news

musk: the most powerful ai training cluster in history is online, and 100,000 more gpus will be added

2024-09-04

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

tesla ceo elon musk. visual china data map
musk claims that the most powerful artificial intelligence computing cluster currently available has been launched, and its size will soon double.
on september 3, tesla ceo elon musk announced on the x platform that the super artificial intelligence training cluster created by its artificial intelligence startup x.ai has officially launched. the cluster is named "colossus".
he revealed that it took the team 122 days to complete the launch of colossus. colossus will also add 100,000 gpus (graphics processing units) in the coming months, of which 50,000 will be more advanced nvidia h200s, which means that colossus's computing power will double again.
on social platforms, the industry responded very enthusiastically, calling it a great achievement in the field of artificial intelligence.
cathie wood, ceo of venture capital firm ark invest, also congratulated the team on their achievement, calling it “impressive” and revealing that xai will have major developments to announce in the future.
as early as the first half of last year, there were constant media reports that musk was purchasing a large number of gpus for tesla. in may of this year, colossus was first made public and began operating in july.
on may 25, musk revealed in a speech that xai plans to build a supercomputer called the "super factory of computing power", which is expected to be four times the size of the strongest competitor on the market. the computer will use nvidia h100 gpu.
on july 22, musk announced at x that the xai team, the x team, nvidia and other supporting companies had started training on the "memphis supercluster", also known as colossus, at 4:20 a.m. local time.
colossus consists of 100,000 liquid-cooled h100 gpus and runs on a single rdma structure (i.e., remote direct data storage structure that solves the latency of server-side data processing in network transmission). it is "the world's most powerful ai training cluster."
musk added that the goal is to train "the most powerful artificial intelligence in the world by every metric" by december this year.
of course, xai’s large model grok will also be trained on its own colossus.
last july, xai announced its official establishment and stated that the company's purpose is to "understand the true nature of the universe." xai stated on its official website, "we are a company independent of x corp, but will work closely with x, tesla and other companies to achieve our mission."
in november 2023, xai released its first large model, grok-1; on august 13, xai announced that the test version of grok-2 was officially launched, and introduced the image generation function for the first time; musk previously revealed that grok-3 will use colossus for training and is expected to be released at the end of the year. he believes it will be "very special."
in may this year, xai announced that it had received $6 billion in series b financing, with major investors including andreessen horowitz and sequoia capital. xai's pre-investment valuation was $18 billion, and its post-investment valuation after this round of financing will reach $24 billion.
however, while leading the artificial intelligence revolution, musk has long emphasized the potential security risks of artificial intelligence and actively promoted the implementation of artificial intelligence safety regulations.
on august 27, musk posted on x that he believed california should pass senate bill sb 1047. the move caused controversy in the tech community, with many worried that it could stifle ai innovation in the united states.
sb 1047 requires some of the largest ai models (costing more than $100 million) to report intelligent safety issues when they are developed and to submit reports assessing the risks associated with the models. the bill has been opposed by xai competitor openai.
the paper reporter wu yuli
(this article is from the paper. for more original information, please download the "the paper" app)
report/feedback