Feeding AI hungry applications efficiently at scale

AI is dependent on data and needs to be fed with huge amounts of data – video, imaging, language processing. This also means you need an infrastructure that can handle large amounts of data very efficiently. At AI Field Day 3, the first presentation was from DDN Storage talking about how to achieve operational AI at scale. They gave a thorough overview of their solution which included customer examples.  Hearing from customers directly with innovative and real-life examples is always inspiring. In the session we heard from Cerebras Systems, who have produced the largest computer chip ever built AND the fastest AI processor on Earth.

We also heard how DDN over the last 4 years have established themselves as a serious AI data platform. They have built a complete and turnkey solution for the AI infrastructure. The platform makes their customers able to pursue AI at scale and their customers include some of the largest AI programs on the planet. During AI Field Day 3, DDN announced that their platform had been chosen for the largest AI supercomputer in Japan

Efficient AI at Scale

AI infrastructure requires a platform that can handle large amounts of data and deliver the data rapidly and efficiently whilst providing the right levels of security for the data. In addition, it’s also required to hold the data for long periods of time. When you combine all of these requirements, it can be a challenge to solve and ensure all of them are in a given solution.  This is where DDN stands out and claims to differentiate themselves from other vendors in the market.

With DDN A3I AI400X2 the AI applications can consume 90 GB/s and 3 million IOPS out of the box. One of the presenters, William Beaudin, Senior Director of Solutions Engineering at DDN, demonstrated real life examples of their customers data sets. DDN power some of the most data intensive workloads and largest ML/DL data sets in the world. The data sets require intensive performance as some of the training applications consume 1 TB/s from storage in real time. With all of the examples of data sets DDN showed, they all required speed at both read and writes. This is an important point and unfortunately some organizations miss this when they implement AI… and why they don’t succeed with their project.

The speakers from DDN shared that they have a unique technology that handles the performance on both read and writes in an optimized way. The platform is seen as a building block where you can add more performance as you go.

Cerebras Systems – The largest computer chip & the fastest AI processor

This was the part of the presentation where all the delegates got very excited (if you’re watching the on-demand presentations, it’s about one and half hour into DDN’s presentation). We heard from Rebecca Lewington, Technology Evangelist at Cerebras Systems Inc. The reason the company was founded was because of the growth in computing requirements that was expanding exponentially each year and to get rid of all the bottlenecks in the infrastructure. Cerebras systems decided to start from scratch and build a chip (with a latency of only nanoseconds!). We were shown the Wafer-Scale Engine (WSE-2) which is a central processor for deep learning computer systems and also the largest computer chip ever built AND the fastest AI processor on Earth.

The numbers are impressive – the chip has 850 000 Cores and 2.6 trillion transistors. It is 56 x bigger and have 123 x more cores than Nvidia A100!  

Cerebras customer portfolio demonstrates their value in the market and some of their customer cases were included in the presentation. They have customers doing research in a variety of areas. We heard about Glaxosmith research publications and also Argonne National Laboratory’s research in COVID-19.

Recommendations to learn more

The DDN presentation was very thorough, and I would recommend watching it when you have time, as it has very good insights on AI workloads at scale. If you don’t have time, I think my fellow delegate, Karen López summed DDN AI data platform up quite nicely with this tweet.

If you’re curious to know more about how DDN achieves AI at scale, you can visit the AI Field day page where you will find recordings and other content from AIFD3 on demand. If you have questions or comments to this blog post you can connect with me on Twitter or LinkedIN.

About Tech Field Day events

Tech Field Day is a series of invite-only technical meetings between influencers and sponsoring enterprise IT companies. Companies share their products and innovations through presentations, demos, roundtables, and more. Over 2-3 days, a panel of delegates interact with these different tech companies on-site in Silicon Valley or remote. The sessions are live-streamed and recorded – You can find the videos, podcasts and blog posts from Tech Field Day events on their website.

*Disclaimer: I am invited to participate as a Tech Field Day Delegate as a guest of Gestalt IT.  I did not receive any compensation to write this post, nor was I requested to write this post. The above post is written of my opinion and not that of Gestalt IT.