Generative AI no longer needs to run in hyperscale data centers. Thanks to the Advantech AIR-520 Next-Gen Edge AI Server, that computing power simply comes to the edge of the network, close to the source of the data. This system is designed for applications where large language models and generative AI are executed on-site. That makes it faster, safer, and often more cost-efficient than cloud-based solutions.
Translated with DeepL.com (free version)
The AIR-520 combines server-level performance with the reliability you expect from an edge system. The system is powered by an AMD EPYC 7003 processor with up to 64 cores and offers space for four NVIDIA RTX GPUs. Combined with Phison’s aiDAPTIV+ technology and Advantech’s Edge AI SDK, this creates a platform suitable for training and running models such as Llama-2 in variants of 13, 33, or even 70 billion parameters. And all this without sensitive data leaving the organization.
Translated with DeepL.com (free version)
Powerful hardware for serious AI
The AIR-520 is a 4U edge server built for heavy and constant AI workloads. The configurations offer ample computing power and excellent expansion options.
- up to 64 CPU cores and 768 gigabytes of memory for intensive AI calculations
- four PCIe x16 slots with a 1,200-watt power supply, suitable for GPUs with a thermal power of up to 700 watts
- support for NVIDIA RTX GPUs and compatibility with NVIDIA AI Enterprise
- flexible configurations from L13B to L70B-Plus, depending on the complexity and scale of the LLM
The combination of powerful processors, multiple GPUs, and optimized storage via aiDAPTIV+ AI SSDs creates a platform built for high data throughput and low latency. That’s exactly what’s needed for real-time insights and generative AI at the edge.
Generative AI close to the source
Generative AI at the edge opens up new possibilities for organizations with sensitive data or time-critical processes. A few practical examples illustrate this well:
- inhealthcare, models can analyze patient data locally without the risk of data leaks.
- financial institutions can make decisions within milliseconds with complete control over their data.
- in retail and manufacturing, real-time analyses are possible that identify trends or predict deviations.
By running large language models on-premises, organizations not only reduce their dependence on external data centers, but also gain greater control over performance, privacy, and costs.
How we make AI workable
At Microtron, we see generative AI evolving from hype to valuable technology. We help companies make the transition from powerful hardware to a working solution on the shop floor. That starts with the right configuration, but also includes everything else: a stable software environment, the right cooling, a suitable network architecture, and a secure data flow.
Our experts have extensive experience with edge computing and AI infrastructures, and understand the practical challenges involved in running heavy models outside the data center. Whether it’s training a language model, optimizing inference performance, or integrating an existing AI framework, we make sure the technology does what it’s supposed to do.
The core of AIR-520
The Advantech AIR-520 Next-Gen Edge AI Server brings generative AI right to where the data is generated. With its powerful AMD EPYC processor, support for multiple NVIDIA GPUs, and integration with modern LLMs, the system provides a robust and scalable foundation for organizations that want to run AI locally. Thanks to our expertise in edge computing and AI integration, we can tailor this technology optimally to the customer’s application.
Generative AI no longer needs to reside somewhere in a distant data center. With the AIR-520, it becomes a practical, secure, and efficient solution at the edge of the network.
Want to learn more?
Want to discover how an edge AI server such as the AIR-520 can strengthen your organization?
Please contact our specialists. We are happy to help you find the right solution for your AI challenge.