The Age of AI-Specific Hardware Beckons
It’s an acronym popping up in workplaces across industry sectors – AI – for the simple reason that the effective use of this technology has the potential to drive greater business efficiencies and critical insights. For developers serious about integrating AI, tailor-designed hardware is not only a “nice-to-have” but, increasingly, a “must-have”.
AI-specific hardware can significantly benefit current AI use cases and, beyond that, will play a defining role in the future of technological advancement.
Right tools for today’s jobs
Across all domains, AI workloads all have one thing in common: an insatiable thirst for data and compute. To be performant and efficient, this requires heavy-lifting processing capabilities from your hardware. It’s this need for next-gen performance (and a host of other AI-relevant benefits) driving a wave of specialised AI-hardware.
It’s also why semiconductor specialists have been hard at work, finessing their AI portfolios. Take Intel’s offerings for example – the company has an end-to-end range that covers a host of hardware increasingly sought after in today’s AI-driven landscape, including processors with built-in AI acceleration, dedicated AI accelerators for training and inference, for Meteor Lake we went with NPU (Neural Processing Unit), and more.
Already available in its second-generation iteration, Gaudi by Intel® subsidiary Habana is a DL (Deep Learning) processor that has met the needs for high levels of training throughput and efficiency. Gaudi2 also provides substantially competitive cost advantages to customers, both in server and system costs. This helps businesses run an increasing volume of AI-led applications and services completing tasks such as image- and gesture recognition in videos, speech recognition, natural language processing, and others. Gaudi2 is designed for easy scalability, and companies have the choice to run it in the cloud or in their data centre. PyTorch and TensorFlow frameworks are available to drive the execution of Gaudi2’s DL models, which significantly increase AI-training performance and operational agility for developers, whether they’re working on the largest multi-modal models or more basic computer vision.
Speaking of the PyTorch or the Tensor Flow Frameworks, these are compatible with another AI-boosting innovation: the Intel® Data Center GPU Max Series. Previously codenamed Ponte Vecchio, this series is based on the scalable graphics and compute Intel Xe HPC micro-architecture, which is designed to deliver exceptional performance and functionality for AI workloads but also for HPC workloads like scientific computing.
Better DL training is critical, but once complete, how quickly and efficiently can AI put what it’s learnt to use? Inference is the ability of DL-trained AI to take new data and use what it’s learnt to deliver insights. A novel solution built into 4th Gen Intel® Xeon® Scalable processors, the Intel® Advanced Matrix Extension (AMX), is an integrated AI accelerator that significantly ramps up the inferencing power of machine learning platforms while improving the efficiency too. In fact, in one real-world case study, it boosted end-to-end inferencing performance by 2.48x, helping the client improve accuracy without sacrificing latency. Both Data Center GPU Max Series and 4th Gen Xeon Scalable processors share the same open programming and multiarchitecture tools, the Intel oneAPI solution.
Coming up next is Meteor Lake with its integrated NPU (Neural Processing Unit), targeting laptops. The NPU makes quick work of processing AI workloads like Microsofts Studio Effects in a way that frees up computing resources on the CPU or GPU.
Increasingly, cost and power efficiencies need to be balanced with superior performance and speed. Compared with general-purpose offerings, specialist AI hardware is truly primed to deliver on these fronts. If the use of AI continues its current trajectory, hardware like this won’t simply be a ‘nice-to-have’ solution for driving competitive edge: it will likely be the norm in years to come; a business ‘must-have’.
The next AI frontier
Cutting-edge hardware is clearly an enabler of AI-led innovation happening right now. What’s exciting is the next evolution. The power of tailored hardware is multiplied when you team it with specialised software in a single technology solution that’s guided by the principles of biological neural computation.
While this sounds theoretical, this is exactly what neuromorphic computing is. Taking a radical departure from the current mainstream AI solutions, this new architectural approach more closely emulates how the human brain operates to deliver capabilities closer to human cognition. It’s capable of more autonomous, cognitive reactions to data in real time and has significantly improved power efficiency.
Working tirelessly to transition neuromorphic computing out of the confines of research labs and into commercial field applications is The Intel Neuromorphic Research Community (INRC). Comprising over 75 partners from all over the world – including research teams from major technology, automotive, aerospace, and academic organisations – the INRC’s ongoing research has led to some key breakthroughs. Intel’s development of its second-generation neuromorphic research chip, Loihi 2, has enabled such innovation. INRC researchers have been able to leverage this chip, alongside Intel’s LAVA open-source software framework to unlock collaborative progress that converges on a common set of tools, methods, and libraries.
If innovations based on the first iteration of the Loihi are anything to go by – researchers trained the chip to learn neural representations of 10 smells, which is a benefit that could help home-use sensors distinguish between carbon monoxide and smoke – the possibilities seem endless. Indeed, they are, with the caveat that we need to help shape communities and environments to nurture this technological evolution.
Encouraging advancements
The UK has played a role in breaking technological ground in the past, and the opportunity is there for the nation to secure its pole position, spearheading new AI advances in the coming years.
A greater push for research investment in neuromorphic technologies is needed. Alongside this, we need to focus on providing specialised training for the next generation of skilled AI processor architects and designers to help scale the expert use of such technology.
Public- and private-sector organisations should join forces in a collaborative effort to usher in this new era of advancement. Multi-stakeholder partnerships should adopt a co-design-based approach to bring together neuroscientists and specialised tech developers and designers so that the entire software-defined, silicon-enhanced stack is created and optimised together.
We should also strive for increased accessibility. If the wider industry encourages a more open software ecosystem for AI, it will help foster innovation while mitigating vendor lock-in. Intel has helped the industry make great strides on this front through the oneAPI initiative.
Staying ahead of the curve
There are a host of exciting AI developments already available, and others are being actively worked on in research labs.
In the long-term, new architectural approaches such as neuromorphic computing offer to power a future of autonomous AI solutions with energy efficiency and continuous learning at their heart. Truly promising to open the doors to exciting new possibilities across sectors and industries.
While businesses help sow the seeds for this next evolution, priming a skilled workforce and championing a collaborative spirit alongside other stakeholders, they can take advantage of efficiency and cost gains in AI right now. Embracing specialised hardware to better manage AI workloads is an important action that will put you in good stead for what comes next.
techUK – Unleashing UK Tech and Innovation
The UK is home to emerging technologies that have the power to revolutionise entire industries. From quantum to semiconductors; from gaming to the New Space Economy, they all have the unique opportunity to help prepare for what comes next.
techUK members lead the development of these technologies. Together we are working with Government and other stakeholders to address tech innovation priorities and build an innovation ecosystem that will benefit people, society, economy and the planet - and unleash the UK as a global leader in tech and innovation.
For more information, or to get in touch, please visit our Innovation Hub and click ‘contact us’.
Tech and Innovation Summit, 19 June
Emerging technologies will be debated and explored at our annual Tech and Innovation Summit, taking place on 19 June. This campaign week will directly feed into the themes discussed at the Summit.
techUK’s flagship Tech and Innovation Summit returns to traverse the extraordinary and ground-breaking discoveries made possible by the application of emerging and transformative technologies.
Upcoming events:
Latest news and insights:
Get our tech and innovation insights straight to your inbox
Sign-up to get the latest updates and opportunities from our Technology and Innovation and AI programmes.
Learn more about our Unleashing Innovation campaign:
