Pytheia Case Study

Cloud GPU vs On Premise - Pytheia Leverages Cloud GPUs to Successfully Power 3D Tracking Software

The Background:

Founded in 2021, Pytheia has developed 3D tracking software that can make any standard surveillance or traffic camera intelligent. Their 3D tracking software leverages AI-as-a-service solutions that enable traffic cameras to build 3D maps that can identify object position, orientation, size, and estimated speed, and track multiple objects over time to allow users to understand what a camera sees in real-time. This data can then be analyzed based on the user’s needs and can be sent to autonomous vehicles for roadway perception.

Pytheia is part of Georgia Institute of Technology’s ATDC program, which assists entrepreneurs across the state. As their software and solutions continued to grow, the team recognized they needed to move to the cloud. Pytheia’s ATDC advisor connected the team with OVHcloud US to join its startup program to leverage cloud mentorship for long-term growth. Pytheia shortly after joined the OVHcloud US Startup Program as its founding member.

The Challenge:

Prior to joining the OVHcloud US Startup Program, Pytheia was using personal computers to run their 3D perception software. However, their personal computers did not have enough processing power for the company to continue developing their AI programs. The team quickly recognized they were ready for the next step and needed to accelerate data processing to improve their AI solutions. Pytheia needed to have access to OVHcloud high-powered GPU servers.

The Solution:

Pytheia had an introduction call with an OVHcloud US engineer who helped Pytheia plan how they needed to leverage GPUs and set up their environment while leveraging startup program credits. To best fit their needs, Pytheia was set up with a t2-45 GPU instance, which market price would be over $1,100 per month. Thanks to the OVHcloud US Startup Program, Pytheia was able to leverage this powerful instance for the price of a ‘standard’ balanced b2-30 instance.

Nine months into the startup program, Pytheia used 100% of its cloud credits and has only been paying a little bit more than $100 out-of-pocket each month for premium GPU access.

“To emphasize how useful those credits were may not seem like a lot, but to us, they were unbelievably useful,” said Pytheia Cofounder and CEO Mark Mote. “The thing we found distinctive about OVHcloud US is that they do not make you jump through 1,000 hoops to access the credits. Quite frankly, I do not think we would have been able to do it with another cloud provider. Additionally, OVHcloud US loved that we were using the credits for our type of work. We do a lot of machine learning, model training, and interference, and that comes with the requirements of GPUs and particularly data center GPUs.”

With these t2-45 GPU instances, Pytheia was able to continue to develop and improve its solution in an easy-to-manage environment. Along the way, Pytheia expanded their needs to OVHcloud US object storage for dataset management and fed that into their GPU instance for training and interference.

“We could easily manage everything we wanted and needed. Sometimes with other cloud services, when I want to do something that should be relatively simple, I would need to add another service to my environment to manage that. With OVHcloud US, there are no unnecessary additions or requirements, and you can do everything through the management console for the instance that you want to do,” said Pytheia Cofounder Ben Mains.

Using OVHcloud US, Pytheia was able to create an instance without needing to train the instance on a different machine. This enabled Pytheia to prototype new instances fast and hook them up to a website to see how the prototypes performed.

“When we were using our own GPUs, we'd have multiple instances running, we'd be doing the training somewhere else, and the inference in a third location. We would then have to pass the data back and forth with the server. Having a data center server that we could reliably host on with GPUs has been very helpful. We do not have to figure out the middle piece of getting two separate instances to talk to each other,” continued Mains.

In addition, Pytheia also leveraged the OVHcloud Public Cloud product line for a more scalable and efficient instance use case.

“Earlier in my career, I used OVHcloud bare metal servers and was happy with them, but Public Cloud was the best strategic option for Pytheia: instances require less work and are more scalable, which is key for us as we continue to build our team and company,” said Mains.

Thanks to OVHcloud US, Pytheia has been successfully developing its 3D mapping software at an affordable price, while leveraging high-performing solutions and having access to the OVHcloud US Startup Program team.

“The startup program has been great, and OVHcloud US help and [Startup Program Lead Francois Giraud] have been very easy to communicate with throughout this process,” said Mains.