GPU and CPU AMD: a roadmap that goes up to 2020

GPU and CPU AMD: a roadmap that goes up to 2020

A few weeks ago AMD has provided information, in conjunction with the PC Cluster Consortim held in Osaka, Japan, on what are their roadmap for future products for the areas of CPU and GPU. The details are very limited but emerge a future scenario that comes up when the solutions expected for 2020.

As of 2016 AMD will release two separate products CPU architectures, both with 64bit support. The first type will be x86, intended to take the place of the family Bulldozer proposals currently on the market. The second will adopt instead the instruction set ARMv8 while the target audience of both will be as server systems to higher performance as client proposals.

The x86 architecture will be called Zen while the ARM will adopt the internal code K12. For the two CPU families will see the debut of the use of 14-nanometer production technology, technology Fin FET.

K12 so as Zen will be characterized by the parallel support many threads, indication which suggests for both technologies to support multi-threading symmetrical in a manner similar to that HyperThreading adopted by several years by many versions of Intel. The implementation HT features a double number, compared to that of physical cores, of threads that can be processed by the processor. AMD did not provide specific guidance, but it seems possible that this can extend to 4 threads per core, perhaps in variations of CPU specifically developed for use in server systems.

A novelty is not indicated in previous AMD roadmap is the inclusion of a High-Performance Computing or HPC APU, expected during 2017. It is a solution for server systems that combines next-generation CPU architecture with an integrated GPU always next generation. The result will be to have a product with high processing power for both the CPU as a component for the GPU, can benefit from the stacked memory HBM (High-Bandwidth Memory) moreover essential thinking of the processing power and bandwidth requirements required by a product of this type.

For GPU shows the will of the US to propose a new architecture every two years, following in this as already outlined by NVIDIA for its products. This does not imply the release of new video cards every two years, since the availability of GPU-based architectures of the new generation with this periodicity. The first concrete is indeed expected in the coming months, with the release of the upcoming video cards Radeon family.

Source: mynavi

GPU and CPU AMD: a roadmap that goes up to 2020 updated: March 30, 2015 author: John Malkovich