Driving Down The AI System Roadmap With Nvidia

In
the early years of the GPU acceleration of application performance – really
from “Kepler” datacenter GPUs in May 2012 to “Volta” in May 2017 – Nvidia, the
world’s most important technology company and still the overwhelmingly dominant
supplier of hardware and systems software for the GenAI revolution, was very
good about putting out roadmaps.

But
for a few years until 2021, the company kept its roadmaps folded up in the
front left inside pocket of co-founder and chief executive officer Jensen Huang’s
leather jacket, but as the GenAI boom went from chemical to nuclear, the
company correctly surmised that with everybody trying to synchronize money, land,
power, cooling, and systems to all come together in the largest infrastructure
buildout the IT market has ever seen, everybody needed a real roadmap, extending
out a few years, so they could plan.The
first such new era roadmap came
out at the end of 2023 in a financial presentation, not in a GTC conference
slide from Huang, and we edited the heck out of it to add missing components
like the some of the GPUs and DPUs and putting the correct calendar years on the
columns, but all along being grateful Nvidia was outlining where it was at and
where it was going. We gathered up all of the roadmaps we could find between
2021 and 2023 and put
them in this story so you would have them for reference.

That
October 2023 roadmap reveal was also when we got the first wind of the annual
cadence of…