Author: Pavel Snoj

Join NVIDIA GTC 2025, the world’s premier event for artificial intelligence, high-performance computing, and innovation. Discover the latest breakthroughs, connect with industry experts, and see how cutting-edge AI is solving today’s biggest challenges.

📅 Date: March 17–21, 2025
📍 Location: San Jose, California & Online

Don’t miss exclusive keynotes, hands-on workshops, and networking opportunities with top AI professionals. Register now and shape the future of AI!

🔗 More information: NVIDIA GTC 2025.

Source: DeepSeek-R1 Now Live With NVIDIA NIM | NVIDIA Blog

To help developers securely experiment with DeepSeek-R1 capabilities and build their own specialized agents, the 671-billion-parameter DeepSeek-R1 model is now available as an NVIDIA NIM microservice preview on build.nvidia.com. The DeepSeek-R1 NIM microservice can deliver up to 3,872 tokens per second on a single NVIDIA HGX H200 system.

Developers can test and experiment with the application programming interface (API), which is expected to be available soon as a downloadable NIM microservice, part of the NVIDIA AI Enterprise software platform.

The DeepSeek-R1 NIM microservice simplifies deployments with support for industry-standard APIs. Enterprises can maximize security and data privacy by running the NIM microservice on their preferred accelerated computing infrastructure. Using NVIDIA AI Foundry with NVIDIA NeMo software, enterprises will also be able to create customized DeepSeek-R1 NIM microservices for specialized AI agents.

Read more on DeepSeek-R1 Now Live With NVIDIA NIM | NVIDIA Blog.

NVIDIA Expands AI Workflows With NVIDIA NIM™ and NVIDIA Blueprints

Source: https://blogs.nvidia.com/blog/nim-agent-blueprints/

NVIDIA offers a wide range of software, including NIM (NVIDIA Inference Microservices) and NVIDIA Blueprints, to simplify the deployment of generative AI across industries. NVIDIA NIM™ provides optimized, cloud-native inference microservices for seamless integration of AI models, while NVIDIA Blueprints offer pre-built workflows for faster development and deployment.

These solutions help businesses accelerate AI implementation, reduce infrastructure complexity, and enhance productivity. Whether in the cloud, on-premises, or hybrid environments, NVIDIA’s new AI tools provide flexibility and scalability.

Learn more about NVIDIA Blueprints: NVIDIA AI Workflows.

Source: GPU Memory Essentials for AI Performance | NVIDIA Technical Blog

The NVIDIA blog highlights the critical role of GPU memory capacity in running advanced artificial intelligence (AI) models. Large AI models, such as Llama 2 with 7 billion parameters, require significant amounts of memory. For instance, processing at FP16 precision demands at least 28 GB of memory.

NVIDIA offers high-performance RTX GPUs, such as the RTX 6000 Ada Generation, featuring up to 48 GB of VRAM. These GPUs are designed to handle the largest AI models, enabling local development and execution of complex tasks. Additionally, they come equipped with specialized hardware, including Tensor Cores, which significantly accelerate computations required for AI workloads.

With NVIDIA’s powerful solutions, businesses and researchers can optimize the development and deployment of AI models directly on local devices, opening up new possibilities for advancements in artificial intelligence.

For more details, visit the official NVIDIA blog: developer.nvidia.com.

Interested in learning more about NVIDIA’s powerful solutions? Contact Xenya d.o.o., and we’ll be happy to help you find the right solution for your needs!

Source: An Easy Introduction to Multimodal Retrieval-Augmented Generation for Video and Audio | NVIDIA Technical Blog

Building a multimodal retrieval-augmented generation (RAG) system is challenging. The difficulty comes from capturing and indexing information from across multiple modalities, including text, images, tables, audio, video, and more. In NVIDIA previous post, An Easy Introduction to Multimodal Retrieval-Augmented Generation, authors discussed how to tackle text and images. This post extends this conversation to audio and videos. Specifically, they explore how to build a multimodal RAG pipeline to search information in videos.

Read more on An Easy Introduction to Multimodal Retrieval-Augmented Generation for Video and Audio | NVIDIA Technical Blog.

Source: Creating RAG-Based Question-and-Answer LLM Workflows at NVIDIA

The rapid development of solutions using retrieval augmented generation (RAG) for question-and-answer LLM workflows has led to new types of system architectures. NVIDIA work, using AI for internal operations, has led to several important findings for finding alignment between system capabilities and user expectations. 

NVIDIA found that regardless of the intended scope or use case, users generally want to be able to execute non-RAG tasks like performing document translation, editing emails, or even writing code. A vanilla RAG application might be implemented so that it executes a retrieval pipeline on every message, leading to excess usage of tokens and unwanted latency as irrelevant results are included.

Read more on Creating RAG-Based Question-and-Answer LLM .