AI services and LLMs
Run language models, document search and business assistants. We will select video memory and server resources for your model and context length.
Start with your project
Tell us what you want to run. We will help choose resources and agree the environment.
Include the model or application name, data volume and rental period.
NVIDIA Blackwell · professional GPU
Servers with NVIDIA RTX PRO Blackwell and 24 to 96 GB of video memory for AI services, image generation and 3D rendering. Choose an environment for your workload without buying hardware.
For workloads that need a GPU
We start with your workload: applications, data volume and expected results.
Run language models, document search and business assistants. We will select video memory and server resources for your model and context length.
An environment for image generation and processing, computer vision and visual models.
GPU rendering, scene processing and visualisation. We will check application and driver compatibility.
Hardware video encoding and decoding, frame processing and media preparation.
One architecture. Different capabilities.
From a compact AI service to large-model computing. Choose video memory for your workload, then select CPU, RAM and storage separately.
Choose video memory for the model, scene and working buffers.
Hardware AI acceleration, including FP4 support with compatible software.
Ray tracing in applications that use NVIDIA RTX capabilities.
NVIDIA video encoders and decoders for media streams.
Choose GPU and resources
One whole graphics card per configuration. Start with a GPU, then choose server resources.
| Graphics card | Video memory | Memory | Memory bandwidth |
|---|---|---|---|
| RTX PRO 4000Blackwell SFF Edition | 24 GB | GDDR7 · ECC | 432 GB/s |
| RTX PRO 4500Blackwell Server Edition | 32 GB | GDDR7 | 800 GB/s |
| RTX PRO 5000Blackwell · 48 GB | 48 GB | GDDR7 · ECC | 1,344 GB/s |
| RTX PRO 5000Blackwell · 72 GB | 72 GB | GDDR7 · ECC | 1,344 GB/s |
| RTX PRO 6000DBlackwell Server Edition | 84 GB | GDDR7 · ECC | 1,398 GB/s |
| RTX PRO 6000Blackwell Max-Q | 96 GB | GDDR7 · ECC | 1,792 GB/s |
Memory bandwidth is a GPU specification, not your application's processing speed. Results depend on the model, software and workload.
GPU Start suits an initial service and model testing. Increase server RAM and storage for larger datasets.
Send the application name, data volume and use case. We will check compatibility and help choose resources.
Software environment
Use root access to configure the system or order environment preparation. We agree compatible software versions before installation.
NVIDIA drivers and GPU computing tools.
Containers with GPU access through NVIDIA Container Toolkit.
An environment for models and machine learning experiments.
Choose tools for your model and the way users access it.
Blackwell requires compatible drivers and libraries. A GPU model name alone does not guarantee speed: we check application startup, memory use and results on your workload.
When needed
Rental provides infrastructure. Configuration and ongoing support can be arranged separately.
Drivers, Docker, one selected stack and a test run. Up to 3 hours of work; model development is excluded.
122.92 USD one-offSeparate storage for project data. The starting package provides 100 GB total backup capacity; we agree the schedule.
from USD 12.42 /mo.OS updates, monitoring and server environment maintenance within an agreed scope.
Choose support →We agree the scope, storage and recovery terms before connecting.
From enquiry to connection
Application, model, data volume, access requirements and rental period.
We check resources, compatibility and availability, then agree the price and connection date.
After agreement and payment, we provide access. Additional configuration follows your selected package.
Before ordering
Each plan assigns a whole graphics card to the customer's virtual machine. The VM receives its entire video memory; the GPU is not divided into plan shares.
Capacity depends on architecture, weight precision, context length and concurrent requests. Cache and working buffers also use memory. Send your model name and use case so we can check the configuration.
The GPU can support experiments and fine-tuning when the workload fits the server's resources. Training large models from scratch requires a different class of infrastructure.
No. The rental format is a virtual machine with a whole GPU. Plans specify virtual CPU cores, RAM and storage capacity.
Rental is paid monthly in advance. The server and GPU are reserved for the paid period, so stopping an application or VM does not stop billing. Final terms are agreed before connecting.
These are additional services. We agree backup capacity, recovery method and setup scope, including a test run. Developing and training your models is excluded from infrastructure rental.
We confirm the connection date after checking resource availability and environment requirements. Workload testing can be agreed before the main order.