At NeurIPS this week, we have been talking about how our IPU-Processor and our IPU-Processor software, which we call Poplar®, were developed together from the ground up for machine intelligence workloads. Just as we have a highly differentiated processor architecture, so we have highly differentiated software.
The Poplar Graph tools contain breakthrough technologies that allow us to provide much more parallel systems than any one else. Poplar is just C++, with Python bindings, so easy to use.
Frameworks like TensorFlow help us to describe high level machine intelligence graphs. The Poplar toolchain takes a high level graph and translates it to a full compute graph that it can then map to our highly parallel IPU-processor.
Dave Lacey, VP Customer Engineering, gave a quick overview of Poplar in this video from NeurIPS 2018 for all those folks who couldn't make the conference.
We love that hat, Dave.
