Longsys Cracks 397B AI Model with Custom SPU and iSA Configuration
Chinese memory and storage company Longsys has successfully run a 397B AI model on its custom Ryzen AI Halo, leveraging a novel compression technique that enables real-time data compression. This achievement surpasses the model's estimated memory requirements, sparking interest in the tech community.
Key points
- Longsys demonstrated a localized version of the 397B-parameter AI model on its custom Ryzen AI Halo, featuring a 16-core Ryzen AI Max+ 395 and 128GB of RAM configuration.
- The custom SPU and iSA configuration allows for real-time data compression, enabling the model to run despite exceeding the available 96GB of VRAM.
- The AI model is derived from Alibaba's Qwen 3.5 397B (A17B), a multimodal foundation model that leverages a Mixture-of-Experts (MoE) approach.
- Longsys's achievement has sparked interest in the tech community, with analysts noting its potential impact on AI model deployment and optimization.
Longsys, a Shenzhen-based memory and storage company, has made a significant breakthrough in the field of artificial intelligence (AI) by successfully running a massive 397B AI model on its custom Ryzen AI Halo. This achievement has sparked interest in the tech community, with analysts noting its potential impact on AI model deployment and optimization.
The custom Ryzen AI Halo features a 16-core Ryzen AI Max+ 395 and 128GB of RAM configuration, which is a notable improvement over the original configuration that required a server room full of GPUs. The AI model itself is derived from Alibaba's Qwen 3.5 397B (A17B), a multimodal foundation model that leverages a Mixture-of-Experts (MoE) approach.
The secret to Longsys's success lies in its custom SPU and iSA configuration, which allows for real-time data compression. This technique enables the model to run despite exceeding the available 96GB of VRAM, which is a significant challenge in AI model deployment. Analysts believe that this achievement has the potential to revolutionize the way AI models are deployed and optimized, making it easier to run complex models on smaller hardware configurations.
While the exact implications of Longsys's achievement are still unclear, it is clear that this breakthrough has the potential to impact the tech industry in a significant way. As the field of AI continues to evolve, it will be interesting to see how companies like Longsys continue to push the boundaries of what is possible.
Sources
The WireByte editorial team synthesises technology news from multiple primary sources, verifies the facts, and links every source. Articles are produced with AI assistance and reviewed under our editorial policy.