There's a lot of hype around artificial intelligence, and it can be hard to sort the signal from the noise.

So why not talk to a real live human? Give us your info and we'll contact you right away.

Contact Us

What’s A GPU Camera And Why Would You Want One?

A “GPU camera” is an AI capable video camera with an onboard graphic processing unit (GPU). These cameras are often used for “fog computing,” or for computations that can’t take place in “the cloud.”

Or, put another way, a GPU camera is an AI capable camera that can process images and video all by itself. No connection to the cloud required!

How A GPU Camera Works

Also called an “edge camera” or a “fog computing camera,” a GPU camera is ideal for AI applications that require a camera which can operate in areas where there isn’t a good connection to the cloud. If, for example, you wanted to put an AI vision system in a remote area of wilderness to try and track the migration of an endangered species, you would need a GPU camera.

Or, more conventionally, if you wanted to put an AI camera up on top of a power pole, but you didn’t want to bother running network cable, you could use a GPU camera.


A GPU camera can operate independently, with no need to be constantly connected to the cloud.

While GPU cameras can be used in any setting, it’s an especially effective solution in places where a steady internet connection isn’t available. GPU camera applications are limitless, but some more common applications include:

  • Counting, tracking, and/or identifying wildlife
  • Track or count people in outdoor areas (like an outdoor shopping mall)
  • Remote monitoring of everything from oil wells to city streets without a big investment in infrastructure for each camera placed

GPU cameras are not the perfect solution for every AI vision system, but they’re a great tool for a great many AI vision applications. Especially applications that are a little outside of the norm or require a more sophisticated algorithm.

The Benefits Of Onboard Processing

The vast majority of AI vision systems rely on cloud data storage and processing. Typically:

  1. A “dumb” video camera is installed wherever video and/or images are needed, along with a high-speed local network that the camera is plugged into.
  2. The camera(s) is networked to a single workstation or server, which then transmits data to a remote server for processing and storage “in the cloud.”
  3. The processed data is often (but not always) returned to the single server/workstation for local review.

This system is not particularly efficient, as data is essentially transmitted twice (once from the camera to the local network node, then again up to the cloud). Additionally, if there’s a network error or a cloud connection error, the entire vision system can fail.

GPU cameras are different:

1. Reduced Or “Distilled” Data

A GPU camera like the DNNCam™ can be configured to store a considerable amount of data locally, which can then either be bulk uploaded if/when a network connection is available, OR stored for extended periods until someone can download the data in the field.

If you have a GPU camera monitoring wildlife out in the wilderness, for example, it can store data for days or weeks until someone can visit the camera and download stored data.

NOTE: The DNNCam™ is available with 4G and satellite modem modules, so no remote retrieval is required. Not all GPU cameras offer this feature.

2. Robust Structure

If you have a network of standard ‘dumb’ cameras, a network failure crashes the entire system. With a network of GPU cameras, one camera failure doesn’t break the entire system. Additionally, each GPU camera can independently observe the same data, allowing for redundancy in situations where accuracy is paramount.

3. Reduced Infrastructure

While GPU cameras often have a higher up-front cost than ‘dumb’ AI capable cameras, they can create big savings when long-term infrastructure costs are considered.

If, for example, an AI vision system needs to be installed in an area without a lot of infrastructure, the costs of providing power and high-speed data to that system can be substantial over a period of months or years. With a GPU camera like the DNNCam™, it’s possible to use a cellular or satellite modem to transmit algorithm results, and a simple solar power system for power.

The DNNCam™ Is Boulder AI’s GPU Camera

DNNCam

The DNNCam™

While we’re a bit biased, we believe the DNNCam™ is the best GPU camera on the market. DNNCam™ features include:

  • IP rated protection from weather: With a maximum rating of IP69K, the DNNCam™ is capable of functioning in nearly any Earth environment. We even have a system for deploying the DNNCam™ underwater.
  • Passive cooling system: Most electronic devices have some sort of active cooling system that circulates air to control temperature. This means that most electronic devices are vulnerable to dust and moisture. Because the DNNCam™ uses passive cooling, we’re able to seal the camera from the elements.
  • Low power requirements: A good GPU camera needs to draw as little power as possible, especially if it’s deployed remotely and reliant upon a simple solar cell and battery combo. This describes our camera to a “T.”
  • Onboard storage: When a GPU camera is deployed without network connectivity, data storage is crucial. The DNNCam™ is currently available with 32GB of onboard data plus a MicroSD storage card slot.
  • High-def resolution: Camera resolution is often crucial to algorithm performance. With 4K resolution and low distortion, the DNNCam™ is built to provide the best possible imagery – even in less than optimal conditions.

Again, we’re biased when it comes to the DNNCam.™ Still, we believe our combination of features is unparalleled. To learn more about our GPU camera the DNNCam™, visit our hardware page here.

Darren Odom

The founder of Boulder AI, Darren’s engineering resume includes consulting for leading-edge research companies, product design and development, custom software application development, and more. Darren holds numerous patents, and is a recognized expert in the fields of deep learning and AI hardware development.