Accepted paper:

The more GPUs you buy, the more money you save: images in data centres

Authors:

Adrian Mackenzie (Lancaster University)

Paper short abstract:

The paper will explore how large accumulations of images affect the architectural dynamics of data centres. It points to the importance of images as operational elements of data centres as technical ensembles.

Paper long abstract:

This paper presents an attempt to speak appropriately about the movement of images into data centre operations and platform infrastructures. The paper will explore the mixture of mundane combinations and unpredictable transformations that have appeared in some spectacular platform demonstrations such as Facebook's 'planetary settlement model' from the standpoint of the GPUs, or graphics processing units, that have become central to many predictive models working with large scale image collections in data centres. It traces the embedding of large image collections in a GPU-based model as a process that transforms collections of images into a highly condensed indexical field, a field that allows platforms to generate predictive propositions or statements about the world. In describing how predictive indexicality takes shape on platforms and machine learning models, the altered collectivity of images under data-centre platform conditions becomes more evident. This indexicality also suggests the need to map data centres as techno-affective architectures in a flow of images.

panel A22
Farming data - collaborations on site