Microsoft has unveiled its new Maia 200 pro accelerator chip artificial intelligence (AI) that is three times more powerful than hardware from competitors such as Google and Amazon, company representatives say.
This latest chip will be used in AI inference rather than training, powering systems and agents used to make predictions, provide answers to queries and generate outputs based on new data fed to them.
The new chip provides performance of more than 10 petaflops (1015 floating-point operations per second), Scott Guthrie, executive vice president of cloud and AI at Microsoft, said in a blog post. This is a measure of performance in a supercomputer where the most powerful supercomputers in the world it can achieve performance of more than 1000 petaflops.
The new chip has achieved this level of performance in a category of data representation known as “4-bit precision (FP4)” – a highly compressed model designed to accelerate artificial intelligence performance. The Maia 200 also delivers 5 PFLOPS of performance in 8-bit precision (FP8). The difference between the two is that the FP4 is much more energy efficient but less accurate.
“In practice, this means that a single Maia 200 node can effortlessly run today’s largest models with plenty of room for even larger models in the future,” Guthrie said in a blog post. “This means that the Maia 200 provides 3 times the FP4 performance of the third-generation Amazon Trainium and the FP8 performance of the seventh-generation Google TPU.”
Chips bye
The Maia 200 could potentially be used for specialized AI workloads, such as running larger LLMs in the future. Until now, Microsoft’s Maia chips have only been used in the Azure cloud infrastructure to run large-scale jobs for Microsoft’s own AI services, notably Copilot. But Guthrie noted that there will be “broader customer availability” in the future, signaling that other organizations could use the Maia 200 through the Azure cloud, or that the chips could potentially one day be deployed in standalone data centers or server stacks.
Guthrie said Microsoft boasts 30% better performance per dollar over existing systems by using a 3-nanometer process made by Taiwan Semiconductor Manufacturing Company (TSMC), the most important manufacturer in the worldallowing for 100 billion transistors per chip. This essentially means that the Maia 200 could be more cost-effective and efficient for the most demanding AI workloads than existing chips.
Apart from better performance and efficiency, the Maia 200 has several other features. For example, it includes a memory system that can help keep AI model weights and data local, meaning you need less hardware to run the model. It is also designed for rapid integration into existing data centers.
Maia 200 should allow AI models to run faster and more efficiently. This means Azure OpenAI users such as researchers, developers, and corporations could see better throughput and speed when developing AI applications and using them as GPT-4 in their operations.
This next-generation AI hardware is unlikely to disrupt most people’s daily use of AI and chatbots in the short term, as the Maia 200 is designed for data centers rather than consumer hardware. However, end users could see the impact of Maia 200 in the form of faster response and potentially more advanced features from Copilot and other AI tools built into Windows and Microsoft products.
Maia 200 could also provide performance boosts to developers and researchers using AI inference through Microsoft platforms. This, in turn, could lead to improved deployment of artificial intelligence in large-scale research projects and elements such as advanced weather modeling, biological or chemical systems, and compositions.
Leave a Reply