Next year the memory type HBM2 will adopt NVIDIA company to create its graphics generation Pascal . They will be manufactured by TSMC using 16-nm technology FinFET. In fact, in the last quarterly conference AMD representatives admitted that they already have examples of similar products in the technological implementation.
We have already reported that the use of memory type HBM2 allow NVIDIA to increase the storage capacity of a single graphics card up to 32GB. Pascal will be equipped with chips HBM2 8 layout. AMD Fiji is now limited 6 layout. However, one of the Pascal version will be equipped with HBM2 memory 4 layout, organized in four blocks. Memory bus, in any case will be 4096-bit, and memory frequency exceeds 1 GHz. The current first-generation HBM operates at a frequency of 512 MHz but uses the same 4096-bit bus.
According to the source, modifying the GPU NVIDIA GP100 4-tiered arrangement of memory is designed for consumer-grade graphics cards, and 8-tier version will get Quadro graphics cards and computing Tesla. In contrast to the Maxwell architecture , which is weak in computing acceleration , Pascal will be able to offer corporate customers a decent performance level . In addition, Pascal provides support for new high-speed interface NVLink, which will ensure the transmission of information between the GPU and CPU at a speed of 80 GB / s.