Currently, the company that leads the artificial intelligence industry, in terms of hardware, is NVIDIA. The company has been preparing for this moment for years and has the most cutting-edge technologies that manage to satisfy the needs of AI. Jensen Huang, CEO of NVIDIAin a recent interview that demand of solutions blackwell architecture It’s “crazy.”
This new architecture focuses on artificial intelligence, offering a huge leap in performance compared to the previous generation. Due to the increased capabilities, companies with AI models developed or under development want these solutions.
Note that this is something known, that the demand is so high that NVIDIA cannot satisfy it. Well, specifically it is TSMC, the foundry that is responsible for producing Blackwell chips, who does not have the capacity to meet demand.
NVIDIA says Blackwell GPU demand is “crazy”
Blackwell is not only the architecture for GPUs intended for AI, it is also the architecture for GPU gaming. Come on, this design will be shared by two quite different segments.
A few hours ago, Jensen Huang, CEO of NVIDIA, was interviewed by CNBC. During the interview he indicated that the Blackwell GPU for AI applications They are in mass production phase. He indicates that everything is going according to plan and that the demand is “crazy.” It suggests that the demand is much higher than what could have initially been anticipated.
As Jensen commented, “everyone wants to have the most, everyone wants to be first.” He also highlighted that, in the future, we will see an annual architecture update.
He has not given more information about it, although a few months ago he already gave a preview of this. During Computex 2024, held at the beginning of summer, he already made a small sketch. Specifically, I highlight the launch of Blackwell Ultra architecture for 2025 and of the Rubin architecture for 2026. You should know that Rubin is a new new architecture and Blackwell Ultra would be an optimization of the current one.
In addition, Huang highlighted that this hour of route could “increase performance year over year by a factor of 2 to 3, like Hopper in Blackwell.” This means “a drop in costs or a drop in energy efficiency.” Huang wanted to place special emphasis on the efficiency factor, highlighting that “energy efficiency will increase from generation to generation.”
You should know that increasing efficiency does not mean reducing energy consumption. Indicates the computing capacity per watt of energy. Energy consumption could increase, but if computing power increases at the same time, efficiency increases significantly.
For example, let’s say an NVIDIA graphics card consumes 100 watts of power and generates 200 FPS. This gives us a ratio of 2 FPS per watt consumption. Now, let’s imagine that it starts to consume 120 watts and generate 360 FPS, thus giving a ratio of 3 FPS per watt. Although we have a 20% increase in consumption, we see that computing power has increased by 50%.
We have to highlight, lastly, that the RTX 50 Series will be based on the NVIDIA Blackwell architecture. A few months ago Huang already dropped that they would arrive on the market with stock limitations.