Hardware

330 TERAFLOPS GPU Server
4 x NVidia 4090


625 TERAFLOPS GPU Server
20 x NVidia A10


140 TERAFLOPS GPU Cluster
16 x NVidia GTX 1080

Old school CPU cluster
To use for tasks that cannot be parallelized



■ The current equipment park consists of:

▪ GPU: 4 x NVidia 4090 - single server with a total compute power of about 330 teraflops (f32).
▪ GPU: 20 x NVidia A10 - single server with a total compute power of about 625 teraflops (f32).
▪ GPU: 16 x NVidia 1080 - cluster of servers with a total compute power of about 140 teraflops (f32).
▪ GPU: 4 x NVidia 2080 Ti, 1x NVidia 1070, 1 x NVidia 3090 - cluster of servers with a total compute power of about 100 teraflops (f32).
▪ CPU: 256 Core Dual AMD Epyc 7742 own built compute server, old school CPU cluster ‐ for tasks that are difficult to parallelize.
▪ CPU: 64 Core Dual Intel Xeon 6430 own built data server with 70 TB fast SSD array ‐ for data keeping and processing.

■ We also have opportunity to rent NVidia A100 equipment within the Amazon EC2 cloud for short, but capacious computations.

■ As the orders become more complex, the we plan to expand the equipment park at the Principality of Andorra.

■ We are planning to order more 4-8 of compute servers with 20 x NVidia Tesla A10 cards.

■ In next year we plan to start mastering IPU technologies and make trial purchases of Graphcore IPU devices.