Twitter Logo Facebook Logo Reddit Logo
1 item added to cart
Found in:

What you should know about NVIDIA Volta and the Titan V

If you’re a little like us here at Unlocked, you’re probably salivating over the recently released NVIDIA Titan V GPU. And if you’re a lot like us here at Unlocked, you’re probably also running a cost-benefit analysis of selling a kidney to afford one, so let’s take a closer look at what makes the new card so appealing.

Aside from scoring points for its gold paintjob, the Titan V packs the best specs we’ve ever seen on a consumer-market GPU. In fact, its potential maximum performance level is so high, NVIDIA has said on record that the Titan V is not meant to be just a gaming graphics card.

The Titan V has 5120 CUDA cores, which is considerably more than even the powerful GTX 1080Ti. Its also got a whopping 12GB of HMB2 memory, and uses a 12nm fabrication process to pack over twenty-one-billion transistors onto the chip.

Most important is that the “V” in “Titan V” stands for Volta, the nomenclanture for NVIDIA’s latest GPU microarchitecture. To truly understand what makes the Titan V a force to be reckoned with, you first have to understand what Volta is.

Volta Microarchitecture

With Volta, NVIDIA keeps to its scheme of naming its technologies for well known scientists, physicists, and mathematicians. The Volta architecture is named for Alessandro Volta, an 18th century Italian scientist known for pioneering work in the field of electrochemicals and is credited with creating one of the first batteries. He also lends his name to the concept of voltage, and the Volt in the International System of Units.

As I said before, Volta is NVIDIA’s latest-generation GPU microarchitecture. A microarchitecture is, on a very basic level, a conceptual pathing arrangement for data flowing to and from a given processor. There has been some slight dilution of the term to refer to the design of any data-handling component within a computer, though this is not entirely accurate.

The key point to keep in mind when discussing ultra-high-end modern GPUs is their intended applications. There are two Volta products available right now, the first of which is Titan V GPU. The second is the Tesla V100 GPU Accelerator, which is intended for enterprise and educational sector data center applications. Consumer market hardware rarely (if ever) supports the physical interconnect to take full advantage of the processing power of the V100, which is known as NVLink. The takeaway here is that neither of these products are actually really intended for gaming, and indeed, it’ll probably be many years before any well-optimized gaming title (looking at you, PUBG) could even begin to challenge these GPUs.

But if NVIDIA is making GPUs so powerful that no game could even take advantage of them, why should average consumers care?

The answer lies with an application known as General-Purpose Computing on Graphics Processing Units, or, mercifully, GPGPU for short. The short version is this: While initially GPUs were only designed and intended to strictly handle graphics data, someone figured out that vastly different types of datasets could essentially be converted to graphics data, which in turn meant that GPUs could be leveraged to handle a variety of workloads and tasks beyond graphical rendering. This was taken to the extreme with NVIDIA’s CUDA Core (Compute Unified Device Architecture,) which gave programmers direct access to the processing power offered by the GPU card.

Volta is the ultimate expression of a microarchitecture optimized for a variety of cutting-edge GPGPU workloads. Data science, robotics and self-driving cars, and machine learning are not only in the realm of possibility, but are well into reality due to Volta’s incredible processing power.  NVIDIA has made that last one, machine learning, a central focus recently. All Volta units come equipped with some number of NVIDIA’s new Tensor Cores, which are processors optimized to handle the arithmetic central to machine learning and neural networks.

Machine learning, of course, leads to the eventual realization of artificial intelligences, which NVIDIA claims can be game-changing for almost any industry. Neural-Network AI has the potential to enhance medical and intelligence imagery processing, market trend prediction, and even protein-unwrapping, to name just a few. Any one of those topics could be the subject of a huge essay by themselves, but they all share one commonality: They require enormous amounts of number-crunching power. In supercomputing terms, that arithmetic is measured in Floating Point Operations per Second, or FLOPS. The concept of Floating Points might be one of the most esoteric mathematical concepts in computing, but the extremely basic explanation of a floating point is a number format involving a representative real number and an exponential identifier. If that doesn’t even make sense, don’t worry; people usually need to get degrees to really understand this stuff.

The point is, in order to actually power an AI, you need equipment the can perform operations involving Floating Point Numbers trillions of times per second, or TFLOPS. The V100 and Titan V can each compute at above 100 TFLOPS (for comparison, a GTX 1080 Ti can deliver between 11-12 TFLOPS), which means as far as AI is concerned, Volta might have what it takes to actually deliver.


The Titan V and the future

So the Titan V isn't intended as a card for gamers. Who is it actually for?

The Titan V isn't bad for gaming. It’s more like current-generation games aren’t good enough for the Titan V, and you'd be paying a lot more than you need to for the performance you'd be getting Another obvious application could be cryptocurrency mining, though there’s no hard data on hashrate to power efficiency yet.

Part of the target market here is whatever demographic would view a gold plated GPU as a worthy status symbol. Dedicated makers and amatuer inventors working on AI projects might find it appealing, as would smaller machine learning startups who can’t fit a DGX-1 platform into their budgets (the DGX-1 is NVIDIA’s self-contained enterprise-level AI server.) These are worthy uses, but none of these are really a factor for the average PC user.

Gamers should be far more interested in what Volta could mean for the future. Let's speculate a bit about what that might look like, as cards as powerful as the Titan V become more commonplace.

We’re already at the point where games are closing in fast on photorealism, at which point the question would be one of where next to direct technological progression for the industry. Volta may have already answered that with its deep focus on AI. If Nvidia 11 series consumer cards are Volta based (as opposed to continuing on with Pascal architecture), then it stands to reason that an increasing number of PCs will have GPUs with AI-optimized Tensor cores. This potentially opens up a whole new direction to go: One with an extreme emphasis on non-player character AI.

Imagine enemies that get harder not because they gain more health or accuracy, but because at each difficulty level they become less and less restrained and are actually allowed to attempt to outthink the player in organic ways. What if one day soon we have a Halo game where Cortana’s lines aren’t just at scripted intervals, but fluid and dynamic, actually based upon what’s happening around the player? Or maybe the next Elder Scrolls RPG leverages Tensor power to finally have NPCs react in a logical way? Maybe they'll even be smart enough to stop trying to mug Jimmy the Dragonborn, Arch-Mage of the College of Winterhold, Legate of the Imperial Legions, and Member of the Circle of the Companions.

We’re standing on the precipice of the AI Era, and that comes with a whole slew of new questions for both players and developers. Imagine trying to design a Sims game where each Sim isn’t merely a collection of almost binary traits or needs, but rather an individual, if uncomplicated, neural net. But a convincing character needs a convincing personality, something we haven't seen much from computers as of yet. Machine learning so far has directed toward the impassive, and the cool, calculated manipulation of data. Can we code emotion, can we simulate an actual, self-aware being?

Technologies like Volta may give us the chance to find out very soon.