Cards like the TITAN X are overkill for the vast majority of people, even those who routinely play intensive PC games. In its announcement blog post, NVIDIA made no attempt to veil just how excessive the new card is. Describing it as “crazy,” “reckless,” “irresponsible” and “The Ultimate. Period,” the company is very clearly targeting PC enthusiasts.
The new TITAN X originated from a bet between Brian Kelleher, NVIDIA’s top hardware engineer, and Jen-Hsun Huang, the company’s CEO. Kelleher went to visit Huang. He bet a dollar that he could get more than 10 teraflops of computing performance from a single graphics chip, a feat never before achieved. Huang “thought that was crazy” but approved the project nonetheless.
The result is the card that was unveiled today. Boasting 11 teraflops of performance, Kelleher won the bet. Using NVIDIA’s new Pascal graphics architecture pushed to the limit, the TITAN X is the largest and most powerful graphics card NVIDIA has ever built. Its 11 teraflops of raw compute power is derived from a record-breaking total of 3,584 individual CUDA cores running at 1.53GHz.
The original TITAN X, itself something of a legend in PC gaming, had 3,072 CUDA cores running at 1.08GHz. The new card is up to 60 percent faster than the outgoing model, likely to set all-new performance benchmarks in the world’s top games. NVIDIA said it’s also engineered for maximum overclocking, allowing customers to squeeze out even more performance once it’s installed in their own systems.
Over 12 billion transistors power the card. There’s a total of 12GB of GDDR5X memory, running at 480GB/s. For comparison, a typical modern graphics card has 4GB of memory.
NVIDIA said it’s “your call” as to whether it’s gone too far with the TITAN X, creating something where pure performance is the core focus. In modern computing, that’s often no longer the case. Power efficiency and miniaturisation tend to be higher up the agenda than record-breaking transistor counts.
The card was unveiled at Stanford University at an event attended by over 500 academics, researchers and students. The extreme power of the new TITAN makes it suited to supercomputing, even though it’s built as a desktop card. NVIDIA CEO Huang handed the first TITAN X to Baidu’s Andrew Ng in appreciation of the Chinese company’s work in deep learning and supercomputing.
In 2012, Ng helped to pioneer artificial intelligence research by using a series of graphics cards to build a deep learning network of artificial neurons. He then used 10 million YouTube videos to train the system, building one of the first deep learning networks.
Today’s version is 50 times faster than the original but the breakthroughs made using NVIDIA GeForce GPUs helped to set the stage for everything that has come since. Ng will be one of the first to receive a TITAN X, a single desktop graphics card capable of powering supercomputing applications.
“Just as electricity 100 years ago transformed industry after industry after industry, I think AI powered by deep learning will now do the same,” Ng said. “It’s hard to think of an industry that will not be transformed by AI in the next decade. If you’re a machine learning researcher having access to a machine that is 2x as fast means that you are 2x as productive as a researcher.”
“This is insane, I’m actually speechless, I didn’t expect something like that,” NVIDIA said one of the event organiser’s “stammered” after the night ended. “This is one of the craziest product launches ever in the history of technology.”
With the new Pascal-based TITAN X, NVIDIA is once again redefining the limits of graphics card technology. The card is likely to make its way into deep learning networks, data crunching systems and some of the most powerful gaming computers on the market, setting new records across the range of applications it could benefit. The card will go on sale on August 2 in North America and Europe for $1,200.
