Many industries are investing heavily in artificial intelligence (AI), and deep learning in particular is generating enormous interest.
The deep learning market is expected to grow at an average annual rate of 42% until 2024 (Source: MarketWatch). One reason for this growth is that extremely large data sets of very good quality can now be accessed.
Nevertheless, a lot of enterprises find it difficult to jump on the bandwagon. This is partly due to the fact that AI today still requires a lot of experimentation, even though development is progressing rapidly.
However, there is now a shorter track to harnessing the potential of deep learning for your business – namely, the Zinrai Deep Learning System (ZDLS).
Why is deep learning still a bit geeky?
Despite all the attention, deep learning is still an area that demands a lot of work from specialists. And for good reason. AI development is progressing rapidly.
AI projects which are being implemented today will encounter considerably more advanced environments, the so-called frameworks, in just a few months. But the infrastructure also falls into this category, and here, too, things are moving forward quickly.
New technologies enable faster project implementations and thus ultimately impact the time to market for the new product and the new innovation, enriched with AI.
Many “AI things” are open source and freely available – and that is good. Freely available neural networks / models are only a mouse click away.
At the same time, getting the best out of neural networks is becoming more complex. After all, software and hardware are being increasingly intertwined and coordinated, and the magic word here is appliance. And this has a significant effect on the training and testing of neural networks.
Fujitsu has developed ZDLS with the uniquely developed Deep Learning Unit (DLU, a processor for deep learning training) to offers an exhaustive range of pre-tested and tuned neural networks.
The most popular and best of the market are included, whether for image recognition, speech, text, your own bot development, and more – all optimally tuned to the infrastructure and powered by the remarkably fast DLU.
Fujitsu brings AI to the data center
By integrating with the data center infrastructure and processes, ZDLS and even multiple ZDLS nodes can be managed with ease by IT administrators, just like any other datacenter system.
In addition, ZDLS guarantees significantly lower power consumption compared to current GPU-based solutions. In most cases, it can also ensure full utilization of the available rack space and required cooling. This ultimately saves a great deal of investment and operating costs.
Depending on the infrastructure, five-to-six-digit savings can quickly be achieved over a period of three years!
The continuous further development of ZDLS, aligned with parallel developments from leading technology partners, is guaranteed.
In my role as Chief Evangelist for the Data Center Business, I recently visited VMWorld 2019 in San Francisco to discuss the latest developments in SDDC as its relates to AI.
One example is a special Fujitsu virtual machine under VMware or any similar solution that allows the storage and loading of development environments (the so-called containers). This can largely prevent the problems of rapid further development of frameworks and the necessary adaptation of the written application to new environments described at the beginning.
You simply load the appropriate environment for the application and develop it further. Done!
Thinking AI is different: get more – pay less
The advantages of an “AI-ready” system are obvious: Companies avoid long planning, experimentation, configuration and test phases. This saves a lot of time and money. The integration in existing infrastructures and processes is, however, only one highlight of ZDLS.
The groundbreaking package of ZDLS takes deep learning applications to a whole new level. The hardware potential can be exploited 100%.
Massive parallel throughput reliably avoids bottlenecks and uses not only FP32, but also the newly developed DL-INT8 functionalities, a mathematical innovation from Fujitsu Labs Japan.
This pays off in practice: AI training is accelerated and guarantees the same or better accuracy compared to traditional GPU-based AI infrastructures. It consumes significantly less energy, and makes optimal use of the space available in the data center.
Therefore, the Fujitsu formula, based on the super-fast Deep Learning Unit, the core of ZDLS, is simple:
Less hardware + less energy + less rack space + less work = More performance
There are many reasons to explore the potential of deep learning for your business – and many more arguments to exploit it with ZDLS. We’ll be talking lots more about the potential this has to transform your results from deep learning – watch this space!