This article is more than 1 year old

Looking for silicon that’s actually designed for AI development? Look to the cloud first

G-Core unwraps Graphcore’s IPU as a service

Sponsored Post If you’ve got an eye on the future, you’ve probably also got an eye on AI and machine learning. Which means you’ve also got an eye on the considerable infrastructure challenges that come with adopting them.

Repurposing your existing CPU-based infrastructure is unlikely to be feasible. Meanwhile, the GPU architectures touted as enabling the AI revolution are both expensive and in short supply. And while GPUs may excel for training AI models, their advantage when it comes to inference is not always so clear cut.

But what if you could access a silicon architecture designed explicitly for AI training AND inference, with an accessible and open tool set?

And what if you could access this in the cloud, bypassing the infrastructure challenges that come with shoe-horning GPUs into your datacenter or ramping up CPU capacity?

That’s the promise of G-Core Labs’ AI Infrastructure as Service which brings Graphcore’s Intelligence Process Unit to the cloud in Europe for the first time.

The massively parallel Graphcore IPU architecture is designed from the ground up for AI training and inference.

Graphcore’s offering ranges from the IPU-POD16 to the latest Bow Pod256, which packs 256 of Graphcore’s latest generation Bow IPUs, capable of delivering large language training runs in hours or even minutes, rather than months or weeks.

The platform supports a comprehensive range of tooling and integrations, including data tools like Spark, PySpark, Kafka, and MongoDB, development tools including Visual Studio Code, Jupyter, RStudio and PyCharm, and ML and AI platforms such as TensorFlow, Keras, PyTorch, and ONNX.

Because this is a cloud service, you can get up and running quickly, backed by G-Core Labs’ 24x7 skilled technical support, 99.9 per cent guaranteed uptime, and S3/NFS storage which offers free egress traffic for public or hybrid solutions.

So instead of focusing on solving infrastructure challenges, you can concentrate on applying Graphcore’s silicon to your specific AI and machine learning challenges, whether that’s proof of concept, training, inference, or full-scale deployment. All on a European-based cloud, meaning fewer data sovereignty concerns.

Graphcore's Bow Pod64 is available now with the Bow Pod256 coming on stream soon. And after registering in your personal account, you could be getting a taste of what a truly AI-targeted architecture can deliver right now.

Sponsored by G-Core

More about

TIP US OFF

Send us news