SK Hynix Adds 4GB HBM2 Memory To Product Catalogue
SK Hynix has now officially added HBM2 memory to their product catalogue with the availability of 4GB memory stacks expected in Q3 2016. This will finally usher in the era of AMD’s high end Vega GPUs as right now Nvidia’s Tesla P100 graphics accelerators are the only ones confirmed to use HBM2.
The company will be releasing two different versions of HBM2 memory, both in 4GB stacks but with different bandwidth specifications and data rates. Two of the fastest HBM2 memory stacks combined can deliver bandwidth equal to a GTX 1070!
The high end HBM2 memory stack, H5VR32ESM4H-20C, will be featuring a data rate of 2.0GT/s and bandwidth of 256GB/s. The other stack, dH5VR32ESM4H-12C, on the other hand will be delivering 1.6GT/s data rate and 204GB/s bandwidth.
To put this into perspective, if the fastest of these new memory chips were used in AMD’s R9 Fury X instead of HBM1 it would have 16GB of VRAM and over 1TB/s of memory bandwidth. Not only does it deliver amazing performance but also has less power consumption which is indeed a great step for a second evolution of any product.
This kind of performance simply outcompetes all competing memory solutions in terms of memory bandwidth per chip and memory capacity per chip.
SK Hynix also has plans to launch 2GB and 8GB stacks as well as a slower 128GB/s variant of HBM2. It is expected that the company will be fully abandoning HBM in order to favor the faster and denser (in terms of storage) HBM2.
Mass production of the 8GB HBM2 stack is expected to begin in the fourth quarter of 2016, hopefully in time for AMD’s upcoming Vega line up of high end GPUs as well as Nvidia’s next generation of cards.