NVIDIA’s CEO, Jensen Huang, seems to have a finely tuned antenna for predicting and directing the convergence of two key 3D technologies: GPUs and artificial intelligence (AI).
NVIDIA’s current growth spurt is based on big bet that Huang made a few years ago, when he understood that he was in a unique position to help advance AI, machine learning and deep learning. But why wasn’t Intel, maker of the world’s CPUs, ideally positioned in the same way?
Why exactly does AI benefit from GPUs over CPUs?
The reason NVIDIA more than doubled its value as a company this year was based on its knowledge years ago that AI could and would need to leverage the superior floating point parallel computation of graphics processing units (GPUs), especially as they continue to grow in processing power.
GPUs and central processing units (CPUs) are measured in floating point operations per second (FLOPS). CPUs, such as Intel's Core i9 Extreme Edition CPU, can perform a trillion FLOPS. That’s quite a bit for a CPU, but it’s nothing compared to the 112 teraflops that NVIDIA’s new GPU can produce.
GPUs have many more cores than CPUs, and they execute instructions in lockstep. CPUs have fewer cores, but they are of a higher quality and execute instructions independently of each other. Anything that can be processed in parallel, like machine learning algorithms, should be directed to the GPU by design. Applications like Microsoft Excel that are single threaded by design should be processed by the higher-quality single processor, which, in this case, resides in the CPU.
Did you notice that the frequency of AI news stories increased dramatically in 2016 and the first half of 2017?
There have been some interesting announcements about AI defeating human counterparts at games. The last two significant ones were Google’s DeepMind beating the world’s best Go player, and an OpenAI bot winning against the world’s best Dota 2 1v1 players. Dota 2 is a complicated e-sport video game where “agents must learn to plan, attack, trick, and deceive their opponents,” as it is described in the Open AI blog.
But why exactly does AI benefit from GPUs over CPUs?
During his keynote at NVIDIA’s GPU Technology conference last May, Huang followed up a demo of the company’s research into using deep learning—which simulated representations of what a finished rendering might look like while the actual rendering (a computationally intensive task) was being processed—with an introduction of Tesla V100, a GPU with some incredible engineering at the limits of photolithography.
Huang then discussed the Tesla V100, highlighting the prized GPU and talking about some of its unbelievable features.