AMD’s Vega Next Generation GPU Architecture Unveiled — How Will It Affect Gamers?

AMD’s Vega architecture has been something that the computing company has been teasing for quite some time. Today at the Consumer Electronics Show in Las Vegas, the company lifted the lid on the new GPU tech, showing off what makes Vega different than its predecessor.

Talking with the company, it’s clear AMD’s Vega architecture tries to move hardware forward on many fronts, but three main sticking points are driving the team: improved visual fidelity over what is available currently, giving content creators an unconstrained canvas upon which to create, and explore into a newer application of GPU tech: machine intelligence.

AMD’s new Vega architecture is being designed to combat the exponential increase of processing workload and how it is outpacing video memory on the graphics card. The rate of growth of video memory is stagnant compared to the amount of relative processing power, or GPU Compute (GFLOPS). Vega is targeted at breaking that paradigm, and it seems poised to do so.

One of the first things people look at when buying a new graphics card is how much video memory the card has. The general idea is that the higher amount of VRAM you have on the board, the more a card might cost, but the tradeoff is being able to play at higher resolution, frame rates, and texture settings. AMD’s Vega introduces what they call the “world’s first scalable GPU memory architecture.” Using their “high-bandwidth cache,” which is replacing a graphics card’s frame buffer, and the high-bandwidth controller, the Vega chip can handle data up to whopping 512 terabytes of virtual address space.

This benefit can be felt in all of AMD’s applications for the Vega GPU, including its newly-revealed Instinct product, but gamers can expect to see this impact their gaming experiences as well. AMD showed two examples of games, The Witcher 3: Wild Hunt and Fallout 4, that benefit from the high-bandwidth cache controller’s fine-tuned data usage. In the graph provided by AMD, both games use less memory traditionally allocated for the games at 4K resolution. And in the event that you do exceed the limit of memory allowed in the cache, Vega won’t stall or quit on you like a more traditional graphics card might. Instead it allows for the data to be paged in small chunks on demand.

Vega builds upon the high-bandwidth memory solution brought about by the company’s Fiji architecture, currently found on its AMD Radeon R9 Fury X, and the Radeon R9 Nano. Vega introduces HBM2, which is a clear step above its predecessor in terms of what it can handle, with two-times the bandwidth per pin over HBM1. Additionally, HBM2 has eight-times the capacity per stack of HBM2, building upon the 4GB memory capacity of HBM1. Finally, HBM2 has over a 50 percent smaller footprint over the more traditional memory model of GDDR5.

AMD’s Vega product also makes the graphics pipeline more efficient than previous GPU architectures. Rendering scenes is one way in which Vega is more efficient, using what Radeon calls “primitive shaders” which will quickly locate and identify polygons that are not visible in the scene, and in effect, it doesn’t waste time on them. An example they showed was from Deus Ex: Mankind Divided. The initial scene shown (shown below) has over 200 million polygons in the scene, however only about 2 million of those are visible to the player. Vega’s new pipeline prioritizes and streamlines those visible polygons and doesn’t waste time on the ones not visible to the player.

AMD also unveiled its new “Next-gen compute engine,” or Vega NCU, which according to the company is capable of 512 8-bit operations per clock, 256 16-bit operations per clock, or 128 32-bit operations per clock. What makes this interesting over the previous compute units is that AMD’s Vega NCU can perform two 16-bit operations simultaneously, something the previous iterations couldn’t do. Additionally, the new geometry engine adds to the efficiency, working with Vega’s “intelligent workgroup distributor,” spreading the load across the GPU much more evenly. According to AMD, the new programmable Geometry pipeline has over two-times the peak throughput per clock, compared to previous GPUs.

AMD has declined, thus far, to provide concrete specifics, such as core clock speeds, memory bus, concrete VRAM, as well as pricing. However, today’s architecture reveal did show what Vega is gearing up to do, and how the team at AMD and Radeon are essentially trying to future-proof the market for the foreseeable future. The Vega architecture by AMD Radeon shows the company’s willingness to move forward and invest in not only enthusiastic gamers, but also give high-end content creators tackling projects, such as machine learning, a look at the future.

Vega takes aim at the high-end market, and with AMD’s Ryzen CPU product, which AMD unveiled that the partner boards and custom boutique computers with Ryzen in mind, 2017 looks to be an interesting year for Team Red.

[Featured Image by TweakTown]