- x NeonPulse | Future Blueprint
- Posts
- 🤖 NVIDIA's announces AI powerchips
Good morning and welcome to the latest edition of neonpulse!
Today, we’re talking about the potential role of AI in sick and elderly care, because a new prototype shows potential…
NVIDIA’s New Powerchip
In a move to stay at the forefront of AI, Nvidia has introduced a new chip, the GH200, designed explicitly for running artificial intelligence models. This strategic step comes as a response to mounting competition from industry heavyweights like AMD, Google, and Amazon, all vying for dominance in the expanding AI hardware landscape.
Currently, Nvidia stands as the undisputed leader in AI chip manufacturing, commanding an impressive 80% share of the market, according to expert estimations. The company's specialization lies in graphics processing units, or GPUs, which have emerged as the favored choice for powering large AI models.
However, the demand for Nvidia's chips has outstripped supply due to the fervent race among tech giants, startups, and cloud providers to secure GPU resources for building their own AI models.
The heart of Nvidia's new GH200 chip is a powerful GPU, reminiscent of the company's top-tier AI chip, the H100. Yet, the GH200 doesn't stop there; it elevates the performance by coupling the GPU with a cutting-edge memory capacity of 141 gigabytes and a robust 72-core ARM central processor.
In a recent conference talk, Nvidia's CEO, Jensen Huang, shared his excitement about the GH200, stating, "We're giving this processor a boost. This processor is designed for the scale-out of the world's data centers." This indicates a pivotal shift toward optimizing data center operations with the enhanced capabilities of the GH200.
Anticipated to be available through Nvidia's distributors in the second quarter of the upcoming year, with sampling options projected by the year's end, the price point for the GH200 remains undisclosed at present.
Firstly, a model undergoes rigorous training using vast datasets, a process demanding substantial time and sometimes thousands of GPUs—such as Nvidia's H100 and A100 chips. Subsequently, the trained model is harnessed by software to make predictions or generate content through inference.
Comparable to training, inference is computationally intensive, requiring substantial processing power each time the software executes tasks—like generating text or images. However, in contrast to training, which is periodic, inference occurs continuously, as the model is deployed for real-world applications.
Jensen Huang aptly explained: "You can take pretty much any large language model you want and put it in this and it will inference like crazy. The inference cost of large language models will drop significantly."
Nvidia's announcement coincides with rival AMD's recent launch of their AI-centric chip, the MI300X, boasting a remarkable memory support of 192GB and a strong focus on AI inference capabilities. Meanwhile, behemoths such as Google and Amazon are diligently crafting their own custom AI chips, geared towards optimizing AI inference tasks.
Do you think NVIDIA will stay at the top? |
Cool AI Tools
🔗 VocalReplica: Clone any voice with just one YouTube video.
🔗 STORI AI: Create branded content with AI.
And now your moment of zen
Source: Humanoid flamingo.
That’s all for today folks!
If you’re enjoying neonpulse, we would really appreciate it if you would consider sharing our newsletter with a friend by sending them this link:
https://neonpulse.beehiiv.com/subscribe?ref=PLACEHOLDER