With NVIDIA extreme co-design, we’ve accelerated AI performance a millionfold over 10 years, and…
2.7M followers
107,866 views
29 Comments
Comments from the original post on X (Twitter)

@kiopmbd
Millionfold but at what power draw per query
3 likes

@NorvegiaMia
The positive energy you carry in your speech and actions is so beautiful 🌺
2 likes

@leyqs97
AI performance has come a long way. Do you think the next 10 years will bring an even bigger leap?😃
2 likes

@tsuyopon24
Wishing you continued evolution and expansion in your endeavors🔜success
1 like

@pianosheet2024
Nvidia likes and respects AMD! That’s why they share videos about AMD!😀
1 like

@SlimAssiliX
A millionfold gain only means something once you define the benchmark, since raw compute, performance per watt, and cost per useful AI task can each tell a completely different story
1 like

@laojiucai1234
wow!great!keep moving!
1 like

@duchenne
A millionfold in 10 years is about 4x per year, every year. Moore's law gave 2x every two years. The cost of intelligence now falls faster than any input in history. By 2035 the scarce input is power, not chips.

@Nusodinkelmi
A millionfold faster in 10 years, and my build still takes exactly one coffee.

@mohammadvazin
A millionfold in ten years, and most of it came from the stack, not the transistor: lower precision, sparsity, interconnect, kernels. That's also why it's hard to compete on chips alone. The open question is how much of that gain reaches the price per token builders pay.

@mayankgulati99
A millionfold in ten years and I'm still hitting rate limits 😭

@mhmdatx
The part that does not get said enough: co-design now reaches the building. A millionfold in performance came with racks going from 10 kW to 250 kW and 800 VDC busbars in the open spec. The chip roadmap is public and on schedule. The facility roadmap is where most of the industry is behind.

@thesimaximalist
NVIDIA's hardware-software co-design is key to that millionfold gain!

@sachintwtss
a millionfold gain in ai performance over 10 years is hard to wrap your head around. co-design across the whole stack is clearly doing the heavy lifting.

@szdyai666
It gets harder and harder to improve as you go on

@MacroBombastic
yeah man that curve is wild, I remember when GPUs were just for gaming lol

@aapakari
Does that millionfold count the software stack, or just the silicon?

@gimgiyeon13
👏

@AndyYang0310
Cloud Bigdata Neural Network Algorithm Deep Learning Divergence Convergence Nvidia 3D AI in collabo w/ Samsung HBM

@AiTraceRoot
Make AI even greater.









