As is known, Samsung skipped the first stage of HBM memory production and immediately proceeded to the issue of High Bandwidth Memory modules improved specifications. This means that the HBM modules can be increased eightfold capacity, to pack a stack of eight memory crystals and exchange rate to provide 256 Gbyte / s per module. This opens the way for video cards with 16 and 32 GB of memory on a substrate with a processor with a data rate of over 1TB / s. Currently, Samsung has released four-layer modules with capacity of 4 GB, and 8 GB production units will start before the end of the year. And what about the company SK Hynix?
In an interview with German site Golem.de representatives of the company SK Hynix said that mass production of the 4-GB modules HBM on improved specifications will be launched in the third quarter of this year. Mass production 8 GB memory company will begin in the fourth quarter. The delay in the release of improved memory may mean that SK Hynix can satisfy demand for HBM memory of first generation or point to the lack of sufficient demand for the memory of the "second generation". While HBM memory in appreciable volumes demanded only by AMD for the graphics cards GPU Fiji. Demand for chips with more volume increased bandwidth will depend on the output speed GPU AMD market by Polaris architecture and GPU NVIDIA in Pascal architecture.
Note the representatives of SK Hynix declined to discuss the production of 2-GB memory HBM improved specifications. Nevertheless, such chips could be registered in the mass cards . Also, 2-GB memory HBM seems most promising for the installation of the APU AMD.
Finally, we draw your attention to the fact that in the second generation HBM will be also chip with the exchange rate typical of chips first generation. This explains the lack of formal specifications HBM definition . From JEDEC specifications, there is only HBM memory without any prefixes. Related Products :
|