If you wondered whether Nvidia would maintain its iron grip on leadership AI compute performance, the new “Blackwell” GPUs unveiled at GTC should emphatically put those doubts to rest. These silicon beasts are simply excessive in the most glorious ways imaginable.
How excessive? Well, consider that each titanic Blackwell die like the flagship B200 model is essentially two distinct, near-reticle-busting GPU chips connected by an ultra-wideband bridge. In layman’s terms, the B200 is a dual-GPU monstrosity measuring roughly double the mouthwatering H100 part Nvidia released just last year.
The performance claims are equally stupendous – up to 30X faster than H100 for certain AI inference workloads, but more conservatively around 2.5-5X higher for training and other tasks. In short, when these chips start shipping later in 2024, they’ll unequivocally become the baddest AI accelerators money can buy.
Nvidia’s Vertical Integration Masterstroke
But as mind-boggling as the Blackwell GPUs themselves are, Nvidia’s masterstroke lies in its unrivaled vertical hardware and software integration providing full-stack AI solutions for enterprises.
The company continues layering networking fabric, CPUs, DPUs, switches, and comprehensive SDKs into optimized, performance-tuned platforms far exceeding mere disaggregated chip sales. NVIDIA increasingly peddles seamless services rather than patchwork products – a key differentiator no upstart can easily match.
Case in point are NVIDIA’s newly announced “NIM” inference microservices, which further extend the company’s subscription ecosystem stickiness and recurring revenue streams. After buying in, you’re drinking NVIDIA’s Kool-Aid for the long haul.
Not that NVIDIA’s stranglehold over the AI silicon game was ever really in doubt. The company’s GTC keynote descended into near rock concert-levels of hype and fan exhilaration rivaling the frenzy incited by the biggest pop culture properties.
AMD, Intel, and that plucky cavalcade of AI startup challengers can keep valiantly jousting at windmills all they want. For the foreseeable future, Nvidia will continue ruling over the AI landscape with an iron GPU fist – armed to the teeth with exorbitant silicon, tightly integrated solutions, and an ever-expanding services empire.