Orquestro Group and FuriosaAI Expand AI Infrastructure Business Based on Domestic NPU
TECHWORLD ·
✦ AI Summary
Orquestro Group and FuriosaAI are collaborating to expand AI infrastructure business based on domestic NPUs.
Orquestro has expanded the support range of its AI inference operations platform, 'CONCERTO A.I.,' from GPUs to FuriosaAI's second-generation NPU, 'RNGD.'
The two companies plan to verify the performance and stability of RNGD-based AI infrastructure and link the results to customer proofs of concept and actual deployment projects.
Orquestro Group and FuriosaAI are cooperating to expand their AI infrastructure business based on domestic NPUs. The two companies aim to extend the scope of their existing GPU-centered AI inference operations environment to FuriosaAI's second-generation NPU, 'RNGD.'
To this end, the two companies have been verifying compatibility and key functions between RNGD and Orquestro's AI infrastructure software since early this year. They have also been building an Orquestro AI infrastructure environment for FuriosaAI's 'NXT RNGD server.'
In addition, the two companies have gradually implemented deployment, serving, and operations technologies for real-world LLM environments based on RNGD. They also plan to link the results of this technical validation to customer proofs of concept and actual deployment projects.
Orquestro Group announced on the 18th that it had signed a memorandum of understanding, or MOU, on 'NPU-based AI infrastructure and platform technology and business cooperation' at FuriosaAI's headquarters in Gangnam-gu, Seoul. The agreement is intended to link the results of technical validation to customer proofs of concept and actual deployment projects.
Orquestro has expanded the support range of its AI inference operations platform, 'CONCERTO A.I.,' from existing GPUs to RNGD. 'CONCERTO A.I.' is a platform that provides integrated support for AI model deployment, serving, computing resource management, and service monitoring. This expansion is focused on easing the burden of having to build an LLM serving system directly in an NPU environment.
Orquestro has enabled models and accelerator resources to be managed on a single platform, improving the ease of building and operating RNGD-based AI services. To that end, it applied Intelligent Routing and Topology-aware Scheduling functions to 'CONCERTO A.I.'
Intelligent Routing and Topology-aware Scheduling support the configuration of inference paths that take request characteristics and accelerator resources into account. This allows heterogeneous AI accelerators such as GPUs and NPUs to be used according to the characteristics of the service and workload.
This agreement focuses on moving the technical validation carried out so far into actual customer business expansion. To that end, the two companies plan to verify the performance and stability of RNGD-based AI infrastructure across various AI models and workloads, and to develop deployment and operations models by conducting BMTs and proofs of concept for customers.
The two companies will target the public, financial, and enterprise markets, and plan to jointly design and propose RNGD-based infrastructure tailored to each customer's AI services and workload characteristics. They also plan to link BMT and proof-of-concept results to actual deployment projects and secure industry-specific use cases.
In the medium to long term, they will expand the scope of their cooperation to heterogeneous AI infrastructure that uses GPUs and NPUs in parallel. At the same time, they aim to advance operations and monitoring technologies for models, inference runtimes, and AI accelerators, and to create an environment in which the most suitable accelerators can be used according to workload characteristics.
Orquestro will connect the results of this collaboration to its own AI and cloud infrastructure software. The software to be linked includes the server virtualization solution 'CONTRABASS,' the cloud-native operations management platform 'VIOLA,' and the multi- and hybrid-cloud integrated management platform 'OKESTRO CMP.' Based on this, it plans to expand the scope of management for AI computing resources, including NPUs.
The two companies will continue cooperation in the field of 'K-AI infrastructure.' 'K-AI infrastructure' refers to the integration of domestic AI models, AI accelerators, and infrastructure software. The cooperation method is to combine FuriosaAI's RNGD with Orquestro's AI infrastructure technologies. Through this, the two companies plan to address demand from the public, financial, and enterprise sectors for private AI deployments and sovereign AI deployments.
Against this backdrop, FuriosaAI CEO Baek Jun-ho said that as AI adoption spreads in corporate and public-sector environments, the importance of software and infrastructure capabilities for the stable deployment and operation of a wide range of AI workloads is growing along with the performance and power efficiency of AI accelerators. He added that the company will pursue the combination of Orquestro's AI infrastructure software stack with RNGD to implement a full-stack solution capable of operating large-scale AI data centers and token factories, while also expanding commercial deployments.
Kim Min-joon, chairman of Orquestro Group, said the two companies have continued their technical cooperation since early this year and have established a foundation for deploying, serving, and operating AI models on NPUs. He added that, with this agreement as a turning point, the companies will link technical validation to customer proofs of concept and commercialization, and will pursue new AI infrastructure use cases in the public, financial, and enterprise markets.
Source: TECHWORLD · Kim Seung-gi
Original: https://www.epnc.co.kr/news/articleView.html?idxno=407212
References
This article was produced with the help of an automated content generation algorithm.
Source: TECHWORLD
View originalThis article was summarized and organized by BizCrush based on the original article from TECHWORLD. For exact quotations and full details, please refer to the original article.