|Description:|| In October we have held our first face-to-face workshop to discuss the goals and plans for our collaborative study of the use of GPU Tensor-Core Correlators for the ALMA telescope. The requirements are impressive (72 dishes delivering 32 GHz instantaneous bandwidth each) and pushes technology to the limit. The challenge is not in computing those correlations on GPUs anymore, but in getting the digitized data into the GPU systems, at a staggering ~100 Tb/s in total. |
The main question to be answered is: “can GPU technology be used for the extreme ALMA requirements within a fixed power budget of 300 kW and within reasonable costs?” The research necessary for this aligns very well with many of the HPC goals on the Smart Backend Research roadmap.
The collaboration is brand new and therefore meeting face-to-face turned out to be an excellent way to get used to each other. The collaboration consists of East-Asia (KASI in South-Korea, NAOJ in Japan) and Europe (University of Bordeaux, ESO in Germany, ASTRON in the Netherlands).