Virtualization Technology News and Information
Article
RSS
SambaNova Systems 2021 Predictions: 3 Trends That Will Define AI Hardware in the Next Year

vmblog 2021 prediction series 

Industry executives and experts share their predictions for 2021.  Read them in this 13th annual VMblog.com series exclusive.

3 Trends That Will Define AI Hardware in the Next Year

By Marshall Choy, VP of Product at SambaNova Systems

The computing industry is poised to undergo a massive technical transformation in 2021, representing the culmination of decades of technological progress hastened by a transformative and disruptive 2020. COVID-19 has reshaped work in America - a true shock test for the digital transformation enterprises have now been undertaking for years. Meanwhile, government and societal crises are asking tough questions about the future of technology. Can AI be a helpful - not harmful - tool for the advancement of the planet and its people?

SambaNova Systems is building the industry's most advanced platform, from silicon to software, custom-designed to run artificial intelligence (AI) and data-intensive applications. Marshall Choy, VP of Product, predicts 2021 will see huge technological innovation in harnessing the true power of AI to respond to seismic global disruptions. That innovation, he believes, will (and must) begin with the hardware that is the foundation and source of AI's potentially world-changing power. Read on for his predictions about what this will look like.

2021 AI Hardware Predictions

  • The end of multicore: Although many people in the space are building systems based on a core approach, which is more suited for the predictability of traditional computing, cores themselves are becoming less and less efficient. Putting more cores together in a multicore chip and in a multicore system only yields a more inefficient system. And, many organizations are moving away from traditional computing and adopting AI systems, which are probabilistic systems that can mold, change and evolve depending on what kind of data - the core approach will not pan out. The speed at which AI models are being developed requires a flexibility that can keep up with where models are today but, more importantly, lead models into the future. I predict that organizations will move away from multicore and seek out more flexible systems to better meet their needs.
  • The convergence of training and inference: Conducting training and inference on a single common platform unites both hardware and software, and creates a much more efficient, fluid system and streamlines operations. In a common platform, one can blur the lines between training and inference and do specific editing of a model or training in real-time as opposed to going back and forth.  
  • The next generation of general purpose applications will expand beyond ML: Computing needs to adapt to fundamentally shift forward. While AI and ML are a hot area of interest in research, they do not fully encompass all workloads. There will continue to be a broad range of application types that are both deterministic and probabilistic. AI, ML and HPC all have different computations so a singular one trick pony type of infrastructure may not be able to serve all those needs for all the different types of applications out there. The evolution of hardware is not only adapting to ML, but adapting to a new world of computer requirements.

These innovations will catalyze a quantum leap in AI power, hugely increasing the speed, efficiency, and ease with which technologists, businesspeople, and researchers can go from massive raw or unstructured data to world-changing insights and information. With the flexibility of software-led hardware design, the time organizations once spend on calibration, finetuning, and troubleshooting can now be spent on what really matters: employing data to make real progress on issues that affect all of us.

##

About the Author

Marshall Choy - VP of Product at SambaNova Systems

Marshall Choy 

Marshall Choy is Vice President of Product at SambaNova Systems and is responsible for product management and go-to-market operations. Marshall has extensive experience leading global organizations to bring breakthrough products to market, establishing new market presences, and growing new and existing lines of business. Marshall was previously Vice President of Product Management at Oracle until 2018, where he was responsible for the portfolio and strategy for Oracle Systems products and solutions. Prior to joining Oracle in 2010, he served as Director of Engineered Solutions at Sun. During his 11 years there, Marshall held positions in development, information technology, and marketing.

Published Tuesday, November 17, 2020 7:35 AM by David Marshall
Comments
There are no comments for this post.
To post a comment, you must be a registered user. Registration is free and easy! Sign up now!
top25
Calendar
<November 2020>
SuMoTuWeThFrSa
25262728293031
1234567
891011121314
15161718192021
22232425262728
293012345