Computex 2024 – NVIDIA Keynote Adventures and Rain

While not officially part of Computex the Gaming and AI powerhouse NVIDIA held their Computex Keynote two days before the official start of Computex 2024.  This was an invite only event that was held on the National Taiwan University campus which is also where Jensen Huang received an Honorary Doctorate several years back.

The Keynote was planned for the evening of June 2nd with a short reception in the nearby park complete with a DJ and several food trucks.  However, being that this is Taipei Taiwan during the Monsoon season there is a high chance that it will rain almost non-stop.

And it did

Funny thing is, I have attended several NVIDIA keynote presentations over the years and depending on the audience it can draw in quite the crowd.  As a result I decided to show up early and secure a place in line with the assumption that I would be inside, or at least under a tent.

No such luck, in fact the event was planned extremely well with scheduled times for when events would happen.  Check-in for the pre-keynote event started right on time and anyone showing up early would be forced to wait.  From that point there was another hour to wait until they opened the doors to the sports arena which also marked the two hour point before the Keynote event started.

An hour before the Keynote they opened the arena and would guide groups of people to their assigned seating location.  From there everyone waited for the event to start.

Needless to say, attending an NVIDIA Keynote speech is quite involved and come a long way from being held in a large popup tent in the middle of a parking lot

The keynote was similar to the presentation that was given during GDC that focused entirely on AI and how NVIDIA is taking charge by not only providing the hardware with the new Blackwell Datacenter Solutions but, also solutions on the software level with the NVIDIA Inference Microservice (NIM) which is used to create AI factories. 

Depending on what NIM’s you select will determine what type of factory you are building Or, the customer can build something custom to fit their unique needs.   

I find it interesting that everything is being hosted in Kubernetes.  Kubernetes is a command and control software used to create clusters using software containers.  The NVIDIA Inference Microservice (NIM) is basically a container that has been renamed.

In addition to enhanced descriptions of what makes up a Blackwell system there were two additional announcements.

First is the hardware release cadence.  This is an important metric as it will define how often the technology will advance while also providing support and EOL metrics for businesses and shareholders.  Back with the NVIDIA TNT video card, the cadence was every 6 months and while things have slowed on the GPU front, NVIDA still loosely follows this model. 

For AI hardware NVIDIA will be on a one year cycle meaning that while Blackwell is currently being delivered NVIDIA is also working on the next generation.

• 2024 and 2025 will be dominated by Blackwell and Blackwell Ultra
• 2026 and 2027 will see the introduction of Rubin and Rubin Ultra.

Everything here is neatly packaged from the one year gaps between product releases to the “Ultra” name being brought over from the GPU side and a strikingly similar “Tick / Tock” arraignment to it all. 

I wonder where I have seen that before

Overall, we just happy to see some really planning behind AI developments along with some solid plans for the future.