Nvidia Quadro RTX GPUs have Samsung’s 16Gbit gddr6 memory

Samsung has announced that its gddr6 memory with a capacity of 16Gbit is part of the new business Quadro RTX video cards based on the Turing architecture announced by Nvidia on Tuesday.

According to Samsung, the gddr6 memory doubles the capacity compared to its own gddr5 memory chips with a capacity of 8Gbit that are baked at 20nm. The gddr6 memory achieves a bandwidth of 14Gbit/s per pin and a throughput speed of 56GB/s. According to Samsung, compared to its own gddr5 chips, this amounts to an increase in bandwidth of 75 percent. The gddr6 chips are also 35 percent more efficient than the gddr5 chips, according to Samsung. The latter is related to the fact that the new chips operate at 1.35V, instead of 1.5V with the gddr5 chips.

The South Korean manufacturer started mass production of the gddr6 vram in January. At that time, the company still reported a speed of 18Gbit/s; Samsung now reports that the chips will not go beyond 14Gbit/s. Similarly, there is also a difference between the maximum throughput. Samsung reported in January that it came out at 72GB/s, while the company now calls it 56Gbit/s. It is unclear how exactly this difference can be explained; perhaps Samsung will release this faster gddr6 memory at a later date. In 2016, Samsung predicted that gddr6 memory would initially have a speed of 14Gbit/s in 2018, which would later increase to 16Gbit/s. The Jedec standard describes gddr6 speeds from 10 to 16Gbit/s, with capacities up to 32Gbit, but Micron has already achieved speeds of up to 20Gbit/s.

On Tuesday, Nvidia introduced the Quadro RTX 8000, RTX 6000 and RTX 5000. These new GPUs, which will be available from October, support ray tracing, among other things. This is a technique in which light rays are radiated from the position of the camera in different directions. How these light rays then reflect is calculated by means of specialized RT cores based on the color and properties of the material.