UNCOVERING THE LAYERS
David Craig , CEO , Iceotope
technologies could result in cumulative savings in capital expenditures and operating expenditures of US $ 100 billion to US $ 1 trillion by 2025 . To fully realise those savings , data processing is going to have to take place on-site . Oil rigs generate about a terabyte of data per day . However , that much data can take as many as 12 days to upload by satellite . There won ’ t be any cost savings to be had if one day ’ s worth of data takes 12 days to transmit . Furthermore , today ’ s CPU and GPU HPC ( high performance computing ) systems offer the processing capability to number crunch that data load on-site , even in hostile environments .
With these changing parameters – more data , AI applications and a move to the Edge – are data centre operators prepared ? Be it an enterprise operator , a colocation provider or a hyperscaler , the industry as a whole is beginning to evaluate the impact of these trends . Rack and chip densities are increasing and are requiring a new way of thinking .
Historically , enterprise racks were configured around relatively low-level heat loads of 3 – 6 kW range and some significant operators today run in the 10 – 15 kW per rack range . All of which is fine for most standard business applications . Now as AI applications are driving GPU processing , data centre operators are coming under greater pressure to deliver 30 – 60 kW racks , something most data centres are not designed for .
The challenge comes in the fact that nearly all components of the data centre environment are designed to be air cooled and not only in the highly mechanised often refrigerant-based cooling systems , which are consuming up to 40 % of a data centre ’ s energy load . They also require technology halls and telecom rooms to be organised to drive www . intelligentdatacentres . com
71