New NVIDIA chip shows ability to adapt to changing AI, graphics landscape

Yesterday NVIDIA formally announced its next-generation graphics architecture called Turing, named after the early 20th century computer scientist credited as the father of artificial intelligence. This new GPU does more than traditional graphics workloads, embedding accelerators for both AI tasks and a new graphics rendering technique called ray tracing.

With the launch, NVIDIA continues to showcase its leadership in the graphics and AI markets, leaving competitors like Intel and AMD scrambling to play catch up at a point when the space is evolving rapidly, often at the direction of NVIDIA designs.

Cards based on this architecture are set to ship in Q4 of this year with price points ranging from $2,300 up to $10,000, targeting professional visualization developers, content creators, and real-time graphics. These fall into the “Pro Vis” segment of NVIDIA’s revenue divisions and represent a significant relative price increase over previous generation parts that lacked AI and ray tracing acceleration. Expect to see margins and income increase once these start shipping.

NVIDIA stated that this market represents 50 million potential customers and a $250B industry. While it is already the leader in the space, more advanced and revolutionary technology like this will accelerate upgrade demand and could lead to increased sales and revenue versus traditional 3-year cycles.

We will see consumer-focused variants of this chip for the GeForce gaming markets later this month.

NVIDIA is proving it will evolve

The most important takeaway from this architecture launch is that NVIDIA is not standing still. Though it built its leadership position in the graphics and AI spaces by utilizing the same underlying graphics architecture that was built for gaming and rendering, CEO Huang and his team understand that evolution is required to adapt in the ever-shifting technology world.

One worry that analysts have had for NVIDIA is that competition from dedicated AI processors like the ones from Google, or even projects on-going at Tesla or Facebook, would undermine the position NVIDIA currently holds. The GPU as it existed five years ago mapped well to the direction of deep learning and AI processing but as the workflows have changed, so must the hardware that powers them. NVIDIA does not appear to be overlooking that.

The same is true for graphics. NVIDIA remains the leader in high performance graphics but continues to push the market forward by integrating accelerated ray tracing compute cores.

Impact of the technology

During the unveiling of the new chip, NVIDIA CEO Jensen Huang called it the “most important innovation in computer graphics in more than a decade.” Those are lofty words that the truth of which will only be decided upon in retrospect. But the implications of what this new technology offers, as well as the weight it puts upon competitors in the market, appear substantial.

This is not the first graphics chip from NVIDIA to offer dedicated processing for artificial intelligence acceleration, called Tensor cores. The company’s Volta design, originally announced in June of last year, was the first to implement those cores with superior deep learning performance. Only a handful of products were released with that design.

Turing implements improved Tensor cores providing 6x the performance of the previous generation architecture, allowing software developers to implement new features for advanced anti-aliasing (removing jagged edges on render lines) or AI-based denoising (removing the imperfections in a real-time rendered image).

New to a graphics chip from NVIDA are ray tracing cores (RT cores). Ray tracing is the act of simulating how light bounces in the physical world in a virtual image, creating the most realistic and life-like results possible. The new Turing chip will offer as much as 25x the performance of the previous generation in ray tracing workloads, an incredibly valuable addition for film effects, professional development, or gaming.

How the competition stacks up

The only question that remains is how great of a lead does NVIDIA have over competitors? In the GPU space, AMD finds itself behind yet again. NVIDIA is now on its second generation of products with integrated AI acceleration in the enterprise and professional spaces and should have the first product out with AI Tensor cores for the consumer later this year. It also is taking a leading position in the race to ray tracing for professional and gaming purposes, combining new RT cores on the Turing GPU with AI features like denoising to bring the holy grail of graphics to life.

AMD has been silent on its graphics plans and strategy since the departure of Raja Koduri with only a mention of the 7nm variant of its current Vega architecture sampling in 2018. This revision will include some accelerators for deep learning tasks but to what extent and performance level is still unknown. Without dedicated ray tracing acceleration, the chip could be deficient in the most exciting avenue of change in the graphics world in several years.

Despite pressure from Google with its TPU project targeting deep learning acceleration, NVIDIA continues to ship products with leadership performance and penetration. Even Google is using NVIDIA graphics chips for its cloud-based AI inference systems, proving that NVIDIA is doing while most others are simply trying to.