Say hello to production-ready Graphcore IPUs

a month ago   •   1 min read

By Daniel Kobran

When we first introduced support for Graphcore IPUs, the goal was to provide a free-to-use sandbox environment for ML Engineers to test-drive this groundbreaking new chip. The response to this offering has been overwhelming. Today, thousands of workloads have run powering all sorts of projects from NLP to GNNs.

Gradient is geared toward machine learning developers at any stage of the ML development cycle. The free plan is designed to test state-of-the-art models and prototype new ideas. When it's time to move beyond the concept phase, Gradient offers access to larger machine types, increased concurrency, collaboration, and other tools that make productionizing models possible. Continuing to build on the successful launch of free IPUs, the obvious next step was to make them available in a way where users could leverage them when building and scaling production applications.  

Starting today, we are excited to share the Graphcore IPUs are now available without the constraints of the free tier. You can access a full range of IPU systems including Bow Pod16 delivering 5.6 petaFLOPS of AI compute. One of the most requested features, to access IPUs on a team workspace, is now publicly available so that any organization can collaborate on their IPU-enabled models.

To learn more about IPUs, including information about sample models, the Poplar SDK, or systems specifications, head over to the Graphcore web page. You can get started by creating or accessing your account:

Additionally, to make onboarding to the IPU fast and seamless, we are offering engineering support to qualified organizations. For example, our team can help you migrate your models from CPUs or GPUs to IPUs. Contact our Solutions team to learn more.

Spread the word

Keep reading