AMD To Make A 30x Powerful, 1.2 Million GPU Supercomputer?

As the world of artificial intelligence (AI) and generative artificial intelligence (GenAI) keeps growing, technology companies and businesses are on the lookout to design, develop and run extremely powerful computer systems AKA supercomputers.

In order to do this, they are required to make use of the most advanced CPUs and GPUs that come with high computational power; infrastructure that’s vital in the training of AI systems and generative AI models.

While the typical number of GPUs used in current supercomputer AI training clusters hovers around 20,000 GPUs, an AMD (Advanced Micro Devices) executive shared “mind-blowing” information about a potential supercomputer in line to be built by AMD, that could eclipse the current best by a large margin.

So, what did the executive say about the supercomputer? Let’s explore!
 

What Does AMD’s Supercomputer Consist Of?

 

• In a recent interview conducted by Timothy Prickett Morgan with The Next Platform, Forrest Norrod, the Executive Vice President and General Manager of Data Center Solutions Business Unit at AMD revealed that AMD had received a request from a customer to build what could only be described as the world’s most powerful supercomputer.

• The supercomputer was to consist of 1.2 million GPUs or a number “in that range”.

• In comparison, current AI training clusters ordinarily consist of a few thousand GPUs connected across several local server racks.

• As of now, the most powerful and largest known supercomputer is Frontier, which possesses 37,888 AMD Radeon Instinct MI250X GPUs.

• Frontier, the first supercomputer to break the exascale barrier, was built at a cost of $600 million.

• AMD’s potential supercomputer in question would effectively be 30 times more powerful than Frontier and could end up costing “very sober people” up to a hundred billion dollars.

• On the other hand, AMD’s shipment of data center GPUs reportedly accounted for just 2% of the total number of shipments. In contrast, NVIDIA’s GPUs consisted of the remaining 98%.

• At the same time, it’s believed that AMD captured 30% of the data center CPU market.

• Furthermore, there are concerns raised regarding the power consumption for such a mammothian supercomputer.

 

What Did AMD’s whatever Say In The Interview?

 

• Timothy Prickett Morgan: Do you think the world can make enough GPUs for the second scenario where everybody wants huge models with large parameter counts?

• Forrest Norrod: I think so. Because, you know, candidly, even for the very largest models you’re talking about, it’s difficult for me to see inference models that are greater than, say, a few racks. Worst case. And I think, because there’s such an incentive, the overwhelming majority of even the largest models are going to fit into a node for inference. But some of the training clusters that are being contemplated are truly mind boggling.

• Timothy Prickett Morgan: What’s the biggest AI training cluster that somebody is serious about – you don’t have to name names. Has somebody come to you and said with MI500, I need 1.2 million GPUs or whatever.

• Forrest Norrod: It’s in that range? Yes.

• Timothy Prickett Morgan: You can’t just say “it’s in that range.” What’s the biggest actual number?

• Forrest Norrod: I am dead serious, it is in that range.

• Timothy Prickett Morgan: For one machine.

• Forrest Norrod: Yes, I’m talking about one machine.

• Timothy Prickett Morgan: It boggles the mind a little bit, you know?

• Forrest Norrod: I understand that. The scale of what’s being contemplated is mind blowing. Now, will all of that come to pass? I don’t know. But there are public reports of very sober people [who] are contemplating spending tens of billions of dollars or even a hundred billion dollars on training clusters.

Do you think AMD’s move to make such a large and powerful supercomputer, when complete, would help it grow its market share in the industry and compete with industry-giant NVIDIA?

Let us know in the comments below!