Chinese Manufacturer Intellifusion Introduces DeepEyes AI Boxes With 96 TOPS

In the burgeoning sphere of artificial intelligence, where high-end hardware often dictates market leadership, China’s Intellifusion is carving a niche with its innovative “DeepEyes” AI boxes. Designed to defy GPU restrictions imposed by the US, these units are redefining affordability in AI technology. For approximately 1000 yuan, or around $140, these units offer a robust AI performance of 48 TOPS, a tantalizing proposition for a market hungry for cost-effective solutions.

Navigating the restrictions with finesse, Intellifusion utilizes an older 14nm node, likely integrating an ASIC, a strategic move that underlines China’s resolve to maintain a competitive edge in AI development. The “DeepEyes” series launches into 2024 with its DeepEdge10Max SoC capable of delivering 48 TOPS in int8 precision, a metric crucial for training AI models. Later in the second half of 2024, the Deep Eyes box will upgrade to DeepEdge10Pro, boasting up to 24 TOPS, with a subsequent model, the DeepEdge10Ultra, promising up to 96 TOPS in the first half of 2025.

Though Intellifusion has not released pricing details for these higher-end models, if they can sustain the starting price point long-term, they edge closer to achieving an ambitious target: AI hardware that is 90% cheaper but still capable of handling 90% of scenarios.

Central to the prowess of the “DeepEyes” series is Intellifusion’s custom NNP400T neural network chip. Equipped with a formidable array of components including a 1.8 GHz 2+8 core RISC CPU and a GPU that powers up to 800 MHz in the DeepEdge 10, the standout feature is the efficient NPU. Such a configuration offers a tempting solution within its market segment.

For context, meeting the demands of what Microsoft dubs an “AI PC,” current systems require at least 40 TOPS from an NPU. Intellifusion’s offerings, according to their developmental trajectory, are on the fast track to meet and possibly surpass this requirement. This is particularly notable as most commercially available NPUs peak at only 16 TOPS. However, Intellifusion will have to contend with emerging competition such as Snapdragon’s X Elite chips, which are anticipated to hit 40 TOPS alongside exemplary iGPU performance later in the year.

Dr. Chen Ning, Intellifusion’s chairman, has illuminated the market’s trajectory, underscoring the potential ubiquity of AI: “In the next three years, 80% of companies around the world will use large models… The cost of training a large model is in the tens of millions, and the price of mainstream all-in-one training and pushing machines is generally one million yuan. Most companies cannot afford such costs.” While the claim about the widespread adoption of AI by companies globally may invite skepticism, it underscores an acute industry pain point – the hefty price tag associated with entering the AI realm and the further expense of developing proprietary models.

DeepEdge chips bolster their appeal through embracing “independent and controllable domestic technology,” with a RISC-V core tailoring its capabilities for extensive AI model training and deployment. This innovation isn’t just about raw performance; it’s also about creating an AI images generator that allows companies to expand their capabilities without the prohibitive costs while fostering an environment where the latest ai news & ai tools are accessible to a larger audience.

As industries delve deeper into artificial intelligence generated images and AI video generators, Intellifusion’s DeepEyes AI boxes are poised to democratize AI, offering scalable solutions that empower more companies to join the AI revolution on their own terms.