How AI and High-Performance Computing (HPC) Work Together Hand-In-Hand

Posted by Annie Paquette on December 27, 2019

Nhehdv3scqqccryunokw

As mentioned in our recent blog post, the increase in AI applications over the past 3 to 4 years, along with the push toward edge computing, has sent many data centers back to the network and computing drawing board. The supercomputers that support AI tax even the most robust network topographies and data processing platforms.  In turn, the mechanical and electrical infrastructures of new and existing facilities are being asked to do more and more, as are the ‘handshake’ devices such as switches and network PDUs that bring those services to the rack.

A recent sampling of articles such as this one from Data Center Knowledge would lead you to believe that today’s largest global issue is the supercomputer.  Forget about nuclear proliferation and trade wars, the real race is to see who can build the biggest and fastest computing machine.  It is the ultimate modern-day ‘mine’s bigger’ contest.  The most prestigious academic institutions are teaming with the largest computer manufacturers in the biggest countries, all vying to out-flop one another.

While these supercomputers are often designed for specific projects or tasks, their not-so-distant cousins, HPC or high-performance computing platforms, are used for merely übercomputing functions.  Once the realm of academic and government applications, HPC devices are finding their way into more corporate and commercial applications today.  This includes recognition as being the go-to platform for delivering the kind of analytics needed to support machine learning and artificial intelligence platforms. They are the answer to question, ‘where will my AI device go to learn and train?’

But as the data center managers who have brought these imposing racks of servers onto the floor well know, it ain’t always easy.

Most people will never forget their first HPC application.  It is one thing to read the specifications and solve new power and cooling challenges, and yet another to make it all happen.  “Sorry, but remind me – how do I bring 60 kW of redundant power and cooling to a single rack again?  And what was that?  This HPC device will have twenty of those racks?”  For most data center managers, it is like that moment when Buddy the Elf learns that he’s 915 Etch-A-Sketches off pace.  Seriously.  Everything needs power, but this is ridiculous.

In fact, it is easy to get caught up in that ‘oh boy’ moment that many forget that it is one thing to get the power there, and yet another to actually distribute it to all of those power-hungry servers.  Needless to say, that is what we are here for.  Our power experts have assisted customers with a lifetime of HPC installations. In addition to ‘Smart’ and ‘Intelligent,’ we should add the ‘Experienced’ moniker to our PDUs, except that it doesn’t really have the same ring, does it? 

At Server Technology, our industry-leading Smart PDUs are providing the solutions that power today’s higher performance computing applications. Our HDOT products are well-suited to high density enclosure and IT equipment racks. For companies large and small, our power strategy experts have been trusted to provide network PDU solutions for demanding data centers worldwide. 

Interested in learning about artificial intelligence and its relationship with other technologies, and the resources it requires? Make sure to read our new white paper, "Powering the Next Revolution: Artificial Intelligence".

How and Why AI is Used for Data Center Cooling