New York City-based Infrastructure-as-a-Service provider Paperspace has announced Gradient, a suite of tools designed to give companies of any size and their developers access to an Uber-grade machine learning tool stack. With Gradient, developers can easily run code on a full GPU cloud, accelerating the development of machine learning-based solutions. There’s also a channel play.
Paperspace is a three-year old startup that was created out of the idea that GPU Compute will drive cloud computing for the next ten years.
“Paperspace makes it easy for developers to integrate GPU Compute into their workflow,” said Dillon Erb, Paperspace’s CEO and co-founder. “GPUs have traditionally been used for gaming and animation. However, as CPUs are coming up against the end of Moore’s Law, GPUs are being used for accelerating applications that can take advantage of the parallel architecture. This leverages the fact that they can do a lot of computations very quickly, which enables you to train machine learning much faster.”
Erb noted that this isn’t the kind of thing where you can just plug in any application and have it become parallel.
“We provide a front end to the GPU Compute,” he said. “It’s a complicated infrastructure that we manage, so that companies can start writing machine learning pipelines as quickly as possible. It’s not sufficient just to have access to the resources. You need tools. We make it easier to get started without having to enlist a DevOps team. It makes developing a modern machine learning pipeline as easy as developing a modern web app.”
The Gradient tools include one-click Jupyter notebooks, a powerful GPU job runner, and a new integration that lets you add 1-line of code to run any code on a full GPU cloud. In just a few clicks, users can begin training and deploying models, leveraging popular frameworks like TensorFlow or PyTorch. Jobs can easily be versioned and reproduced and can run in parallel for rapid iteration.
“Until now, Paperspace has been focused at a lower level, around GPU management and orchestration,” Erb said. “This adds a higher level of abstraction – a layer on top. Out of the box, you can now plug into a sophisticated tool stack. Gradient allows companies of any size to have access to an Uber-grade machine learning tool stack. Going forward, it will be our flagship offering.”
The biggest competition in Paperspace’s early days was homegrown solutions, but because of the scale of the problem, and the opportunity, other solutions have been emerging from traditional Big Data and data science companies.
“The big cloud providers are interested, and Google has their take on it,” Erb said. “Every startup has to stay ahead of the curve with the big public cloud providers. They have hundreds of offerings, and they take you 80 per cent of the way there. You can go further though, if you have resources to leverage, and we can take it 100 per cent of the way there. We run Gradient on top of the public cloud in many places. Long term, our differentiator is that we build out the entire stack.”
Paperspace’s go-to-market model at its core is based on the GitHub model, onboarding through the developers.
“Getting companies to decide to integrate this into their workflow has been about getting in front of developers, so we work with large online courses and through universities,” Erb said. “Anyone can sign up for the basic version, but there is an enterprise tool stack.”
Paperspace does have a channel program, and they are expanding their channel as well.
“We work with MSPs and VARs, because partners are also an important way to get into businesses,” Erb said. “With Gradient, we will roll it out to MSPs and other channel partners as the next step.”
Erb said they are continuing to get requests from the channel and would like to get more.
“We would love for them to reach out,” he stated. “The channel is also seeing that companies are looking for solutions that will make it easier for them to become machine learning-enabled.”