Samsung reveals GDDR6 memory technology is coming to graphics cards - but in 2018, not 2017

Graphics cards will need more power and more bandwidth to better handle virtual reality

The GDDR6 memory standard will be introduced to graphics cards in 2018 and not in 2017 as had been widely expected - at least according to Samsung, one of the world's biggest makers of memory chips.

GDDR6 should offer bandwidth between memory and graphics processing unit (GPU) of more than 14 gigabits per second (Gbps) - perhaps up to to 16Gbps according to some reports.

Based on these raw bandwidth speeds, the standard is also expected to be able to offer total memory bandwidth of up to 512 gigabytes per second (GBps) via a 256-bit memory bus, and up to 786GBps on a 384-bit wide memory bus.

The increased bandwidth will enable GPUs to address more data, and also enable graphics cards to come loaded with even more memory for especially intensive applications, such as virtual reality.

That compares to a bandwidth of up to 12GBps from the most recent GDDR5X parts launched earlier this year by Micron Technology.

It will also offer power consumption reduced by about 20 per cent and an improved memory controller, LP4X, which is more efficient at management the balance between voltage and clock speeds.

The news was revealed by Samsung in a presentation at the Hot Chips technology conference in San Francisco, California this week, which placed GDDR6 firmly in 2018 - not 2017 as widely expected. The conference had also earlier revealed ARM's plans to take its microprocessor designs into supercomputing.

However, even if it is introduced in 2018, it may still take some time for the technology to percolate down from the ultra-high-end into the mainstream.

Graphics card makers have only just started shifting to GDDR5X, with the NVIDIA GeForce GTX 1080 becoming the first to use it when it was launched in May 2016.

The 1080 was the first graphics card to deploy NVIDIA's new GP100-series GPU, which was also the debut for the company's Pascal micro-architecture. The GP100-series parts are the first GPUs to be based on a 16-nanometre process technology, compared with the 28nm manufacturing process used on the previous Maxwell-architecture parts.

NVIDIA's main rival in graphics cards, AMD, has rushed out its Radeon 400 series graphics cards in response, based on its 14nm Polaris GPU, which will be closely followed by Vega - perhaps before the end of the year. It also unveiled in July an ultra-high-end graphics card with one terabyte of memory - compared to the more usual two-to-four gigabytes of mainstream graphics cards todays.

The new GPUs will increasingly need to support virtual reality games and applications, with the technology starting to take off this year, while NVIDIA also foresees its GPUs being used more widely in other applications, too, such as self-driving car systems.