news

한어한어Русский языкРусский языкEnglishEnglishFrançaisFrançaisIndonesianSanskritIndonesian日本語SanskritDeutsch日本語PortuguêsDeutschΕλληνικάPortuguêsespañolΕλληνικάItalianoespañolSuomalainenItalianoLatinaSuomalainenLatina

in nvidia's revenue landscape, gpu is undoubtedly the largest source of contribution. recently, huang renxun said that the demand for his next-generation blackwell gpu chips is "crazy." however, the spectrum-x series of ethernet network products are also developing at a faster speed.

jen-hsun huang: demand for next-generation blackwell ai chips is “crazy”

ceo jen-hsun huang told cnbc's "closing bell overtime" that demand for the company's next-generation artificial intelligence chip, blackwell, is "insane."

"everybody wants to have the most, everybody wants to be no. 1," huang said in an interview that aired wednesday. nvidia shares were up about 3% on thursday morning.

blackwell, which is expected to sell for between $30,000 and $40,000 per unit, is hotly sought after by companies such as openai, microsoft, meta and others that are building ai data centers to support products like chatgpt and copilot.

nvidia has been a major beneficiary of the artificial intelligence boom, with its shares up about 150% so far this year. the company's second-quarter revenue continued to soar to $30.04 billion, a year-on-year increase of 122%. the company expects sales to reach $32.5 billion this quarter.

huang renxun said in an interview with cnbc: "in an era when technology is developing so rapidly, we have the opportunity to triple our investment and truly promote the innovation cycle, thereby increasing production capacity, increasing output, reducing costs, and reducing energy consumption. we are working hard in this direction. , everything is going according to plan.”

chief financial officer colette kress said in august that the company expected blackwell revenue to reach billions of dollars in the fourth quarter.

jensen said nvidia plans to update its ai platform every year to improve performance by two to three times.

spectrum-x: nvidia’s next money printing machine?

in addition to blackwell chips, nvidia's spectrum-x series of ethernet network products are rapidly contributing to its revenue.

kress made the above remarks in the second-quarter financial report released by nvidia at the end of august. he also said that the company's ai ethernet revenue, including spectrum-x products, doubled from the previous quarter.

she said: “spectrum-x has broad market support from oem and odm partners and is adopted by cloud service providers, gpu cloud providers and enterprises (including elon musk’s x.ai) to connect the world’s largest gpus computing cluster.”

kress claimed that spectrum-x "enhances the capabilities of ethernet for ai processing and improves performance to 1.6 times that of traditional ethernet" and said that nvidia plans to launch new spectrum-x products (as well as new chips) every year to "support the need to scale computing clusters from tens of thousands of gpus today to millions of gpus in the near future."

in 2019, nvidia acquired high-speed interconnect provider mellanox technologies for us$7 billion and began to aggressively enter the network field. since then, mellanox's technology has played an important role in nvidia's strategy to become a "data center scale computing company," as nvidia ceo jensen huang once put it.

while the company sells a variety of networking products, from ethernet and infiniband switches to data processing units and smartnics,but nvidia uses these products as basic building blocks for its ai servers and data centers, which it designs in-house and increasingly sells to customers ranging from enterprises to cloud service providers.

nvidia recently joined the super ethernet alliance along with other tech giants to develop specifications for high-speed networking solutions, and the solutions provider's chief engineer said that if nvidia eventually expands its network sales to cover the broader data center market, it will he wouldn't be surprised.

"i think they're probably going to rely on the fact that instead of just deploying the mellanox network and (the company's) basepod and superpod (ai supercomputer clusters), they're going to start deploying it as part of the enterprise network," he said. because they're starting to take this super ethernet alliance type of approach."

recently, nvidia hired jp vasseur, a famous cisco engineer and top inventor, to vigorously promote its network business. this is yet another sign that the ai ​​computing giant is making networking a top priority.

"with over 32 years of networking experience, including the past 13 years focused on artificial intelligence and networking products, i'm excited about the tremendous opportunities ahead to drive innovation in these areas," he wrote in a linkedin post.

more than a month ago, nvidia chief financial officer colette kress said the company's spectrum-x series of ethernet networking products for data centers "is expected to begin shipping a multibillion-dollar product line within a year." ".

vasseur received his phd and master's degrees in computer systems networking and telecommunications from telecom paris in france and stevens institute of technology in new jersey, usa respectively. he has been working at cisco for the past 25 years and is a well-known engineer and top inventor. , working on high-priority initiatives.

according to vasseur's linkedin profile, he previously served as vice president of network machine learning and artificial intelligence engineering at cisco, where he led "large-scale language model and generative ai use case specification, system architecture design, rapid engineering, model tuning and knowledge database and retrieval development of complex technologies such as augmentation generation. he previously served as an engineering lead in several high-profile areas including predictive networking, iot, and advanced threat detection.

he announced in a linkedin post on wednesday that he will join nvidia this month as nvidia's senior distinguished engineer and chief architect of artificial intelligence and networking.

cisco called vasseur a "top inventor" at the company in a 2019 blog post, with 483 patents to his name that "cover everything from machine learning and artificial intelligence in enterprise security to key aspects of the internet of things." technology field".

"he is one of only 19 cisco fellows out of 30,000 engineers who continue to push the boundaries of technology," cisco said in a blog post.

when vasseur announced his departure from cisco two weeks ago, he said, "one of the most exciting parts of my career has been contributing to the birth and explosive growth of a new industry: the internet, to which cisco has made significant contributions. "

now at nvidia, vasseur seems to be saying that he views his job at the artificial intelligence computing giant as another journey that will bring about huge changes.

"jensen huang's vision for nvidia to leverage ai factories to build new frontiers of accelerated computing and industry is incredibly inspiring, and i'm excited to join this journey and a talented team!" he wrote on linkedin .