The DGX-2 integrates 16 Nvidia v100 Tensor Core GPUs in one chassis that is not larger than a small fridge. It is the world's first single server to have a calculation speed of two petaflops. The system is designed such that complex models can use the combined power of the 16 GPUs at once without sacrificing performance. When training new AI systems, researchers often have to wait several hours to get a result. A good combination of CPUs (central processing units) and GPUs (graphics processing units) is crucial in this regard. As the DGX-2 is also equipped with a dual Intel Xeon Platinum CPU of 24 cores each and has 1.5TB of RAM onboard, jobs that are in need of both CPU and GPU performance, can easily be run on the DGX-2.
The DGX-2 at IDLab Antwerp is part of the broader GPULab testbed, which is a testbed set up jointly by IDLab Antwerp and IDLab Ghent. As this is joint infrastructure, any GPULab user on the DGX-2 can also gain access to the rest of the GPULab testbed.
Researchers and industry can access the DGX-2 infrastructure through a Fed4Fire project, enabling them to collaborate on deep learning, human-like AI etc.
- The Flanders AI Research Program focuses on leading-edge AI research and pioneering proofs-of-concepts for healthcare, industry 4.0, governments and their citizens. IDLab Antwerpen is involved in the Flanders AI Research program, where several researchers are leveraging the power of the DGX-2 to conduct their research.
- Smart Waterway is a project on enabling autonomous barges in urban waterways through low-cost sensor systems. The DGX-2 has been used by researchers to train object recognition models to be able to distinguish ships and bridges.
- AI training
- Object recognition
- Human-like AI