GPU

Dec. 11 (Bloomberg) -- IBM has announced the development of a new optical technology that can train AI models at the speed of light while saving significant energy. By applying this breakthrough to data centers, the company says that training one AI model saves as much energy as 5,000 U.S. homes consume in a year. The company explained that while data centers are connected to the outside world through fiber optic cables, copper wires are still used internally. These copper wires are connected to GPU gas pedals, which spend a lot of time idle while waiting for data from other devices...

October 27 news, Denmark launched the country's first AI supercomputer, named after the Danish mythological goddess Gefion, aimed at promoting breakthroughs in quantum computing, clean energy, biotechnology and other fields, NVIDIA CEO Jen-Hsun Huang and the King of Denmark attended the unveiling ceremony. Gefion is a NVIDIA DGX SuperPOD supercomputer powered by 1528 NVIDIA H100 Tensor Core GPUs using NVIDIA's Quantum-2 InfiniBand net...

At a meeting with analysts last week, billionaire Oracle co-founder and CTO Larry Ellison told the audience that he and Elon Musk, the world's richest man, took NVIDIA CEO Jensen Huang to Nobu Palo Alto for dinner and "begged" Huang to give them more GPUs. "I would describe the dinner as Oracle - Elon and I begging Jensen for GPUs," Ellison recalls. "Please take our money. Please take our money. By the way, I'm having dinner. No, no, no, eat more. We need you to have more...

According to Jiemian News, at the 2024 REAL Technology Conference held today, Luan Qing, general manager of SenseTime Digital Entertainment Division, introduced that the domestic artificial intelligence computing cluster invested and built by SenseTime currently has 54,000 GPUs, with a maximum computing power of 20,000P. Luan Qing said that SenseTime is investing in the construction of the country's largest artificial intelligence data center in Lingang, Shanghai, and the country's computing nodes are spread across Shanghai, Guangzhou, Chongqing, Shenzhen, Fuzhou and other places. According to previous reports by IT Home, SenseTime's semi-annual report data as of June 30, 2024 showed that in the first half of 2024,…

A research report released by Meta shows that its 16,384 NVIDIA H100 graphics card cluster used to train the 405 billion parameter model Llama 3 experienced 419 unexpected failures in 54 days, an average of one every three hours. More than half of the failures were caused by the graphics card or its high-bandwidth memory (HBM3). Due to the huge scale of the system and the high synchronization of tasks, a single graphics card failure may cause the entire training task to be interrupted and need to be restarted. Despite this, the Meta team has maintained more than 90% of effective...

Recently, GPU cloud computing company Lambda announced the launch of its new 1-Click cluster service, where customers can now get Nvidia H100 GPU and Quantum2 InfiniBand clusters on demand. This innovative service enables enterprises to obtain computing power only when needed, especially for those companies that do not need to use GPUs 24 hours a day. Source Note: The image is generated by AI, and Robert, co-founder and vice president of Midjourney Lambda, a picture licensing service provider, said:

Musk announced on July 9 that his artificial intelligence company xAI is building a supercomputer with 100,000 Nvidia H100 GPUs, which is expected to be delivered and start training at the end of this month. This move marks the end of xAI's negotiations with Oracle to expand its existing agreement and lease more Nvidia chips. Musk emphasized that this will become "the most powerful training cluster in the world, and the lead is huge." He said that xAI's core competitiveness lies in speed, "which is the only way to close the gap." Prior to this, xA…

Hugging Face, the world's largest open source AI community, recently announced that it will provide $10 million in free shared GPUs to help developers create new AI technologies. Specifically, the purpose of Hugging Face's move is to help small developers, researchers, and startups fight against large AI companies and prevent AI progress from falling into "centralization." Hugging Face CEO Clem Delangue was interviewed by The Verge...

Intel made it clear at its first quarter earnings conference call at the end of last month that the Falcon Shores GPU will be launched in late 2025. According to foreign media HPCwire, the processor is being redesigned to meet the needs of the AI industry. Intel CEO Pat Gelsinger said that Falcon Shores will combine a fully programmable architecture with the excellent system performance of the Gaudi 3 accelerator, allowing users to achieve a smooth and seamless upgrade transition between two generations of hardware. Intel said that the AI industry is turning to Python…

On the 24th, the Beijing Municipal Bureau of Economy and Information Technology and the Beijing Municipal Communications Administration issued the "Beijing Computing Infrastructure Construction Implementation Plan (2024-2027)". The "Implementation Plan" proposes that by 2027, the quality and scale of computing power supply in Beijing, Tianjin, Hebei and Mongolia will be optimized, and efforts will be made to ensure that independent and controllable computing power meets the needs of large model training, and computing power energy consumption standards will reach the leading domestic level. Key tasks include promoting independent innovation in the computing power industry, building an efficient computing power supply system, promoting the integrated construction of computing power in Beijing, Tianjin, Hebei and Mongolia, improving the green and low-carbon level of the intelligent computing center, deepening computing power empowerment industry applications, and ensuring computing power foundations...

Once upon a time, Nvidia's H100 GPU for artificial intelligence computing was in short supply. However, according to Digitimes, TSMC's Taiwan general manager Terence Liao said that the delivery waiting time for Nvidia H100 has been greatly shortened in the past few months, from the initial 3-4 months to the current 2-3 months (8-12 weeks). Server foundry manufacturers also revealed that compared to the situation in 2023 when Nvidia H100 was almost impossible to buy, the current supply bottleneck is gradually easing. Although the delivery waiting time has…

The massive GPU clusters required for generative AI star Stability AI’s popular text-to-image generation model, Stable Diffusion, also appear to have been partly responsible for former CEO Emad Mostaque’s downfall — because he couldn’t find a way to pay for them. The UK model-building firm’s sky-high infrastructure costs allegedly depleted its cash reserves, leaving it with just $4 million as of last October, according to an exhaustive report citing company documents and dozens of people familiar with the matter. Stab…

In the AI industry, especially in the field of generative AI, the rapid development of technology and the broad prospects of application have attracted a lot of investment and attention. However, the high cost of this field has also caused widespread discussion in the industry. Recently, a report in the Wall Street Journal pointed out that companies in the AI industry spend 17 times their revenue on purchasing Nvidia GPUs, which is a shocking figure and has also triggered in-depth thinking about the future development of the industry. AI startup Cognition Labs, backed by well-known investor Peter Thiel, is seeking a valuation of $2 billion, and its valuation…

On March 28, according to a report by the Nikkei today, Nvidia's cutting-edge image processing semiconductor (GPU) H200 is now available. H200 is a semiconductor for the AI field, and its performance exceeds the current flagship H100. According to the performance evaluation results released by Nvidia, taking the processing speed of Meta's large language model Llama 2 as an example, the processing speed of generative AI derived answers of H200 is up to 45% higher than that of H100. Market research organization Omdia once said that in 2022…

NVIDIA has officially released AI Enterprise 5.0, an important product designed to help enterprises accelerate the development of generative artificial intelligence (AI). AI Enterprise 5.0 includes NVIDIA microservices and downloadable software containers that can be used to deploy generative AI applications and accelerate computing. It is worth mentioning that this product has been adopted by well-known customers such as Uber. As developers turn to microservices as an effective way to build modern enterprise applications, NVIDIA AI Enterprise 5.0 provides a wide range of...

According to people familiar with the matter, Nvidia is currently in deep negotiations with Israel's artificial intelligence infrastructure orchestration and management platform Run:AI to discuss acquisition matters. The value of the transaction is expected to reach hundreds of millions of dollars, and may even climb to a high of $1 billion. Currently, the negotiations between the two parties are still ongoing, and the specific acquisition details and conditions have not been disclosed. Run:ai is an AI optimization and orchestration platform specifically for GPUs. Run:ai provides a range of tools and features, including CLI and GUI, workspaces, open source frameworks, metrics, resource management...

As we all know, "AI PC" is one of the hottest topics in the consumer electronics industry. For some consumers who don't know much about the technical details but are interested in this concept, they believe that "AI PC" can help them complete some unskilled operations more intelligently or reduce the burden of daily work. But for users like us, who have high expectations for "AI PC" but are relatively familiar with it. Many times, we think that AI PC has already appeared, but why is it only being promoted now? How old is AI PC?…

Meta announced two new data center clusters through an official press release on the 12th. The company is hoping to stand out in AI-focused development through Nvidia's GPUs. It is reported that the sole purpose of these two data centers is to conduct AI research and development of large language models in consumer-specific application areas (IT Home Note: including sound or image recognition). Each cluster contains 24,576 Nvidia H100 AI GPUs, which will be used to train its own large language model Llama 3. The two newly built data centers…

Facebook parent company Meta is investing heavily in artificial intelligence technology as part of its "Technology Roadmap to 2026" to develop a new AI recommendation model for all its video products, including TikTok-like Reels short video services and traditional long videos in Facebook Feed. Meta executive Tom Alison revealed that the company currently usually uses a separate recommendation model for each product such as Reels, Groups and the main feed, but will develop a unified...

Tesla CEO Elon Musk made a brief appearance via video at the just-concluded Bosch Connected World Conference, talking about topics such as generative AI, super-intelligent AGI, self-driving cars, and robots. Musk expressed his concerns about artificial intelligence, emphasizing that without regulation and safeguards, the technology could get out of control and have negative consequences. He is currently suing OpenAI and its CEO Sam Altman because they abandoned the startup's original vision (to provide everyone with free...

Nvidia’s H100 GPU delivery cycle for artificial intelligence (AI) and high-performance computing (HPC) applications has reportedly been drastically reduced, from 8-11 months previously to just 3-4 months. This has led to some companies that have been hoarding trying to sell their excess H100 80GB processors, as it is now more convenient to rent chips from large companies such as Amazon Web Services, Google Cloud, and Microsoft Azure. According to The Information, some companies are trying to sell off their excess H100 80GB processors due to reduced scarcity and the high cost of maintaining unused inventory.

Source @Kepler_L2 revealed that AMD will launch a revised version of the AI accelerator MI300 using HBM3e to compete with Nvidia B100 at a low price, and AMD's next-generation Instinct MI400 accelerator will be released in 2025. AMD announced two Instinct MI300 accelerators in 2023, namely the pure GPU MI300X and the APU architecture MI300A, both of which use HBM3 memory with a capacity of 19…

Social media giant Meta plans to deploy a custom second-generation AI chip, code-named "Artemis," in its data centers this year. The new chip will be used for "inference," the process of running AI models, in Meta's data centers, Reuters reported. The goal of the move is to reduce reliance on Nvidia chips and control costs for AI workloads. Meta also offers generative AI applications in its service and is training a...

Nvidia is still strong in the field of artificial intelligence, and it is expected that the revenue of its data center business will reach a staggering $46 billion in 2024 (IT Home Note: Currently about 330.28 billion yuan), setting a new record. Speaking of Nvidia's performance in the field of artificial intelligence, there is no need to say more. In fiscal 2023, the company's outstanding financial performance in this field attracted much attention. According to the latest estimates of Wells Fargo Equity Research, Nvidia's data center business revenue and market share will continue to grow steadily in 2024. It is reported that Nvidia currently occupies more than 90% of the world's...