10 Apr Pure Storage And NVIDIA Announce AIRI Converged Infrastructure Reference Architecture


AIRI (AI Ready Infrastructure)

Last month, I attended NVIDIA’s GTC 2018 conference, which has turned into one of the premier AI, ML and DL conferences in the world. You can read my analysis here. Part of the reason for this reputation is NVIDIA’s leadership in machine and deep learning training which it helped to create and popularize using its highly programmable and performant GPUs.

With less fanfare and credit, Pure Storage has taken a similar approach to storage in AI, ML, and DL, and have brought out AI-optimized FlashBlade storage systems to take advantage of the parallelized AI workflows. An AI solution without the right storage is like a fiddler crab with one giant claw- it can only get around as fast as that one, giant claw. We have previously covered this potential imbalance in a blog and a paper that both talk about the importance of parallelized storage when it comes to parallelized AI workloads.

NVIDIA and Pure Storage are now pulling their combined “super-powers” together to a create what we normally expect from HPE, Dell EMC, IBM, and Lenovo- that is, a converged infrastructure platform, called “AIRI”. AIRI (AI Ready Infrastructure) is an AI-optimized converged infrastructure reference design, available from Pure and NVIDIA partners and contains different combinations of Pure Storage FlashBlades, NVIDIA DGX-1s, and Arista switches.

AIRI is targeted at enterprise, not CSPs, who design and then get ODMs to build their own systems. Holistically, enterprises have two choices to get into AI. They can choose on-prem systems like AIRI or from vendors like HPE, Dell EMC, IBM, and Lenovo or use AI cloud services from AWS, Azure, GCP or IBM Cloud or Watson. Enterprises do not want to fall behind like they did with the public cloud, inevitably costing them both time and efficiency. Amazon.com AWS had a seven-year public cloud head-start, and many enterprises were caught flat-footed with the private cloud. Therefore, enterprises are hoping to get a much quicker start on AI applications that help to address security, cost or latency problems, and should be local and not in the public cloud.

(Source: Patrick Moorhead)

AIRI is not a specific device, delivered to the door from Pure Storage or NVIDIA, it is a reference architecture , meaning that it has a ‘cookbook’ with the full meal built and served by an authorized AIRI reseller. The current AIRI authorized partners are World Wide Technology, Trace 3, FusionStorm, ePlus , Groupware Technology , PNY Technologies and XENON Systems . These resellers may not be household names but are well known in the HPC and AI space.

The benefit of a reference architecture is that the combined Pure Storage, NVIDIA and Arista system is specified and tested to operate in a high-performant and compatible manner with NVIDIA machine learning software. Given that it is delivered by partners, your mileage will vary, of course, like any reference architecture. One of the benefits of converged infrastructure systems from the biggest OEMs like Dell EMC, HPE and Lenovo is that they will deliver it to your doorstep, warrant and service it, but OEMs require many, many months to test and validate, which is why I believe Pure Storage and NVIDIA are doing this.

I really like what Pure Systems and NVIDIA are kicking off with AIRI. AI is difficult enough for enterprises to grasp and the last thing they need to worry about is how well the piece parts work together from a performance and compatibility standpoint. With AIRI, you do get both Pure Storage’s and NVIDIA’s promise, but at the end of the day, it will come from the HPC and AI resellers detailed above who need to build and service it.  I also like the time to market element of this as I expect future AIRIs to deliver the latest FlashBlades and the latest DGX models. I’m not concerned about AIRI starting with DGX-1 as a single DGX-2 is $399K and not available yet. It’s not as if DGX-2 replaces DGX-1, DGX-2 is a higher-level configuration, costing more. AIRI is a good start and I am looking forward to getting even more details as they are available from Pure and NVIDIA as well as talking with some AIRI customers.

Note: Moor Insights & Strategy writers and editors may have contributed to this article.