Model training

Accelerate your model training and build something extraordinary! Leverage Nebius AI infrastructure powered by NVIDIA® H100 GPUs.

GPUs for different workloads

We offer AI-tailored NVIDIA A100 and H100 GPUs in DELTA HGX Baseboards with 8 GPUs connected by NVLink, as well as L40s and other GPUs in the PCIe form factor.

InfiniBand fabric

Get the most of multihost training on thousands of GPUs with full mesh connection. Our latest InfiniBand network supports up to 3.2Tb/s per host.

Marketplace

Leading vendors' AI-specific tools including OS images and Kubernetes® apps will make a perfect workplace for ML engineers.

How to choose GPU for training

V100

Good for dealing with small and middle-size models that do not require BF16 precision support.

Much more affordable than other types of GPU.

А100

Cost effective for single-node training of conventional models.

Great for domains where CNN, RNN models are popular, e. g. computer vision or medical diagnostics.

Н100

Best choice if speed is your top priority.

Perfect for bigNLP, LLM, and all models with Transformer architecture.

H200

The world’s most powerful GPU for supercharging AI and HPC workloads coming soon!

Solution architecture

This set of Nebius AI services will enable you to create an environment and a data pipeline for supervised, semi-supervised, unsupervised or reinforcement learning.

Let’s find the best possible technical solution

If you want to use a specific database or third-party software for your project, our team of solution architects is here to assist you every step of the way.

FAQ & basic terminology

What is model training?

Model training is the process of teaching a machine learning algorithm to recognize patterns and make predictions or decisions based on data. During the training, the algorithm learns from labeled data, adjusting its parameters to minimize the difference between its predictions and the actual labels.