Sunday, May 28th 2023
Computex 2023: NVIDIA Keynote Address by CEO Jensen Huang
The 2023 Computex will be the first fully-fledged in-person tech expo in Taiwan in over 3 years, and there's no better event to kick it off than a Keynote address by a man who loves his job—Jensen Huang, co-founder and CEO of NVIDIA, a company that has broken serious ground in the areas of computer graphics, supercomputing, and AI. We are in Taipei, and we will be live-blogging the event.
"I haven't given a public speech in 4 years! Wish me luck! - Jensen"
03:05 UTC: Jensen highlights how RTX real-time ray tracing transformed accelerated ray tracing.
03:08 UTC: RTX 4060 Ti is "re-unveiled"
03:09 UTC: NVIDIA ACE (Avatar Cloud Engine) for animating digital characters using AI.
04:55 UTC:
03:10 UTC: The future of AI and MMORPGs could be incredible with digital character animations.In the near future, GPTs will drive NPCs, so you can have lengthy conversations. The GPT will follow the character's backstory.
03:13 UTC: IBM System 360 formed the basis for software investments for 60+ years. The future is AI-written software.
03:17 UTC: The computer is the data-center, not the individual server. GPU-accelerated data-centers have lower TCO.
03:24 UTC: Data-center TCO explained:
03:27 UTC: Full volume production of the H100 "Hopper" has commenced. H100 has a transformer engine within it.
03:35 UTC: Magnum IO is a data-processing/training/optimization OS.
03:38 UTC: In the future, companies will have AI development departments along the lines of HR.
03:42 UTC: Generative AI is a force of nature.
03:44 UTC: Grace Hopper GH2000 is now in full production. Nearly 200 billion transistors, with 600 GB of memory coherent between the CPU and GPU.
03:56 UTC: GH2000 will accelerate recommender engines unlike anything else.
03:59 UTC: Cadence EDA, CAE, CAD, is accelerated by Grace Hopper.
04:04 UTC: Grace HopperDGX GH200 GPU Supercomputer can play Crisis, when it's not pushing the boundaries of AI.
04:05 UTC: NVIDIA Maxine 3D video conferencing taps into AI to generate 3D faces using 2-D camera inputs.
04:10 UTC: A 5G network can run on a Grace Hopper at no "cost."
04:22 UTC: Spectrum 4 is an AI-accelerated network chip. 100 billion transistors, 128x network ports.
04:38 UTC: "You've been selling me things every time I came to Taiwan. Omniverse is the first thing I will sell you.": Jensen
04:55 UTC: This concludes the keynote.
03:05 UTC: Jensen highlights how RTX real-time ray tracing transformed accelerated ray tracing.
03:38 UTC: In the future, companies will have AI development departments along the lines of HR.
03:42 UTC: Generative AI is a force of nature.
04:22 UTC: Spectrum 4 is an AI-accelerated network chip. 100 billion transistors, 128x network ports.
04:55 UTC: This concludes the keynote.
17 Comments on Computex 2023: NVIDIA Keynote Address by CEO Jensen Huang
EDIT:: oh ok, I didnt see this thread getting updated and it was literally just him saying he hasnt done a speech.
They should be unlaunching this 4050 and have 4060 4608cc 12GB and 4050 4352cc 8GB instead. 4060 3072cc is actually a 4030.
Since you can't predict what an AI would give you.
It is so easy for an AI to just copy paste someone else's copyrighted material and presented to you as the 【Result】.
There will be more and more copyrighted materials carrying similar characteristics exists online, generated by AI using the most popular prompts.
It will be like a mine field to try to avoid getting copyright strike using AI generated materials in commercial use.
Then you need another AI to check if the first AI copied too much copyright materials.
Then you need a third AI to check the result of the second AI...It is just a death spiral.
On the other end, AI 's liability is very questionable.
Who is gonna take responsibility when an AI makes mistakes?
The Company who decided to use it? The operator? The maker of the AI software? Or blame the fault dataset it trained with?
Having AI being a black box within your business process could introduce serious uncertainties.
Simply put, Nvidia are losing interest.
Why make relatively low margin consumer parts when they can make far higher profits in the balooning AI bubble?You've barely scratched the surface of the legal minefield these new techs have laid.
AI is going to get a lot of lawyers rich.
Very, very rich.
AI divisions.
Hard as it may be to hear but NV have cut GPU wafer requirements due to falling sales while
They can sell their AI chips twice over.
Just talk about AI, next day stock price goes up.
Nvidia is clearly milking this way beyond what's reasonable, but it's working for them so you can't really blame them.