We sat down, virtually, with Thomas Hall, Head of Technology at Above to ask him about Above's experience with Proline by Exacta. Above required high-quality high-performance systems with immense processing power. Read on to find out what Above had to say...

Introducing Above

Above is a UK company at the heart of the digitalisation transformation within the global solar industry, critical in ensuring its rapid and sustainable deployment. Founded in 2016 by Will Hitchcock, they have worked on over 3500 solar assets and are now supporting customers across six continents. Working with many of the world’s leading solar energy companies to deliver solar plant data and analytics through their interactive platform, SolarGain. Above’s mission is to speed up the renewable energy transition through innovation, with the vision of zero-carbon energy for all.

What type of systems did Above use previously?

“At Above we used a handful of under-desk workstations for a mixture of CAD, photogrammetry and 3D reality modelling, and AI processing. They were single-CPU, therefore limited in core count and RAM. The most powerful of these had 4 NVIDIA A4000 GPUs. Our AI processing was distributed over a network of PCs in the office with standard GPUs.

The main reason for purchasing a new processing server was to have cores, RAM and GPU in a single machine for the purpose of processing very large photogrammetry/3D reality modelling projects quickly, and without hitting resource limits. We also wanted a Linux system that could be remoted into by various members of the team to check on progress, whilst they could continue CAD work on their own workstations. Finally, we wanted the flexibility to split server resources into multiple virtual machines (VMs) for processing AI workflows overnight.”

What software packages does Above run on these systems?

• Ubuntu
• Agisoft Metashape Pro (photogrammetry, 3D reality modelling)
• KVM Virtual Machine manager for Windows VMs
• Python, PyTorch (AI workflows – mostly neural network (NN) inference, but also NN training and non-NN computer vision and machine learning).

Since switching to Proline, how has Above's workflow changed?

“Being able to split the server into eight VMs, we have essentially added 8 powerful machines to our AI processing pipeline for running NN inference. Owing to the reliability of the dedicated server components, the efficient cooling, and the location in a comms room, these machines can be run 24 hours per day, rather than just overnight. For our workflows, this is much more efficient even than cloud processing, because our data throughputs are very high, so network and I/O speeds are a meaningful bottleneck.

In “beast mode” (bare metal on all 8 GPUs), the system has allowed us to process photogrammetry/3D reality modelling projects that were previously unimaginable for us. In fact, the projects we can process now, even significantly exceed what was possible by renting cloud-servers, because we can commit effectively unlimited RAM, and the U.2 NVME storage massively reduces I/O latency. Our new record is processing topographical (3D) mapping for an area of over 1500 hectares, requiring 40,000 images, and generating over 4 billion measurements of the Earth’s surface, with each point accurate to within the size of a ping-pong ball!

I’m confident the server can process even bigger projects because we still had plenty of resource headroom. I really like the fact we can ‘accelerate’ in the near future by upgrading GPUs, rather than replacing the entire system.”

Tech Summary from Proline System Architect Adam Lister

“It was a pleasure to assist Tom and the team at Above with their first step away from large deskside compute, and into the world of server-grade HPC hardware. As Tom intimates, they were looking for improved resource orchestration, and much more scalable, predictable, performance. The addition of the hypervisor layer enables them to fluidly assign resources to those that require it within their job schedules. And then, with the hypervisor bypassed in 'beast mode' they can share their workload across all eight GPUs via NVLink, and saturate all CPU cores and memory together, giving them previously unthinkable performance. We look forward to collaborating with Above as they grow their HPC cluster in the coming months and years.”

Learn More about Proline Systems 

Discover Above