Innovation in computing has always been at the low end.
When people were able to get their own personal computers, that's when the revolution sparked.
Nvidia, in a weird reversal of the norms of computing, is pushing AI into "GPU Mainframes" like the Nvidia A100 - super giant, super powerful, super expensive AI/GPU number crunchers.
That's the playfield where Nvidia CEO Jensen Huang wants the AI revolution to play out - in the cloud, in the hands of the high priests of cloud computing at Amazon, Azure and Google, using Nvidia's most ridiculously expensive hardware.
But I have bad news for Jensen - software developers hate that. Consumers want to do it themselves, they want it to be cheap and they want to be in control. The industry abandoned the mainframe, and no matter how much Nvidia wants that era back, its finished.
What that means is that the AI revolution is not going to play out in the cloud, where GPUs are expensive and hard to get access to and controlled by the high priests, who decide who is allowed mainframe computing time.
Nope, the AI revolution will be on retail gaming spec GPUs from AMD and Intel. In fact Nvidia may not be part of this low end AI party, because Nvidia imposes artificial constraints on its drivers so you CAN'T do the AI you want to on their retail gaming GPUs.
Software developers will develop AI for what they can get their hands on. They'll smash and squeeze the software till it fits and works on what's available. They'll find innovative ways to connect multiple gaming GPUs together.
The forces of open source will rally against Nvidia's attempt to make AI into a new mainframe era.
It's happening now.
Andrew Stuart
9 Aug 2023
andrew.stuart@supercoders.com.au
