I just watched the CES 2025 Keynote by NVIDIA CEO Jepsen Huang.
I’m feeling the hype.
I’ve never heard him speak before. He’s a nerd. I love it, even if it is a bit corny. The keynote was very well staged.
I kinda want to get the 5080FE if I can swing it when they release on the 30th. Wish me luck!
The 4080 Super is already more than twice as powerful as the 2070 Super that I have. (See here.) And if the numbers bear out, the 5080 should be
Along with the NAS, this is the year to replace my GPU to avoid having a full computer to buy and build in 3-4 years. What’s the point of an machine with upgradable components if you don’t do it one piece at a time?
I have already checked and my exisitng PSU should be enough to power it. I can also get an official cable for the dramatic power connector it requires, instead of a tricky adapter set up.
The 2070 Super I have is unfortunately 7mm too long to simply migrated into the NAS case though. Either I’ll need to do some surgery to it, or the case itself to make it fit. The overkill PSU I’m putting into it could certainly power it at least.
Related, I recently read The NVIDIA Way by Tae Kim, while sitting by a pool in San Jose del Cabo, and the flights there and back. It was an interesting read! I liked learning the story of the company, along with the major milestones in the hardware development process. I was taken back to my point of view from growing up as a gamer and CS student.
It did briefly make me want to try out CUDA programming, but the urge passed.
Overall it seemed too finicky to do more than a passing hack or two. If it does come up, I’ll be sure to write about it here. There are a number of existing Go packages that bind to the CUDA libraries.
I liked the blend of history, interviews and technical details the book had.
It did not make me want to work for Jepsen though. I don’t think I’ve ever been that ambitious. But then, if I were, I’d be a CEO wouldn’t I?
The Keynote covered the new GPUs, a few demo, and a multitude of collaborations they have with clients for their various AI powers.
I’m still a bit negative on the current GenAI/LLM trends, in particular for all the ways it generates spam, or devalues real artists and creators. But the use in simulations seems useful enough. We need more automation, and this is a valid way to enable that.
I’d probably feel differently if I started to try and find ethical uses for it, and the power of the hardware I have. As much as one can do it within modern society at least.
But that’s a topic for another time.