090 builds target heavier AI, rendering, and inference workloads, while RTX 4070 Ti SUPER builds provide strong CUDA acceleration at a lower monthly rate. Typical AI workloads can see 10-50x speedup over CPU-only systems.

What bandwidth is included?

Bandwidth depends on the selected plan. The RTX 5090 build includes a 1Gbps port with 50TB bandwidth. RTX 4070 Ti SUPER builds include 1Gbps bandwidth with upgrade options available. No cloud-style egress surprise.

How quickly are GPU servers provisioned?

Standard Intel and AMD configurations are typically online within 24 hours of payment. Custom GPU builds may take 2–3 business days depending on hardware availability.

Are GPU resources dedicated or shared?

These are 100% bare metal dedicated servers - no virtualization, no noisy neighbors. Your selected GPU, CPU cores, RAM, and NVMe storage are exclusively yours.

Can I get a custom GPU server configuration?

Yes. We can source different GPU models, additional GPUs, higher RAM, custom CPU options, and more. Submit a ticket with your requirements and we respond within 24 hours.

Still have questions about GPU hosting?

Our support team is available 24/7 to help you choose and deploy the right GPU dedicated server for your AI and ML workloads.

Contact Support
DEPLOY IN 24 HOURS

Start Training
Tomorrow.

RTX 5090 or RTX 4070 Ti SUPER. Dedicated bare metal hardware. Optional AI stack setup. Flat monthly rate with no cloud billing surprises.

RTX 5090 Available
RTX 4070 Ti SUPER Option
1Gbps / 50TB Option
Up to 96GB DDR5
Flat Rate - No Surprises
24h Provisioning