Dedicated GPU in the cloud –
NVIDIA T4 VPS on VMware ESXi
AI inference, video encoding, CUDA computing – with a dedicated physical
T4 GPU, no shared resources. From a NATO-certified Hungarian data
center, billed monthly.
- 🖥 Dedicated NVIDIA T4
- ⚙️ VMware ESXi PCI passthrough
- 🛡 NATO-certified DC
- 🇪🇳 Personal support
- 📅 Monthly billing
99,9%
Guaranteed uptime
16 GB
T4 GPU VRAM
320 GB/s
GPU memory bandwidth
24h
Setup time from order
New product
NVIDIA T4 GPU VPS
The NVIDIA T4 Tensor Core GPU delivers 320 GB/s memory bandwidth and 65 TFLOPS FP16 performance. With VMware ESXi PCI passthrough, the card is assigned exclusively to your VM – not shared, not sliced, physically dedicated.
Unlike other providers offering vGPU slicing, we give you full, exclusive access to the physical card. This is a critical difference for AI and ML workloads.
GPU
NVIDIA T4
VRAM
16 GB GDDR6
Performance
65 TFLOPS FP16
Bandwidth
320 GB/s
Virtualization
PCI Passthrough
Platform
VMware ESXi
Use cases
AI / LLM inference
Run Llama, Mistral, Whisper, Stable Diffusion on your own infrastructure
Video encoding
NVENC hardware encoding, 4K/8K transcode, streaming pipelines
Data science
PyTorch, TensorFlow, RAPIDS, Jupyter – CUDA-accelerated computing
Rendering
Blender GPU render, archviz, 3D visualization
Plans
Choose the right fit
All plans run on VMware ESXi, Sun/Oracle and Cisco hardware, in a NATO-
certified Hungarian data center. Prices are net (excl. VAT).
Entry
Vszerver 1
Small websites, dev environments
5 000 Ft/hó
+ VAT
Advanced
Vszerver 2
Webshops, business apps
7 900 Ft/hó
+ VAT
High availability
Vszerver HA
Mission-critical systems, 99.9% uptime
9 200 Ft/hó
+ VAT
Fault tolerance
Vszerver FT
Zero downtime, Fault Tolerance
12 800 Ft/hó
+ VAT
- GPU
Dedicated GPU
Vszerver T4
AI, ML, video encoding, CUDA
Custom
Why vszerver.hu?
NATO-certified data center
The Proserver data center holds "NATO Supply Eligible" certification. Physical security of the infrastructure is guaranteed at the highest level.
Enterprise hardware
Sun/Oracle servers, Cisco networking, VMware ESXi virtualization – enterprise-grade reliability with no unexpected surprises.
Personal support in your language
No ticket system, no chatbot. Real engineers, reachable by phone and email, in Hungarian and English.
Flexible billing
Monthly subscription model, Barion payment, electronic invoicing. No annual commitment, no setup fee.
How to get started
Three steps and your server is running
01
Reach out
Send us a message or call. We discuss your needs – CPU, RAM, storage, GPU requirements.
02
Receive a quote
We send a concrete quote within 24 hours. If it fits, you pay online and sign the agreement.
03
Your server starts
You receive SSH/RDP credentials and VMware Client access. Ready to use immediately.
FAQ
Frequently asked questions
What is the difference between a regular VPS and a GPU VPS?
A regular VPS provides CPU and RAM only. A GPU VPS also includes a dedicated NVIDIA T4 card used exclusively by you – not shared. This enables CUDA-based computing, AI models, and hardware video encoding.
What does "dedicated GPU" mean – isn't vGPU enough?
vGPU splits the physical card between multiple VMs, limiting performance and CUDA compatibility. We use PCI passthrough: the entire physical T4 is assigned to your VM, as if it were physically installed in your machine.
Which operating systems can I run?
Linux (Ubuntu, Debian, CentOS, Rocky Linux) and Windows Server are both available. We choose the OS together during the consultation.
Is there a minimum contract period?
No annual commitment. We operate on a monthly subscription – cancel with one month’s notice when you no longer need the server.
Where is my data physically located?
In the Proserver data center in Hungary, which holds “NATO Supply Eligible” certification. Data does not leave the country.
How do I access the server for administration?
Via SSH (Linux) or RDP (Windows), plus VMware Client with unique credentials. Full root/admin access.
Ready? Write or call us now.
We send a tailored quote within 24 hours. From GPU VPS to basic VPS, any size.