Slides show details on faster gddr5 variant of Micron

Slides of a gddr5x presentation by Micron have appeared on the internet. The manufacturer pushes this extension of the gddr5 standard forward as the new generation for video memory. This makes it a competitor for high bandwidth memory.

Micron announced work on gddr5x in early September and promised to officially unveil the technology in 2016, but Expreview is already publishing a presentation detailing the gddr5 expansion. Increasing the speed of gddr5 is a challenge, according to Micron. When the standard was developed, the throughput was expected to be a maximum of 5Gbit/s, but it is already heading towards 8Gbit/s. In addition, the clock speed of the memory is said to be at its maximum and the command/address protocol is also reaching its limits.

However, where there is still some stretch is the prefetch buffer, claims Micron. By doubling the size of the buffer from eight ‘data words’ per memory access to sixteen ‘data words’, an initial throughput rate of about 10 to 12 Gbit/s could be achieved. In the long run, 16Gbit/s should be within reach. In combination with a 384-bit wide bus, this yields a bandwidth of 480GBps initially and 768GBps at a later stage.

For comparison: the high bandwidth memory or hbm used in AMD’s Fury GPUs, when using four stacks with four layers, each with a data rate of 1Gbit/s, comes out at 512GBps. At hbm, however, there is a lot of room for improvement, both in the number of layers and the number of stacks. AMD’s new Radeon generation is equipped with hbm and future Nvidia cards will receive it as soon as the new generation hbm is ready. According to rumors, Nvidia will provide the mainstream GPUs based on the upcoming Pascal architecture with gddr5x memory.

The first video cards with gddr5x memory should appear on the market in 2016. According to Micron, the advantage of the memory is that it uses the same form factor, allowing manufacturers to continue to use today’s proven production methods.