Nvidia, a leading chipmaker, has recently announced the launch of its highly anticipated Blackwell series, a new generation of artificial intelligence (AI) chips and software designed specifically for running AI models. This groundbreaking development was unveiled at Nvidia’s developer’s conference in San Jose, where the company aimed to solidify its position as the go-to supplier for AI companies.
The first chip in the Blackwell series, known as the GB200, is set to be released later this year. According to Nvidia, the Blackwell-based processors offer a significant leap in performance, with 20 petaflops in AI performance compared to the previous H100 series, which offered only 4 petaflops. This means that AI companies will now have access to more powerful processing capabilities, enabling them to train larger and more intricate AI models.
One of the key features of the Blackwell chip is its “transformer engine,” which is specifically designed to run transformers-based AI. This technology is crucial for running AI models like ChatGPT, which rely on transformers as their core architecture.
The Blackwell GPU is a remarkable piece of engineering, combining two separate dies into a single chip manufactured by TSMC. Additionally, Nvidia will offer the GB200 as a complete server called the GB200 NVLink 2, which consists of 72 Blackwell GPUs and other Nvidia components designed to facilitate AI model training.
In terms of accessibility, major cloud service providers such as Amazon, Google, Microsoft, and Oracle will offer access to the GB200 through their cloud services. Amazon Web Services, for example, plans to build a server cluster with an impressive 20,000 GB200 chips. This level of infrastructure will enable companies to deploy models with up to 27 trillion parameters, surpassing even the largest models currently available in the market.
With regards to cost, specific pricing details for the GB200 and its associated systems have not been disclosed by Nvidia. Analyst estimates, however, suggest that the previous generation H100 chip, which the Blackwell series aims to replace, costs between $25,000 and $40,000 per chip, with whole systems costing as much as $200,000.
In addition to the hardware advancements, Nvidia also revealed a new software product called NIM. This software aims to make it easier for companies to utilize older Nvidia GPUs for inference, the process of running AI software. NIM enables companies to continue leveraging the GPUs they already own, making it a cost-effective solution for businesses that want to run their own AI models instead of relying on AI-as-a-service providers.
Nvidia’s ambition behind the introduction of NIM is to encourage customers who purchase Nvidia-based servers to also subscribe to Nvidia enterprise software. With an annual license cost of $4,500 per GPU, the Nvidia enterprise software provides businesses with the necessary tools to efficiently run their AI models on their own servers or on Nvidia’s cloud-based servers. This integration seamlessly integrates with popular AI platforms like Microsoft and Hugging Face.
In summary, Nvidia’s Blackwell series represents a major step forward in AI processing technology. With its powerful performance capabilities and innovative features, the Blackwell chips and accompanying software provide AI companies with the necessary tools to train and deploy larger and more sophisticated AI models. As Nvidia continues to evolve into a comprehensive platform provider, it is poised to remain at the forefront of the AI revolution.
—
FAQ:
Q: What is the Blackwell series?
A: The Blackwell series is Nvidia’s new generation of AI chips and software designed for running AI models.
Q: When will the first Blackwell chip be released?
A: The first Blackwell chip, the GB200, is expected to ship later this year.
Q: How does the Blackwell series differ from the previous H100 series?
A: The Blackwell series offers a significant performance upgrade compared to the H100 series, with 20 petaflops in AI performance versus 4 petaflops.
Q: What is the transformer engine feature of the Blackwell chip?
A: The transformer engine is a specialized component of the Blackwell chip designed to run transformers-based AI, a key technology in models like ChatGPT.
Q: Will the Blackwell chips be accessible through cloud services?
A: Yes, major cloud service providers such as Amazon, Google, Microsoft, and Oracle will offer access to the Blackwell chips through their cloud services.
Q: How much will the Blackwell chips cost?
A: Specific pricing details have not been disclosed, but analyst estimates suggest that the previous generation H100 chip costs between $25,000 and $40,000 per chip.
Q: What is NIM?
A: NIM is a new software product introduced by Nvidia that aims to facilitate the use of older Nvidia GPUs for inference, the process of running AI software.
Q: How much does the Nvidia enterprise software subscription cost?
A: The annual license cost for the Nvidia enterprise software is $4,500 per GPU.
Nvidia’s Blackwell series is a major development in the AI chip industry, aiming to provide AI companies with more powerful processing capabilities for training larger and more complex AI models. Market forecasts suggest that the demand for AI chips and software will continue to grow as AI becomes increasingly prevalent across industries.
The AI chip industry is highly competitive, with major players like Nvidia, Intel, and AMD constantly pushing the boundaries of performance and efficiency. Nvidia’s Blackwell series is expected to solidify the company’s position as a leading supplier for AI companies, particularly in the field of transformers-based AI models.
According to market research, the global AI chip market is projected to reach a value of $83.3 billion by 2026, with a compound annual growth rate (CAGR) of 35.2% during the forecast period. This growth is driven by the increasing adoption of AI technologies in various sectors, including healthcare, automotive, finance, and retail.
One of the main issues related to the AI chip industry is the high cost of hardware. While specific pricing details for the Blackwell chips have not been disclosed, estimates suggest that the previous generation H100 chip costs between $25,000 and $40,000 per chip. The cost of whole systems can reach as high as $200,000. These prices pose a significant barrier to entry for smaller companies and startups looking to incorporate AI into their operations.
In an effort to address this issue and provide more affordable solutions, Nvidia has introduced NIM, a software product that enables the utilization of older Nvidia GPUs for inference. This allows businesses to leverage their existing GPU investments and reduce the need for expensive hardware upgrades. By offering a cost-effective solution, Nvidia aims to attract customers to subscribe to its enterprise software, which provides additional tools for running AI models efficiently.
As for market accessibility, major cloud service providers such as Amazon, Google, Microsoft, and Oracle will offer access to the Blackwell chips through their cloud services. This opens up opportunities for companies of all sizes to harness the power of AI without the need for significant infrastructure investments. For example, Amazon Web Services plans to build a server cluster with 20,000 GB200 chips, enabling companies to deploy models with unprecedented parameters.
In conclusion, Nvidia’s Blackwell series represents a significant advancement in AI processing technology. With the increasing demand for AI chips and software, the industry is poised for substantial growth. However, the high cost of hardware remains a challenge, which Nvidia aims to address with its NIM software. By providing more powerful and accessible solutions, Nvidia is well-positioned to remain at the forefront of the AI revolution.
For more information on the AI chip industry and market forecasts, you can visit Market Research Future.
The source of the article is from the blog japan-pc.jp