In co-operation with Cirrascale Cloud Services, Graphcore is making available its Mk2 IPU-POD scale-out cluster, offering a simple way to add compute capacity on-demand, without the need to own and operate a datacentre.
When it comes to AI there are a unique set of computing challenges; model size is growing rapidly, and the bar for accuracy is constantly being raised. If customers are to take full advantage of the latest innovations, they need a tightly integrated hardware and software system built specifically for artificial intelligence.
Graphcloud is a secure and reliable IPU-POD family cloud service that allows customers to access Graphcore’s Intelligence Processing Unit (IPU), as they scale from experimentation, proof of concept and pilot projects to larger production systems.
There are two Graphcloud offerings available at launch, with larger scale-out systems being added in coming months:
- IPU-POD16 - delivering 4 petaFLOPS of AI compute (16 GC200 IPUs in 4 IPU-M2000s)
- IPU-POD64 - delivering 16 petaFLOPS of AI compute (64 GC200 IPUs in 16-IPU-M2000s)
IPU-POD systems deliver a lower total cost of training with improved time to solution compared to the latest GPU systems.
Graphcloud system instances are pre-installed with Poplar and systems software. Sample code and application examples are available locally for state-of-the art models used in Graphcore benchmarks, including BERT, EfficientNet, among others. Users also have access to comprehensive documentation to help them get started quickly with multiple frameworks, including PyTorch and TensorFlow.
Commenting on the launch of Graphcloud, Graphcore CEO Nigel Toon said, “With Graphcloud, it has never been easier to start using the IPU - whether you’re evaluating our hardware and the Poplar software stack for the first time, or scaling-up your AI compute resource.
“We’re delighted to be working with Cirrascale to bring Graphcloud to the world. It is another important part of Graphcore’s now comprehensive offering to customers, along with our global partner program, in-house sales and engineering support.”
