Making The Rounds: Part 1 — GE's HPEC CoE

9 July 2014
figure-1.jpg

I recently took over the reins as the GE Intelligent Platforms marketing leader for Embedded - Military/Aerospace computing and beyond. I’m excited about this because I get to broaden my focus beyond military networking and communications. I'm now involved in telling the story for all our embedded products, services, customers and markets. This is going to be a lot of fun!

I’ve begun making the rounds, spending time with all our product managers and marketers to gain a deep understanding of their products, roadmaps and customers. I started with our High Performance Embedded Computing Center of Excellence (HPEC CoE) conveniently located an hour up the road from my Foxboro, MA office in Billerica. This is the first in a series of posts in which I plan to share my travels.

What is an HPEC CoE anyways?

First and foremost, I feel compelled to outline the correct pronunciation of Billerica. It does not rhyme with "America" and it is not "Bill-er-icka" (as most out-of-towners pronounce it). Pretend you don't see the "e" in the name - Bill-ricka-ah (or if you want to sound like a real local - Brick-ah). It is located north and west of Boston surrounded by high-tech companies large and small. It also happens to be conveniently located close to the universities and research labs that produce exactly the kind of engineers you need to hire when setting up a High Performance Embedded Computing CoE.

So what exactly does a HPEC CoE do? In a nutshell, it helps customers take very high-performance, computationally intensive applications, often running on racks of air-conditioned servers, and deploy them in real-world embedded, often remote, harsh environments such as radar installations and ISR platforms in the battlefield. That’s it: take a high performance data center closet and shrink it to the size of a shoebox that can be installed in a rugged ground vehicle or a military aircraft.

That’s what it is. But why did we build it? Because we have a bunch of engineers who like to play with cool computer hardware and software? Well, no. Here’s the thing. We did it to help our customers save time, save money, minimize program risk and get their sophisticated solutions out there faster than they would otherwise do. Really.

A high performance computer in a shoebox

The fact is: some signal processing applications like radar are really processor-intensive. Then there is the whole Connected Battlefield thing; modern military forces are more technically advanced, more reliant on data, and more connected. Military machines are getting smarter. All those autonomous vehicles, weapons systems, C4ISR platforms, voice and high-definition video feeds need to talk to one another. They do this using the same big data crunching, cloud computing infrastructure found in the commercial world. But, unlike the commercial world, they don’t always have the luxury of an always on high speed communications network - so pieces of the cloud or the data center must sit out on the edge of the network, close to the clients that need them. If the soldier can’t get to the cloud infrastructure and all of its computing resources, bring the cloud to the soldier.

HPEC engineers get creative and resourceful

A GE CoE is just that - a center of excellence, and it’s deigned to help our customers rise to those challenges. The HPEC CoE houses a team of computing experts - engineers who live and breathe computing. The team has developed an in-depth understanding of specialized processors – Intel, PowerPC, ARM, many-core processors, GPUs, digital signal processors, SoCs and so on - and the features/advantages/disadvantages/trade-offs they bring to the table. They study the ins and outs of all these processor architectures developing neat tricks to squeeze out just that bit more performance without increasing power or heat. This takes a real knack - with a little help from some smart software tools.

The HPEC team gave me a demo of one of their lab systems. Customers often buy these off-the-shelf platforms to try out their applications. Once the applications are wrung out and tuned in this environment, GE rugged military systems experts take the architecture and turn it into a fully-deployable, certified production system. A typical lab system is approximately 10"W x 23"H x 14"D – so, larger than the target deployment platform, but a whole lot smaller than an IT closet. It can pack up to 16 compute cores and over 1500 GPU cores providing about 3TFLOPS of computing performance. A beast of a computer sitting on a lab bench.

The team used a high definition image processing application that generated enough compute load to demonstrate the system's capabilities. This is where those smart software tools come into play.  It's one thing to watch the HD video feeds and gauge performance by how smooth the video is – but it's another thing when you can visualize on the screen exactly what each processor and core is doing. This is what the AXIS software suite developed at the CoE is for (and it’s not just for our internal use, either – it’s available also to our customers). Using the AXIS tools, we could see each core in the system and how it was executing application code. Not only that, but we could zoom in and get very granular information about timing and dataflows - at least, that’s what the engineers told me I was seeing. It was  just a lot of colorful lines to me. AXIS is not just a visualization tool however. It also enables the software developer to compile and tweak code to optimize performance for the underlying multi-compute environment. For example, you might determine that your code is overloading some processor cores, while leaving others idle. A quick few button clicks, a recompile and now your code is optimized - balancing across all available cores.

The ah-ha moment - it's not just hardware.

To be honest with you, I never knew how difficult it was to write applications that run efficiently on high performance computers. I always thought it was a matter of throwing more processors or more cores at the problem. The engineers in our HPEC CoE see it differently. They work closely with their customers to understand their applications and architect the highest performance, but most efficient system platform from the get-go. Then they provide software tools such as the AXIS suite and time-savers such as specialized math algorithms that help customers port or write code quickly and with fewer bugs. Everything is built using a modular open systems architecture to take the whole platform from concept to deployed product in as short a time and with as little risk as possible. I now have a whole new appreciation for what those guys are doing up there in Brick-ah.

Want to see some of this stuff for yourself? We have a bunch of videos on YouTube  where you can see our HPEC systems and the AXIS software suite in action. We also have a few white papers that outline the applications and challenges HPEC systems are designed to address. Then there is our resident HPEC expert right here on the blog - you might have seen some of his posts.

So - the next time you need TFLOPS of computing performance in a small rugged system - you know who to call.

Rubin Dhillon

Rubin has spent over 20 years in the embedded computing world, in roles ranging from support to sales to product management and even garbage collector. He experienced the huge growth (and crash) of the telecom industry, and he's spent time dabbling in medical, industrial, transportation and military applications. Rubin figured he has so many stories to tell, he should get into marketing and so he is now our VP of Marketing. Connect with Rubin on LinkedIn and he'll explain the "garbage collector" story…