Edge AI server solutions for generative AI

24-06-2024 | Advantech | Test & Measurement

Advantech has announced a groundbreaking Edge AI server solution for generative AI, featuring Phison's patented aiDAPTIV+ technology. The AIR-520 Edge AI Server, powered by an AMD EPYC 7003 series processor, incorporates SQ ai100 AI SSDs, NVIDIA RTX GPU cards, an Edge AI SDK, and NVIDIA AI Enterprise to deliver a ready-to-deploy solution.

Generative AI tools such as LLMs transform enterprise knowledge management by automating data organisation, retrieval, and analysis, boosting productivity and improving decision-making. Custom LLMs improve accuracy, while edge training increases data privacy, though it can be more costly. This solution supports LLM fine-tuning with 1-4 GPU cards and SQ ai100 AI SSDs, allowing enterprises to train LLMs cost-effectively while keeping sensitive data secure at the edge.

The company offers four options: AIR-520-L13B/L33B/L70B, and L70B-Plus, tailored for different scales and applications. The L13B is ideal for real-time applications such as chatbots and language translation. The L33B is suited for more complex tasks, improving productivity and innovation in content creation. The L70B excels in sophisticated data analysis and decision-making for specialised domains. Also, the L70B-Plus, equipped with the NVIDIA AI Enterprise software platform, delivers end-to-end, reliable and optimised AI SDKs with long-term support and expert consulting services, providing efficient deployment of business applications.

All solutions include SQ ai100 AI SSDs, which leverage Phison's aiDAPTIV+ technology. These SSDs act as an extension of GPU vRAM, enabling the system to fine-tune LLMs with minimal GPU cards. This approach not only eases the budget barrier but also makes the Edge AI Server more compact than traditional large rack-mount servers. The AIR-520 Edge AI Server has been designed for various edge AI applications. Its size is comparable to a desktop PC and can be rack-mounted with the appropriate accessories. Its low profile permits easy deployment of an edge AI fine-tuning environment, eradicating concerns about space and maintenance.

In addition to LLM fine-tuning capabilities, the company provides an Edge AI SDK with the GenAI Training Studio, preloaded with Llama-2 13B/33B/70B models for applications like chatbots and data analysis. This simplifies and accelerates customer-specific LLM model training and inference evaluation on the AIR-520. Furthermore, its DeviceOn provides OTA software/container updates and remote management, enabling efficient edge AI orchestration and long-term maintenance.

The company's European DMS team offers a wide range of local design and manufacturing services for customers requiring customised solutions.

sebastian_springall.jpg

By Seb Springall

Seb Springall is a seasoned editor at Electropages, specialising in the product news sections. With a keen eye for the latest advancements in the tech industry, Seb curates and oversees content that highlights cutting-edge technologies and market trends.