Partners RISE and Peltarion have already used the new infrastructure for projects connected to the Federated Learning and Swedish Medical Language Data Lab. Peltarion had great results with over 850% faster processing - a job that would take them 1 h 50 minutes on the machines they would normally use could now be done in only 13 minutes!
The A100 GPU:s outperform the V100 GPU generation, previously included in the Data Factory, greatly in almost every single usage area and provides several times the performance in many use cases. It comes with the new multi-instance GPU (MIG) technology that allows each A100 to be divided into multiple chunks of smaller GPU:s that act and perform like physical GPU:s. This allows for the DGX A100 platform to be split into a maximum of 56 GPU:s that each can provide up to the same compute power as a previous V100 GPU. This allows for great flexibility and the ability to adapt resources to workload, ensuring maximum usage of available compute resources.
“We at CGit are proud to be able to contribute to research and innovation together with AI Sweden. Through this platform, based on NVIDIA's A100 GPUs, AI Sweden will increase the computing power it can offer in projects by 20 times compared to today and thus greatly accelerate the exciting initiatives and projects it is now working on. The platform is one of the very first of its kind in the Nordic region and AI Sweden will be able to provide computing power on a scale that usually is reserved for supercomputing centers. CGit is also on-site with staff at AI Sweden and contributes with our unique competence and experience in the area of AI Infrastructure. “ says Mattias Bergkvist, CEO at CGit.
The new infrastructure will allow partners to use more compute resources for model training, run bigger hackathons, do extremely compute intensive projects and much more. It will also be a very important resource in the collaboration projects that AI Sweden runs together with partners.
In the upcoming challenge arranged in collaboration with AstraZeneca, the hackathon teams will be able to run on one A100 GPU each, which is an amazing opportunity for the soon to be selected teams. Hackathons are a great way for partners to make use of the infrastructure. The particular challenge that AstraZeneca is running aims to accelerate the drug development process through machine learning and could, as such, provide results that can benefit all of society.
Providing this platform with a full integration AI optimized infrastructure gives AI Sweden the ability to scale and support the need for resources such as networking, super high-performance dataset storage, long term storage, supporting CPU infrastructure and even additional GPU resources. The new infrastructure will also allow for future deployments of new infrastructure solutions, making it possible to provide solutions to complex problems such as security and isolation of data, toolchains and other infrastructure related challenges.
AI Sweden and CGit are already planning for a second A100 that is planned to be up and running in December 2020.
Want to know more?
Ebba Josefson Lindqvist
Project Manager Data Factory
+46 (0)732-54 29 03