Data Factory Infrastructure
The Data Factory is currently in a pilot phase and is under constant development. The Data Factory provides storage and compute to enable our partners and projects to explore AI algorithm training as well as serves as a testbed for our partners to explore data factory solutions.
Since 2019, we have been operating a pilot Data Factory in the Gothenburg-area. Initially, it comprised one DGX-1 through a donation from our partner CGit. This enabled us to provide 1 PetaFLOPS capacity (40 000+ cores) to our partners. In August 2020 AI Sweden upgraded the capacity to the new DGX A100 platform and we have access to 5 PetaFLOPS capacity (over 55 000 CUDA cores). A true state-of-the-art solution.
The current Data Factory implementation has the following specifications:
NVIDIA DGX-A100 comprising
- 8x NVIDIA Tesla A100 GPU:s
- Performance: 5 petaFLOPS
- GPU memory: 320GB
- NVIDIA CUDA Cores: 55296
- System memory: 1TB
- PCle Gen 4
- OS: Ubuntu (Nvidia BaseOS)
Tier 1 storage | 11 x 17TB NVMe FlashBlades
- Total capacity: 153TB
- Performance 11GB/s throughput
- ∼ 1,1 Million IOPS
Tier 2 storage | Access to over 2PB of local archive storage
Tier 3 storage | 2x Arista 7280 switches
- 100GbE deep buffer (8GB buffer)
- Up to 3.6Tbps "wire speed performance"
In-depth specifications of the DGX A100 can be found here.
Engage in the Data Factory
If you are interested in more information about the technical setup of the current Data Factory, want to know more about how to get access to the Data Factory and datasets, or are interested in sharing a dataset or a model, please contact us. You can also reach out to our partner AIXIA which hosts the current infrastructure of the Data Factory.