GTC 2024

[Deleted member 2197]

* Nvidia GTC (GPU Technology Conference) is a global artificial intelligence (AI) conference for developers that brings together developers, engineers, researchers, inventors, and IT professionals.[1] Topics focus on AI, computer graphics, data science, machine learning and autonomous machines. Each conference begins with a keynote from Nvidia CEO and Founder Jensen Huang, followed by a variety of sessions and talks with experts from around the world. GTC 2024 will have 1035 discussion sessions for participants.

It originated in 2009 in San Jose, California, with an initial focus on the potential for solving computing challenges through GPUs.[2] In recent years, the conference focus has shifted to various applications of artificial intelligence and deep learning, including: self-driving cars, healthcare, high performance computing, professional visualization, and Nvidia Deep Learning Institute (DLI) training.[3] * Wikipedia

March 18 - 21

Catalog sessions
Date & time, participants and live stream links:

NVIDIA #GTC2024 Conference Session Catalog

Register now. Streamed online. March 18-21, 2024.

www.nvidia.com

 

[LordEC911]

Ian had a youtube video about it and mentioned a raffle/giveaway of a few (5?) RTX4090.
Think he said after you register, you have to watch the keynote and one other session to qualify.
Free online registration for the "virtual" pass and live-streaming of the sessions.
There was maybe 10 or so sessions/panels that seemed quite interesting to me but a few were scheduled at the same time as the others.

Edit- Link to Ian's video for more accurate info than me trying to remember.

 

[Xmas]

I'll be going in person this year. Anyone else?

 

[Deleted member 2197]

I'll be going in person this year. Anyone else?

It is possible @Arun might be there since he plans on publishing something before GTC 2024 begins.

 

[Arun]

Yep, I’m in San Jose now, happy to meet up with anyone here at some point

Not planning to publish anything beforehand anymore, scope creep got ahead of me, could have posted a braindump but not sure that'd have been very useful (does anyone really want to hear me rant about why CUTLASS only makes use of 116 rather than 120 SMs with thread block clusters? Seems smarter to try to ask them directly in person tbh...)

 

[nAo]

Damn, I wish I was down in CA! Enjoy GTC

 

[Deleted member 2197]

 

[LordEC911]

Super creepy intro...

 

[trinibwoy]

He was visibly annoyed at the lack of applause

 

[LordEC911]

He was visibly annoyed at the lack of applause

Oh, I meant the intro video with the intentionally obvious computer generated voice talking about all the stuff "AI" is doing before Jensen walked out.

Jensen's opening wasn't great but normal Jensen things. LoL

 

[trinibwoy]

Oh, I meant the intro video with the computer generated voice talking about all the stuff "AI" is doing before Jensen walked out.

Jensen's opening wasn't great but normal Jensen things. LoL

Oh lol I missed the AI intro. His opening jokes were real cringey. But you’re right that’s his brand.

 

[LordEC911]

Did I mishear? Or did he misspeak?
Thought I heard Jensen say 208trillion transistors...

Edit- Really weird to see him stumble over the naming and interchanging then correcting chips/dies.

 

[Picao84]

That's a massive die!

 

[troyan]

No, two massive dies. 104 billions transistors per chip.

 

[Bondrewd]

waow, it's just SM spam.
makes sense given the pricing.

 

[trinibwoy]

No, two massive dies. 104 billions transistors per chip.

Think he said 10 TB/s between the dies but was pretty light on the details. Not sure if there’s any secret sauce there vs other chiplet implementations.

 

[troyan]

From their PR:

World’s Most Powerful Chip — Packed with 208 billion transistors, Blackwell-architecture GPUs are manufactured using a custom-built 4NP TSMC process with two-reticle limit GPU dies connected by 10 TB/second chip-to-chip link into a single, unified GPU.

NVIDIA Blackwell Platform Arrives to Power a New Era of Computing

New Blackwell GPU, NVLink and Resilience Technologies Enable Trillion-Parameter-Scale AI Models New Tensor Cores and TensorRT- LLM Compiler Reduce LLM Inference Operating Cost and Energy by up...

nvidianews.nvidia.com

 

[LordEC911]

Referencing Silicon Valley's hotdog detector?
Edit- Or I guess more likely, SV's hotdog detector is based on Nvidia's Alexnet.

 

[vola]

waow, it's just SM spam.

it's 160 SMs

 

[Bondrewd]

it's 160 SMs

Per die.
There are two!