As data analysis becomes more complex — with people looking to sort through, process and visualize information from massive volumes of data — traditional computers are proving ill-suited to carry out such deep learning / artificial intelligence work. Graphics processing units (GPUs), however, are a great fit for these applications. Today, Cybera is pleased to announce that it has made GPUs freely available to all Albertans on its Rapid Access Cloud.
The Rapid Access Cloud provides Alberta-based innovators, researchers (and those who are just curious) with free, instant access to online resources to build their own cloud computing environment. Following a successful pilot program, GPU instances can now be accessed for short-term use by all Rapid Access Cloud users.
Greg Burlet is an Edmonton-based entrepreneur who was an early user of the Rapid Access Cloud’s pilot GPU program. He wanted to use technology to solve an issue common to many musicians: how to take a song composed on a guitar and transcribe it into sheet music for others to read. Burlet theorized a solution that would allow a recorded song to be analyzed by a computer algorithm. This algorithm would use deep learning to identify the exact notes of the tune, and transcribe those notes into written music.
But to prove his idea could even work, Burlet needed access to powerful GPUs to test the computer algorithm. After four months of building the technology, and verifying its accuracy on the Rapid Access Cloud, his company, Frettable, is now preparing to launch its first iOS beta app.
“I didn’t know if my idea would scale up to a minimum viable product, and had priced out the GPUs I would need to buy at $49,000 (plus power constraints and labour), which was an infeasible amount of money to spend at that point,” says Burlet. “The Rapid Access Cloud helped me prove my idea could work, and we’re now looking to launch our product this spring. So this has been an incredibly valuable tool for me!”
Similarly, Ryan Hayward, a computing science professor from the University of Alberta, needed access to GPUs to solve another theoretical problem: how did the AlphaGo artificial intelligence system beat humans at the game of Go (a result once deemed impossible)? And could this solution be applied to the similar game of Hex?
Hex is a strategy game involving two players trying to create an unbroken chain of tokens from one end of the board (a hexagonal grid) to the other. For a computer to master this game, it must use image recognition and a tree-search algorithm to plot each step in the game. “We are trying to replicate the AlphaGo system in a more open environment, to make their algorithms available to a more general audience,” says Hayward (pictured middle with Jakub Pawlewicz and Aja Huang, co-authors of the award-winning Hex player MoHex).
Hayward’s students used the Rapid Access Cloud GPUs to build and test these algorithms. Aside from creating a “super-human” artificial Hex player, he says their work could help solve a variety of complex problems that require deep learning, such as energy regulation.
“Games have long been used to test the top minds among humans, and they are a convenient testbed for new algorithms,” says Hayward. “This project is about demonstrating an artificial intelligence that could have long-lasting social implications.”
Cybera is funded by the Government of Alberta and members to accelerate the adoption of advanced technology in the province. In one of its core roles, Cybera manages CyberaNet, the ultra high-speed broadband network that connects Alberta’s post-secondary institutions and business incubators to the international system of research networks. Over 650,000 Albertans now connect to CyberaNet.
What is the Rapid Access Cloud?
- The Alberta-based Rapid Access Cloud is funded by the Government of Alberta to:
- Help researchers utilize cloud computing environments for testing, analysis and experimentation
- Give Albertans a competitive advantage by offering a staging ground to test their ideas for cloud-based services, before moving to a commercial cloud platform
- Provide a learning environment for people to explore the possibilities and benefits of cloud computing
What are Graphics Processing Units (GPUs)?
A Graphics Processing Unit (GPU) is a special purpose processor, optimized for running multiple tasks or calculations simultaneously, which is useful for creating computer graphics, as well as running deep learning, analytics and engineering applications. (Note that these jobs require custom code capable of leveraging GPU resources.)
What GPUs are available on the Rapid Access Cloud?
The GPUs available are Tesla K80 GPUs, each of which has 2,496 CUDA cores and 12GB of on-board RAM. Each default allocation will allow a user to launch 1 virtual machine capable of accessing 1 GPU.
Who is eligible?
The Rapid Access Cloud, and its GPUs, are available to any Albertan who would like to use the cloud computing environment for pre-production, pre-commercial research, experimentation, or testing purposes.
How do I sign up?
New users can sign up for the Rapid Access Cloud using a Google sign-in here. Existing users can simply launch a GPU enabled instance through the Rapid Access Cloud dashboard. Click here to read the background documentation.