Bitfusion’s New Cloud Adaptor Lets Developers Use GPUs And FPGAs In The Cloud

Disrupt Battlefield alum Bitfusion wants to help developers get to best compute performance out of their hardware, be it their legacy desktops or specialized high-performance compute clouds.

With Boost, Bitfusion showed how it can speed up existing applications by optimizing their libraries for the hardware in a given machine. After testing Boost in private alpha for the last few months, Bitfusion has now moved the service into beta and you can sign up for it here.

Bitfusion co-founder and CEO Subbu Rama and COO Maciej Bajkowski tell me that a lot of its early users were especially excited about using Boost for machine learning and data science projects. For now, Boost only works with a number of specific applications (Blender, ImageMagick, Octave, Matlab, Torch, for example), but the team plans to add more over time.

The company sees Boost as its flagship product. The company, however, also now offers the Bitfusion Cloud Adaptor — and in many ways, it’s really the more exciting product.

2015-11-24_1032The Cloud Adaptor allows developers to write apps on their local machines (and still access GPUs and FPGAs in the cloud. The applications are made to believe the laptop plays host to a lot of exotic hardware, but the actual devices live in the cloud.

“We launched Cloud Adaptor because it was the next logical step after Boost,” Rama told me. “With Boost, we offer in-node acceleration. But if you don’t have the right devices in your machine, you’re up a creek.” With the Cloud Adaptor command-line tool, developers can access GPUs in the AWS, Rackspace and Softlayer cloud, as well as FPGAs in Bitfusion’s own cloud it developed in partnership with Rackspace.

Most developers don’t exactly have easy access to an FPGA, but with this new service, they can easily use the cloud to access one (or at least not until Intel launches its first Xeon chips with built-in Altera FPGAs).

The difference here is obviously that this is for new applications that developers write to take advantage of this exotic hardware. Developers can pay based on work done or API calls.

Until now, Rama and Bajkowski argue, developers were able to code their apps for GPUs and FPGAs on their machines but weren’t able to test them. Now, they will be able to use a simple tool to run their applications in the cloud and instead of having to invest in exotic hardware they only when they run their applications.

Bitfusion believes giving developers easy access to high-performance computing will open this market to a lot of tinkerers who would never have had access to this hardware without the service.