quarta-feira, 19 de abril de 2017

Moore's law challenges and solutions


We are in the midst of true turning point computing, as it has changed the point of contact with our daily technologies.

We have most of the equipment we use Embedded sensor and rapid deployment of internet connection is a "smart device" capable of responding to voice commands, a large amount of data is generated, becomes a hub and cloud of computer network I will.

However, virtualization augmented reality has begun to adopt these technologies require significant computer processing and graphics processing to gain more realistic experience. This is to investigate large amounts of data, or to provide information timely and contextually, with astonishing progress in machine learning applications that can inherit these new applications day by day, We are challenging to provide more computing resources at a more affordable price.

Demand supply is so, more calculation is particularly difficult. The progress of Moore's Law was postponed.

Moore's Law is defined as the trend that the minimum circuit leads to twice the number of transistors on biennial chips that leads to performance and energy efficiency. Previously, in order to promote the next generation of low-cost computer chips, we were able to rely on each generation of semiconductor technology.


The laws of physics are never deceived. The miniaturization of transistors is approaching the physical limit. The new semiconductor technology node has not reached the historical improvement rate in order to achieve low power consumption and substantial reduction in size and cost increase over the next ten years.



In this way, we are facing the dichotomy as merely a slow historical improvement of Moore's Law. A new application of intensive calculation, you need to have many resources exponentially. This more data and multiple data processing, more real time information is always triggered by the consumer's demand for the high speed service required. Automotive, drone, all robotics, real-time information processing massive, reasoning requires interpretation.

For proof of malfunction or prompt response, all calculations can not be executed in the cloud. We will need to calculate to make sure that you are close to the user beyond the edge of the network. In intelligent applications and AR / VR interfaces, cars, houses, emerging businesses, cell towers, yet still require local high performance computing capabilities and while connected to the cloud.

An explosion of data to disseminate sensors to the device by millions of "Internet's - seeds (things thing)" by scanning most of our work and personal lives Occurred. A huge amount of this data needs to be analyzed by processing large amounts of data in real time. We can use this data hybrid environment in VR / AR in a new way, you can just mix the space with the space before the rendered image, you can see the duplication of visual information there is. Such demands, computer performance

This power, in addition to the real world point of the graphic overlay related to the realistic image and context, to reproduce the real, to achieve the view, information and augmented reality and virtual reality, to reproduce the real.

At the Royal College of Medicine, we imagine AR Overlay, record the surgery on the VR, and the surgeon provides real-time information so that it can be done more accurately. These applications are highly destructive.

Is this confusion possible to calculate after continuing pace, keeping the computer on the collapse of Moore's Law in the industry best condition, which has an impact on the industry? How can I provide more calculations?

It turned out that engineers have many features that can be manipulated to improve future performance. This is the details of Moore's Law. Engineers are more creative and interdisciplinary, you need to take advantage of collaboration with industry. Moore's Law, the four major elements,
And it is integrated with the cost effectiveness of new package semiconductor devices and a slight increase in interconnected technology. This allows flexibility to match chip technology in a new way.
Use special promoters and heterogeneous computing processors (CPU and GPU) to provide high memory data for such engines.
Development and easy-to-use open source software programs can use heterogeneous computing framework programs.
Application VR / AR ecosystem software that can be programmed using advanced computer learning, data analysis, simple provisioning application.

Plus Time Moore's Law will promote universities and industries by applying these leverage performance. Before casting, extreme ultraviolet lithography, it can be a useful technique to promote small node production process leading to new small transistors. These will be low resistance and will be sent together with a new metal structure. Semiconductor manufacturing will be more advanced.

For future applications, PCs, mobile devices, servers, etc., you will need more memory. In the case of servers, especially machine learning, virtualization applications, you need more memory on certain workloads such as database processing. However, the increase over the previous year has reduced memory. Again, it leads to new benefits, as seen in innovation, new memory nonvolatile memory and stack.

We are also proceeding with packaging by inexpensive technology for connecting multiple dies to an organic package. CPU, graphics, stacked memory, and 3D are stacked die other elements of the connected chips without biscuits at the bottom more. A while ago, you can place the optical interconnection of the mold for native. These are important for system design flexibility, performance for a deeper new approach to computing that can effectively connect high-density nonvolatile permanent memory, without loss of content when the power is turned off .

In my opinion, to make it easy to program these new approaches, the effort to keep the Law of Growth Rate at 18 to 24 months of Moore is that it is not helping at all. The ecosystem is in order of CPU, but in order to use the GPU and other accelerators, you need to have an open approach. Some took a unique approach, but it works, but it is expensive.

AMD co-architecture based heterogeneous (HSA) -based CPU, GPU fixed function, can accelerate ligand including FPGA shared memory, which can be optimized from the current system perspective, promoter.

Moore's law, the need for collaboration with universities and the need for open standards to continue the development of the semiconductor industry, manufacturing industry will facilitate the creation of the program. I am the reason why companies were convinced that they could manage the cost curve to add more transistors.

These continue to further progress of computing and summarize. According to Moore's law, instead of slowing down the pace of Moore's Law, we will accelerate the interruption.
Share:

0 comments:

Postar um comentário

Arquivo

Tecnologia do Blogger.
GeraLinks - Agregador de links

Blog Archive

Blogger templates