Recently, Korean sources reported that generation of NVIDIA graphics solutions Pascal will be released on 16 nm technology using FinFET at the facilities of TSMC. Colleagues from the Swedish website SweClockers got hold of slides from the Japanese conference at which NVIDIA representatives will perform today.
Mentions the use of HBM, NVIDIA is not in a hurry, indicating the appropriate memory simply as chip with three-dimensional layout. However, judging by the bandwidth of the memory interface at the level of 1TB / s, it certainly has been going on memory generation HBM2. Moreover, the amount of onboard memory, and this is indicated - at least 16 GB significantly higher than the allowable amount of memory such as HBM, which in the case of AMD video cards is limited to four gigabytes.
By the way, there is information about the interface bandwidth NVLink - it reaches 80 GB / s. Unified Memory Architecture Supports a more effective exchange of information between the CPU and GPU.
In addition, it confirmed the information about the possibility of working with graphics processors Pascal floating point double precision (FP64).
Related Products :
|