AMD Radeon Vega GPU Architecture 223

AMD Radeon Vega GPU Architecture

Improved Memory Management »

Introduction



Vega is the common name of α Lyr, a star located some 25 light years from our Sun, which the scientific community regards as being the "second most important star in the sky after the Sun." Over the centuries, data from observing this star has been used to make great discoveries in the field of astrophysics. It also earned its place among several cultures and mythologies. AMD named its next-generation graphics processor architecture after this majestic, pale-blue celestial body. It is also AMD's most important development by way of high-performance silicon design after the "Zen" micro-architecture on which the company is pinning its resurrection in the CPU market.



"Vega" isn't a fancy name for the 5th generation Graphics CoreNext architecture, a new iteration of AMD's number-crunching machinery inside the GPU, but an entirely new GPU design. The GPUs will still employ GCN compute units, but will augment them with several new components never before featured on AMD GPUs, which will improve the chip's efficiency to cope with complex graphics and general processing workloads.



The idea behind "Vega" is that conventional approaches to making GPUs, so AMD believes, will not be able to cope with emerging workloads. The industry has already made a great leap from years of stagnating at 1080p to 2K and 4K resolutions that demand a proportionate increase in GPU performance. What's more, the workload itself requires optimization at the silicon level, without AMD having to optimize them at the software level through regular driver updates. This means giving the GPU the ability to learn the way a 3D app behaves to have it optimize itself to the app, which isn't easy as the GPU needs a second set of lightning fast memory that works in tandem with the video memory itself.



The GPU also needs expanded geometry processing capabilities to cope with the ever-growing realism in today's photorealistic 3D scenes. To work the tremendous amount of math that goes into it, the GPU needs a new Compute Unit design with added performance. Lastly, it needs an updated pixel-engine that puts it all together to draw the 3D scene. AMD claims to have made big changes to all four areas with "Vega."

In this article, we take a brief stroll through the various changes AMD made to the modern GPU with the "Vega" architecture.
Next Page »Improved Memory Management