“For this reason AMD cannot compete” The Nvidia Manner writer explains why the AI race is not shut

Nvidia has been main the cost in the case of AI, serving to it rework into one of many largest firms on the planet, mingling with the likes of Apple, Microsoft, and Amazon. In fact, Nvidia continues to take care of its shopper graphics card facet of the enterprise, just lately releasing 4 playing cards within the RTX 50 collection, most just lately the RTX 5070.

Nonetheless, it’s AI that brings within the huge bucks, and Crew Inexperienced has been investing closely in infrastructure to take care of its lead. It doesn’t come as a shock to see 200,000 Nvidia H100 GPUs powering Grok 3. Now, we will get additional perception into Nvidia’s secrets and techniques to success after an interview with The Nvidia Manner writer Tae Kim, who has been in shut contact with the corporate.

Nvidia is effectively forward of the competitors because of its AI servers

Massive firms have been counting on Nvidia GPUs to coach their AI fashions. Nvidia has even teamed up with DeepSeek, with CEO Jensen Huang just lately praising the startup on its “world-class” open-source reasoning mannequin. So, it’s secure to say that it’s forward of the competitors proper now. Kim says it is because Nvidia has the {hardware} and interconnectivity that its rivals don’t.

“Previously, even only a yr in the past, a typical AI server would have eight GPUs, it was a Hopper AI server. Now the present Blackwell AI server that’s delivery proper now could be 72 GPUs in the identical house, it’s like one rack, it weights one and a half tons. For this reason AMD can’t compete, they don’t have a 72-GPU AI server that has all of the interconnects, all of the networking, every part optimized.”

Tae Kim, writer of The Nvidia Manner: Jensen Huang and the Making of a Tech Large

“You may stack a row of those [GPU racks] collectively and construct out a 100,000 GPU cluster,” says Kim; that is what powered earlier variations of the Grok chatbot. Nvidia invested in GPU graphics past gaming and GPUs for AI sooner than its friends, and it has paid off in an enormous approach. Kim continues, “the rivals don’t have CUDA, they don’t have all of the optimized libraries and the bugs ironed out after 10 years of battle testing”. Moreover, Nvidia’s acquisition of Mellanox in 2019 was an enormous increase to its networking division.

AMD has been slowly catching up on the AI entrance. On the gaming facet, its new FSR 4 upscaler is now “a lot nearer in high quality” to DLSS 4, Nvidia’s newest model, with Crew Purple now making use of a neural community very similar to Nvidia. As for AI servers, although, Nvidia remains to be effectively within the lead with its Blackwell structure – AMD’s different is its EPYC Processors.

If you wish to be taught extra about The Nvidia Manner, you’ll be able to examine the e book (or audiobook) on Amazon. Kim additionally talked a bunch about Nvidia CEO Jensen Huang, who he calls “blunt and direct” – in an extremely great way.