IT Infrastructure

AI: Hype or reality?

Share this post:

 

We have seen machines beating humans; IBM’s Deep Blue became world champion chess player in 1997, IBM Watson became the best Jeopardy player in 2011 and Ke Jie lost from Google’s AlphaGo in 2017. Just a few examples of the power of artificial intelligence (AI) in recent history, growing from artifical intelligence to machine learning to compute learning. Where in the beginning AI consisted of feeding the machine with all possible outcomes, and using lots of computing power to pick the best one, you will see computers training themselves and gaining ‘intelligence’ along the way.

The next generation of AI goes beyond super-fast in- and output. This deals with the possibilities to make new information based on data that has been fed to the machine and data that it picks up real-time by lots of different kinds of sensors. We’re talking machine learning or deep learning. When it comes to real AI we ain’t seen nothing yet. Although the IT-industry talks about neural networks for decades; now is the time. Because now both the hardware and software are ready for AI.
Speed is key; from a hardware perspective. Machines do no longer only rely on their CPU, but can expand to the massive parallel compute power in GPU’s. Wich allows more training sessions in the same time and period, leading to more intelligent networks. The biggest benefit here is this will lead to higher accuracy of the model when in production.

On top of this IBM and NVIDIA have built the NVLink to enable 2,5x the data throughput compared to PCIe, enabling data to move superfast between GPU’s but also from GPU to CPU.
From a software perspective, you will see that open sourced frameworks like Caffe, Torch, Theano, Tensorflow, etc. have grown in maturity and scalability and to make life even better, IBM has bundled the most popular machine learning frameworks and their dependencies, built for easy and rapid deployment. IBM’s most recent addition in this space is adding Distributed Deep Learning to it’s PowerAI bundle which enables you to scale over multiple servers with GPU’s (up to a cluster of 64 servers with a total of 256GPU’s at 95% efficiency.

The combination of PowerAI software and superfast and scalable POWER8 hardware has led to a new world record, beating Facebook’s record from June 2017.

 


Inspired to get started?


Linux on Server Solution Seller

More IT Infrastructure stories

Data-driven asset management with IBM Maximo Application Suite and Cloud Pak for Data

IBM has enhanced its Enterprise Asset Management platform, IBM Maximo Application Suite (MAS), with IBM Cloud Pak for Data: a supporting platform which provides a framework for combining a variety of data from different areas of an organization. How does IBM Cloud Pak for Data help organizations gain additional asset management insights from available data? […]

Continue reading

Being a data driven organization: What does this mean at Allianz?

Reading Time: 8 minutes From calculating risks and premiums to understanding customer behavior, data is of vital importance in the insurance business. Allianz, a multinational financial services company that focuses on insurance and asset management, has recently transformed its operations on a data level to reinvent its insurance business. Read an extract from a recent […]

Continue reading

Hardware is not dead – Power10 is paving the way for hybrid cloud and energy savings

  IBM has recently launched the first Power10 server. A new generation of servers, which due to changes in structure, components and functionality, provide significant improvements in terms of performance, computing power and energy consumption. These servers are paving the way for a more sustainable and flexible business based on a hybrid cloud. But what […]

Continue reading