Macnica begins offering AI agents and physical AI testing environments using the NVIDIA DGX B200 system in the AI TRY NOW PROGRAM

Macnica (Head office: Yokohama, Kanagawa Prefecture; Representative Director and President: Kazumasa Hara; hereinafter referred to Macnica) CLAVIS Company has launched the AI TRY NOW PROGRAM, a support program that can be used when considering the introduction of the latest AI solutions on the NVIDIA development environment. ^{* 1}"NVIDIA DGX™ B200^{* 2} Today, we are announcing that we will add a system to provide a pre-test environment for AI agents and physical AI, which are at the forefront of generative AI solutions.

■ Addressing the new AI frontier
Advances in generative AI technology are driving rapid progress in the development of AI agents and physical AI. High-performance computing capabilities are essential in these fields, and expectations are rising for the latest NVIDIA Blackwell architecture GPUs. This GPU's support for FP4 ^{* 3} has attracted particular attention for its ability to provide an efficient learning and inference environment with reduced memory requirements. While an increasing number of companies are eager to adopt these technological innovations, challenges such as a lack of product knowledge and information on implementation benefits required for budget approval continue to hinder companies' rapid adoption of AI.

Accelerating AI Agents and Physical AI with Macnica 's "AI TRY NOW PROGRAM"
For companies facing these challenges, Macnica is adding the NVIDIA DGX B200 system to its "AI TRY NOW PROGRAM," providing an environment that accelerates the verification and testing of AI agents and physical AI using the latest technologies, including NVIDIA NeMo™, NVIDIA NIM™ Microservices, and NVIDIA Cosmos™. For the "AI TRY NOW PROGRAM," the Company engineers have pre-selected the optimal technologies and built the environment, eliminating the need for customers to prepare their own GPU environment. This allows developers to efficiently train and test complex AI models and simulate operational environments, supporting implementation decision-making at unprecedented speed.

■ Use cases for AI agents
As a use case in the environment provided by the "AI TRY NOW PROGRAM," customers can try out AI agents using the NVIDIA NeMo Agent Toolkit and NIM. This allows them to run a series of tests, such as launching an AI agent using the NeMo Agent Toolkit or NIM on the NVIDIA DGX B200, sending prompts via API, and receiving reports from the AI agent.

Example: Test configuration of an AI agent incorporating RAG and Web Search

The NeMo Agent Toolkit mentioned above is a library that allows you to easily implement AI agents by incorporating the LLM, Function (tool), and Workflow you want to use into a configuration defined in a YAML file. It covers the entire lifecycle of an AI agent, from implementation to planning and maintenance, making it a worthwhile solution for both those who are already developing AI agents and those who are thinking about starting. By conducting actual testing, the "AI TRY NOW PROGRAM" will help you speed up the implementation of an AI agent that improves business efficiency, for example.

AI agent for improving business efficiency based on the results of the verification

Use cases of Physical AI
The "AI TRY NOW PROGRAM" also allows users to try out physical AI, which has recently become popular. When considering physical AI, it is possible to train a model using synthetic data (video data) generated by NVIDIA Cosmos and then simulate it in the NVIDIA Omniverse™ environment, enabling pipeline verification.

Example: Verification of NVIDIA Cosmos and Omniverse for Physical AI

Macnica conducted a test using Cosmos-Transfer1 to generate synthetic data for Physical AI model training by adding prompt instructions to video data and segmentation data.

Synthetic data generation using Cosmos-Transfer1

■ Future outlook
As a primary distributor of NVIDIA products, Macnica will continue to develop and support the implementation of new AI technologies, further evolving Japan's generative AI technology and contributing to solving even more problems.

*1: About the "AI TRY NOW PROGRAM"
As an authorized NVIDIA distributor, Macnica not only provides infrastructure but also offers the "AI TRY NOW PROGRAM," a support program that can be used when considering the introduction of the latest AI solutions that run on NVIDIA accelerated computing. the Company engineers have already carefully selected the AI technology that is best suited to solving the problem and have created an environment, so customers do not need to prepare the environment themselves. By deepening their understanding of the latest AI solutions and being able to measure KPIs of performance and functionality that are close to those of a production environment, it is expected that the decision-making process for introduction will be accelerated.

More details here:
https://www.macnica.co.jp/business/semiconductor/manufacturers/nvidia/products/143484/

*2: About the NVIDIA DGX B200 system
The NVIDIA DGX B200 system is a next-generation AI platform equipped with eight NVIDIA B200 Tensor Core GPUs and 1.4TB of large-capacity GPU memory, delivering approximately three times the training performance and up to 15 times the inference performance compared to the previous generation. Optimized for a variety of AI workloads, including large-scale language models, generative AI, and recommendations, the system powerfully supports the entire AI pipeline, from development to production.

More details here:
https://www.macnica.co.jp/business/semiconductor/manufacturers/nvidia/products/7002/

*3: Regarding FP4 (4-bit floating point)
FP4 is a new numerical representation format that enables efficient calculations while reducing data memory usage. Compared to conventional 32-bit or 16-bit floating-point numbers, FP4 allows for more efficient use of computing resources, making it particularly effective for LLM, which requires a large amount of GPU memory. The NVIDIA B200 Tensor Core GPU supports FP4, which can significantly improve the performance of AI model training and inference processing. The introduction of FP4 enables more data to be processed using less memory, a benefit that is particularly evident in real-time applications and edge devices.

[For product inquiries, please contact here]
Macnica CLAVIS Company Company Product Manager
Tel: 045-470-9821

*Company names and product names mentioned in this text are trademarks or registered trademarks of Macnica and each company.
*The information published in the news release (including product price, specifications, etc.) is current as of the date of announcement. Please note that the information may be subject to change without prior notice.