The cloud is a natural fit for CIOs that wish to evaluate artificial intelligence (AI) because it lends itself to proof of concept (POC) and experimentation without the need for buying hardware that might ultimately not be required.
Using AI capabilities in the cloud also makes sense for organisations beyond POC that are ready to scale out AI deployments requiring the flexibility and proven technological capabilities that cloud platforms offer.
“Cloud has agility and the ability to scale quickly and is very interesting in the domain of AI applications which are new and fast-moving,” says Chris Feltham, industry technical specialist for cloud at Intel.
“If you are at the exploratory stage, cloud allows you to fail fast and learn fast. AI is the perfect fit for cloud because of its infancy for a lot of customers,”
If AI is the answer, what is the question?
Intel focuses on helping customers figure out how best to understand and deploy AI. But first, CIOs need to pinpoint the fundamental business question to be solved - a critical step which is often missed due to all the hype around AI.
“Certain companies are doing AI seriously and others are experimenting. There is a lot of buzz around AI and there is a feeling of, ‘I need to do AI’, but if AI is the answer, what is the question?” says Feltham.
“Sometimes, AI might not be the right option - there may be a quicker route with more traditional methods. AI is just a tool.”
In some cases, organisations might have a specific problem that can be solved using an off-the-shelf solution such as a chatbot or robotic process automation. Other organisations may want to develop the capability to build their own bespoke AI. Either option is valid but will require different approaches.
“From your own solution to an off-the-shelf solution, there are different flavours to choose from. A company may want a speech-to-text solution that just happens to be based on AI or it may want to develop its own AI solution and choose the right environment. All options are open, but first ask, ‘What is the question I am trying to answer and what am I trying to achieve?’ AI is not the goal,” says Feltham.
Accelerate AI adoption
Intel’s AI Builders Programme aims to accelerate the adoption of AI across Intel platforms. Specific solutions can be found for specific business requirements that have AI capabilities. However, without knowing the goal, the risk is that continuous experimentation and “POC paralysis” leaves companies playing with AI for the sake of it, warns Feltham.
Without a plan in place, and an enabling culture, the risk of frustration is high. The business sees the power of AI after a POC but is unable to do anything meaningful in production and the journey stalls and fails.
Intel aims to work with customers to get the most from their investments in Intel platforms. The company’s second generation Xeon Scalable Processor is the only general purpose processor with built-in AI acceleration and is up to 30 times faster than its previous incarnation for image inferencing tasks.
“We help customers to get started simply. First of all, customers must understand what the data is they need, where it is, access to it and its use for AI,” says Feltham.
Flexible AI is the goal with a focus on frameworks, speed of development and deployment, and ability to scale. Intel works with cloud providers and - as most of the cloud is run on Intel - can help customers get the best out of CPUs and infrastructure to achieve business goals.
“The value for the customer is that we understand the capabilities in our silicon and can ensure the right hardware and software is paired up to maximise performance,” says Feltham.
Cloud synergies
Within the healthcare sector, for example, Philips used the OpenVINO toolkit to optimise an AI model for medical imaging and improve performance of deep learning inference performed on patient x-rays and CT scans. Xeon-based servers gave speed improvements of 188 times faster for x-rays and 38 times faster for scans. No accelerators nor modifications to the hardware were necessary, saving money and improving patient outcomes.
These synergies with Intel-based cloud infrastructure and AI make business sense to customers. Intel’s optimised deep learning frameworks offer compelling performance enhancements for developers. For example, Intel demonstrates 11 times performance improvement using the Intel-optimised version of Tensorflow.
“If you develop AI and use Tensorflow, why not use the Intel-optimised version as we can give you a better result and a significantly better performance than the standard version?” says Feltham. Intel takes a similar approach to optimising other frameworks including MXNet, Caffe, PyTorch, Theano and Chainer.
Achieving Transformative Application Performance with Vexata Systems and Intel® Xeon® Processors and Intel® SSDs
Read this brief to learn how Vexata and Intel have collaborated to solve the digital transformation challenges using a solutions-based approach that brings the best out of applications running on Intel processors using Vexata storage systems.
Download NowDelivering the best AI performance
By combining the right tools with the right hardware, ensuring the access and availability that cloud provides everywhere, AI solutions become easy to deploy for the best possible performance.
“When looking at AI, CIOs must evaluate cloud partners and skills. Be clear about the outcome you want as it will better shape your choice of partner. The value that Intel offers is that Intel-based architecture for cloud is flexible, tested and proven,” says Feltham.
The result is success without breaking the IT budget to explore the promise of AI, and without being overtaken by the hype.
“AI is another workload on an Intel platform, and we can ensure it performs well,” says Feltham.
“CIOs don’t have to buy a new silo of different technology. They can use the capability available which Intel understands and can apply to AI.”
