As generative artificial intelligence (AI) technology becomes more sophisticated, the need to power AI tools is getting a boost. amd has launched what it calls the most advanced GPU for AI, a challenge for Nvidia, which currently dominates the AI chip market. The chip, dubbed the MI300X, will ship later this year and Amazon may be the first customer.
The US-based company aims to tap into the AI market as a replacement for Nvidia’s products. AMD CEO Lisa Su said AI is the company’s “biggest and most strategic long-term growth opportunity.”
“We envision the growth of the data center AI accelerator (market) from about $30 billion this year, at a compound annual growth rate of more than 50%, to more than $150 billion by 2027,” she said.
Why are AI GPUs needed?
GPU chips are needed for companies that offer generative AI programs and tools. “At the heart of this are GPUs. GPUs enable generative AI,” Su told investors, as reported by CNBC.
Large language models (LLMs) for generative AI applications require a lot of memory and computing power due to an increasing number of calculations. AMD demonstrated the MI300x with a 40 billion parameter model called Falcon. To provide more clarity, OpenAI’s GPT-3 model has 175 billion parameters.
According to AMD, the MI300X chip and the CDNA architecture are designed for large language models and other advanced AI models.
It can use up to 192 GB of memory, which essentially means that the GPU can easily run larger models. Nvidia’s rival H100 only supports 120 GB of memory.
“Model sizes are getting much bigger and you actually need multiple GPUs to run the latest major language models,” Su said. She noted that with the added memory on AMD chips, developers wouldn’t need many GPUs.
A ‘good news’ for AMD?
AMD hasn’t disclosed a price, but the launch could pressure Nvidia to rethink its pricing strategy of its AI GPUs, such as the H100, which can cost $30,000 or more. Lower GPU prices mean lower costs for operating generative AI applications.
Initially, AMD will offer an Infinity Architecture that combines eight of its M1300X accelerators into a single system — similar to Nvidia and Google systems that combine eight or more GPUs into a single box for AI applications.
Amazon can use AMD GPU chips
According to a report from Reuters, Amazon Web Services is considering using new AI chips from AMD, but the world’s largest cloud computing provider has yet to make a final decision.
“We’re still working together on exactly where that will land between AWS and AMD, but it’s something our teams are working on,” said Dave Brown, vice president of Elastic Compute Cloud at Amazon.
“There we benefited from some of the work they’ve done around the design that connects to existing systems,” he added.
end of article