Реклама партнера — Название партнёра
UNIT.City — місце, де люди працюють... КРАЩЕ! Обирай свій простір просто зараз 👉

NVIDIA's cutting-edge H200 accelerators are available in the De Novo cloud

National cloud and data center operator De Novo announces the availability of NVIDIA H200, the most productive tensor accelerators for large language models. This is the most advanced solution for generative artificial intelligence tasks, which is now available in the Ukrainian cloud. Now De Novo has the largest GPU infrastructure in Ukraine — H200, H100, A100, L40S, L4. It allows businesses to train complex AI models without transferring data abroad.

Leave a comment
NVIDIA's cutting-edge H200 accelerators are available in the De Novo cloud

National cloud and data center operator De Novo announces the availability of NVIDIA H200, the most productive tensor accelerators for large language models. This is the most advanced solution for generative artificial intelligence tasks, which is now available in the Ukrainian cloud. Now De Novo has the largest GPU infrastructure in Ukraine — H200, H100, A100, L40S, L4. It allows businesses to train complex AI models without transferring data abroad.

The key advantage of the H200 is not just «more power», but a revolutionary subsystem using high-speed HBM3e stack memory. Previously, the development of AI was held back by data transfer speed, not processor power. The new cards remove this barrier, allowing for significantly larger amounts of information to be processed for AI models.

Technical specifications in numbers:

  • Memory subsystem bandwidth: 4.8 TB/s (43% faster than its predecessor H100).
  • Memory capacity: 141 GB (76% more than H100).
  • Efficiency: Query processing time (inference) by Llama 3 70B level models is accelerated by up to +50%.

Thanks to the increased memory capacity, large language models (LLMs) can now run on a single graphics card without the need for sharding (dividing the task into parts), which significantly increases performance. The new high-speed NVIDIA NVLink fourth-generation communication interface, with a bandwidth of up to 900 GB/s, allows you to combine up to eight H200 accelerators into a single computing node with a total memory pool of over 1.1 TB. This opens up new horizons, allowing Ukrainian companies to solve extremely complex tasks.

In the field of generative AI (GenAI), the H200 is ideal for building high-accuracy RAG systems that work with closed corporate knowledge bases, for example, to automate legal analysis, medical data processing or technical documentation. Such power is also critical for training large-scale LLMs with billions of parameters and creating high-performance chatbots or recommender systems.

For users working with LLM and inference in development and production environments, the De Novo cloud offers AI Studio, a pre-configured environment for running and optimizing models. AI Studio offers ready-made container environments with support for CUDA, PyTorch, TensorRT, vector databases, and tools for RAG systems. With integrated APIs, developers can quickly deploy models on the H200 without manual infrastructure preparation, focusing directly on the product, not on configuration.

All calculations take place in the secure De Novo cloud (ISO 27001, PCI DSS, KSZI certificates), which guarantees legal data purity. The new cards are fully compatible with popular tools (CUDA, PyTorch, Kubernetes), which allows developers to instantly get a performance boost without rebuilding their IT infrastructure. Thus, H200 in the De Novo cloud is a tool that allows Ukrainian companies to enter a new era of multimodal and generative models and increase competitiveness in the global AI market.

Have important news to share? Message our Telegram bot

Key events and useful links in our Telegram channel

Discussion
No comments yet.