NVIDIA CEO Talks Titan X, Next-Gen Pascal, Deep Learning And Elon Musk At GTC 2015
NVIDIA CEO Jen-Hsun Huang kicked off the company's 2015 GPU Technology Conference today with his keynote that focused heavily on "deep learning" and analytics that essentially teach computers to think, recognize and react like a human (AI). Huang disclosed a number of details regarding the just-launched GeForce GTX Titan X as well, along with some news regarding the company’s upcoming Pascal GPU architecture and future-looking initiatives in autonomous driving, accompanied by a special guest, Tesla CEO, Elon Musk.
In case you missed out on our full review, we have all of the GeForce GTX Titan X’s juicy details chronicled for you all right here. The card is based on NVIDIA’s 8 Billion transistor, Maxwell-based GM200 GPU, and is packing 3072 CUDA cores and a whopping 12GB of memory. As you’d probably expect, performance is exceptional; the Titan X put up the highest, single-GPU scores we have seen from a graphics card yet.
NVIDIA, however, was tight-lipped on pricing, right up until Jen Hsun spilled the beans during his opening keynote. Although some folks though the Titan X would arrive at a $3000-ish price point, like the Titan Z, NVIDIA is selling the card for $999—just like the original Titan that arrived in 2013.
Though it wasn't specifically noted, it does look like NVIDIA's next-gen Pascal will hit later in 2015
Jen Hsun also discussed a number of features coming with NVIDIA’s next-gen Pascal GPU architecture. Huang noted that Pascal-based GPUs could process workloads up to 10 time faster than current architecture with regard to deep learning applications like machine vision, image recognition, neural networks, etc., due to a handful of specifically targeted features.
NVIDIA Pascal Will Have 3X The Memory Bandwidth Versus Previous Gen Maxwell Arch.
However, high level features were only disclosed, including the fact that Pascal will employ 3D-stacked memory and offer 3x the memory bandwidth of Maxwell. In addition, Pascal will offer four times the FP16 floating point throughput.
Pascal To Offer A 10X Speed-Up In Deep Learning Apps
In addition to having up to 32GB of memory, or 2.7X more than the GeForce GTX TITAN X, Pascal will feature mixed-precision computing capabilities, in addition to its 3D-stacked memory, which alone can result in up to 5X improvement in deep learning applications according to NVIDIA. Pascal will also feature support for NVIDIA’s high-speed interconnect dubbed NVLink, which links together multiple GPUs and can push performance up to a 10X improvement in deep learning with over twice as much bandwidth than the previous generation.
We caught some of the keynote highlights here but we're going to follow-up with some additional coverage of Jen-Hsun's chat with a very special guest. In the meantime, hit play to catch some footage from the floor of GTC 2015...
NVIDIA Digits DevBox With Quad-Titan X Cards
Another notable that Jen-Hsun took time to break out the dog and pony for was the company's new Digits DevBox. This is a beast of a machine with a quad-GeForce GTX Titan X setup that is, according to Huang, "optimized in every way to wring out every last FLOP" from their GPU architecture. The DevBox is architected specifically for the task of "deep learning" in neural networks. Reportedly, the DevBox comes "preinstalled with all the software data scientists and researchers require to develop their own deep neural networks. This includes the DIGITS software package, the most popular deep learning frameworks – Caffe, Theano and Torch – and cuDNN 2.0, NVIDIA’s robust GPU-accelerated deep learning library. "
And then things got even more interesting as Jen-Hsung invited Tesla CEO, Elon Musk to the stage for a discussion of machine learning, autonomous vehicles and machine vision.
NVIDIA CEO Jen-Hsun Huang And Elon Musk Talk Autonomous Cars And AI
Coming from the perspective of a guy that recently was quoted as saying developing real computer AI could be like "summoning the demon," we were anxious to hear Musk's view on the subject of deep machine learning as it relates to every day life, driverless cars, etc. Interestingly, Musk sort of dodged the question when Jen-Hsun asked specifically about that comment, noting only, "it could." However, he carried on further to note that autonomous vehicles, though a difficult technology to master and develop, are not that far off in the future at all. In fact, his opinion was of course, if done right and with proper regulation, they can be both safer than human drivers and possibly inherently more secure.
We'll have more detail on Musk's talk with NVIDIA's Jen-Hsun Huang in a bit of video footage that we'll load up shortly. In the meantime, speaking of cars and because NVIDIA's CEO is so fond of the HOT wheels in general, of course the company had a few Tegra-enabled vehicles on hand to showcase their infotainment and car automation chops. Feast...