The new Champollion supercomputer was built by HPE in collaboration with Nvidia using AMD-based Apollo systems equipped with Nvidia A100 GPUs.
Champollion combines high-performance computing and AI technologies
for machine learning and faster results. HPE already provides computing resources and AI systems at its facilities in Grenoble for customers and researchers. HPE also plans to provide access to Champollion for scientists around the world to speed up AI model testing and research.
Champollion equipment includes:
- 20 HPE Apollo 6500 Gen10 Plus server nodes ;
- 160 Nvidia A100 GPUs ;
- Nvidia Quantum InfiniBand network .
The HPE ML system starts with four nodes, but customers have the ability to scale. Each Apollo node will have 4 TB of memory and 30 TB of NVMe local storage with additional HPE Parallel File System Storage . The HPE Machine Learning development environment provides an integrated platform for creating and training models compatible with the popular TensorFlow and PyTorch machine learning platforms. According to HPE, Champollion is now available to a narrow circle of users and in the near future will be available to the entire research community so that users can start developing and training their models.
Recall that the DeepMind research laboratory has released an artificial intelligence system called Gato. AI is
a general-purpose system and is capable of performing 604 tasks, including creating a description for
images, participating in a dialogue, laying blocks with the help of a real robot arm and
playing an arcade.