Soldato
- Joined
- 2 Jan 2012
- Posts
- 12,338
- Location
- UK.
An interesting article published by AMD of IEEE Micro outlines the strategy and the future projection of the house Sunnyvale within the class of exascale supercomputers (ie systems with very high computing power capable of performing a trillion [10 ^ 18] of operations per second).
According to AMD also 'exascale computing requires a diverse approach - which combines the CPU and GPU (in one chip) structures on multi-node cluster to a park right software - as it is the only solution that can guarantee a computational power high quality at relatively low cost and low energy consumption (cost reduction is achieved by using general-purpose architectures instead of expensive dedicated hardware solutions).
What AMD wants to do, it is to bring the concept of HSA (Heterogeneous Systems Architecture) within HPC servers via a new APU. APU powerful enough to be renamed with the acronym EHP (exascale Heterogeneous Processor) and consists of a chip with 16 cores Zen x86 (can process 32 threads in parallel), with a powerful integrated GPU belonging to the next generation "Arctic Islands "(something like a report DP / SP equal to 1/2, NDA), combined with memories HBM2 on-package.
As for the launch period, the roadmap of the HPC / Workstation of the arrival of the new AMD APU is expected between 2016 and 2017.
Source: http://www.bitsandchips.it/9-hardware/5858-amd-exascale-heterogeneous-processor
Source Translated: https://translate.google.co.uk/tran...le-heterogeneous-processor&edit-text=&act=url
Last edited: