GPUs

I have recently become a bit obsessed with GPUs. I have been making paintings of imagined data centres for a while now without giving much thought to what is inside the real things. In my exhibition, data centres serve as a recurring motif which, according to the accompanying text that I wrote (but which now seems interestingly alien to me), serve as ‘metaphors for the instability of digital memory and the volatility of outsourced recollection’. I have now discovered that data centres are built from thousands of interconnected GPUs and this has chimed with me in unexpected ways.

This train of thought was started by reading a long piece by the tech blogger Ed Zitron (https://www.wheresyoured.at/the-case-against-generative-ai/). The article is hilariously bombastic but also very convincing. I was already pretty sure that there will be some sort of significant adjustment due to what seems like an AI bubble but the stuff in this piece is truly hair-raising. Zitron basically contends that the chip maker NVIDIA and Open AI are colluding in a demented GPU Ponzi scheme that cannot not end up in some sort of car crash (especially as Foundational AIs just cannot seem to do the things that these people are claiming/ promising). This is all grist to my ever-expanding mill but also, GPUs! I have played video games my whole life and so I have been aware of NVIDIA for a long time. I bought my first home computer (not counting my childhood C64) in something like 2001. I had it built and stretched my finances to get an NVIDIA graphics card. It was worth it and until I switched to console gaming, I always bought NVIDIA. As such they are one of those brands that I have always kind of liked. It’s weirdly disappointing to discover that they are largely responsible for the current age of AI bullshit.

I also think that there is something interesting in the fact that AI is powered by technology that was originally designed to render the unreal, imagined, and silly as convincingly as possible. It also explains why AI is so expensive (financially and environmentally) as GPUs are built to prioritise throughput over latency – they basically do one thing at a time, really quickly – which means that the sort of scaling of routine AI use that would be required to make it economically viable requires an almost infinite number of GPUs and the data centres to hold them.

Inevitably this has had me looking at pictures of GPUs and the racks that hold them. There is an interesting aesthetic to this sort of photography; there are the pictures designed to sell, that use dramatic lighting, interesting angles, and exaggerated scale and then there are the dull pictures of server rooms that seem designed to shout ‘nothing to see here’ with their repetitious anonymity. As such the GPUs seem to sit in a curious space at the moment, simultaneously infrastructure and commodity; hidden and hyped.

I have provisionally pitched that I will build a paper collage/ sculpture for the interim show. I think that the GPUs could provide some interesting source material for this. On a surface level they provide lots of interesting shapes and speak to some of the tangled structures that I have built in the past.

On another level I think that there could be something interesting in reimagining the GPUs in collage. Paper is relatively weak, disposable, cheap, and prone to failure, all qualities that NVIDIA would not want associated with their GPUs. For the interim show this feels like it could be a useful experiment. This would not be a resolved idea but rather an early articulation of my current train of thought. It would also allow me to open my project out again whilst keeping painting in mind as whilst the collage would be pleasingly ephemeral, I will document it thoroughly with a view to potential paintings.