Gaming and graphics are the performance applications for processors and memory. As such, leading-edge technology debuts here and eventually migrates to mainstream computing, mobile, and consumer electronics applications over time. State-of-the-art GPUs deliver functionality including photorealistic game characters and environments, support for multiple simultaneous displays, 3D image processing and video output, and full HD 1080p resolution. In order to support this functionality, the number of graphics processor cores and transistor counts per chip are skyrocketing. High-end GPUs have over 2 billion transistors and more than 1000 graphics processor cores up from less than 100 just 5 years ago.
Historically, these performance increases have come with a commensurate rise in power consumption. However, because of thermal, power supply and cost constraints that trend cannot continue. Top-of-the-line dual-GPU graphics cards and game consoles can draw as much as 300 watts (W) of power and must allocate a significant portion of the bill-of-materials (BOM) for the cooling system. While demand for higher performance will be ever present, power efficiency will increasingly become a first-order requirement.
GPU’s must also be scalable to support a broad range of performance levels and price points. Although they are the performance drivers, high-end graphics cards make up only a small percentage of the overall market. A single GPU platform must be configurable through the use of multiple memory types, or a single memory with a wide performance range.
The combination of these factors puts tremendous demands on the graphics memory system. Bandwidth requirements for next-generation gaming and graphics systems will exceed 500 gigabytes per second (GB/s). Meanwhile the total power budget must remain constant or even decrease. Similarly, price points must remain essentially unchanged for each of the respective performance segments.