Skip to main content

Huawei Shows Off 384-Chip AI Computing System That Rivals Nvidia's Top Product

3 weeks 6 days ago
Long-time Slashdot reader hackingbear writes: China's Huawei Technologies showed off an AI computing system on Saturday that can rival Nvidia's most advanced offering, even though the company faces U.S. export restrictions. The CloudMatrix 384 system made its first public debut at the World Artificial Intelligence Conference (WAIC), a three-day event in Shanghai where companies showcase their latest AI innovations, drawing a large crowd to the company's booth. The CloudMatrix 384 incorporates 384 of Huawei's latest 910C chips, optically connected through an all-to-all topology, and outperforms Nvidia's GB200 NVL72 on some metrics, which uses 72 B200 chips, according to SemiAnalysis. A full CloudMatrix system can now deliver 300 PFLOPs of dense BF16 compute, almost double that of the GB200 NVL72. With more than 3.6x aggregate memory capacity and 2.1x more memory bandwidth, Huawei and China "now have AI system capabilities that can beat Nvidia's," according to a report by SemiAnalysis. The trade-off is that it takes 4.1x the power of a GB200 NVL72, with 2.5x worse power per FLOP, 1.9x worse power per TB/s memory bandwidth, and 1.2x worse power per TB HBM memory capacity, but SemiAnalysis noted that China has no power constraints only chip constraints. Nvidia had announced DGX H100 NVL256 "Ranger" Platform [with 256 GPUs], SemiAnalysis writes, but "decided to not bring it to production due to it being prohibitively expensive, power hungry, and unreliable due to all the optical transceivers required and the two tiers of network. The CloudMatrix Pod requires an incredible 6,912 400G LPO transceivers for networking, the vast majority of which are for the scaleup network." Also at this event, Chinese e-commerce giant Alibaba released a new flagship open-source reasoning model Qwen3-235B-A22B-Thinking-2507 which has "already topped key industry benchmarks, outperforming powerful proprietary systems from rivals like Google and OpenAI," according to industry reports. On the AIME25 benchmark, a test designed to evaluate sophisticated, multi-step problem-solving skills, Qwen3-Thinking-2507 achieved a remarkable score of 92.3. This places it ahead of some of the most powerful proprietary models, notably surpassing Google's Gemini-2.5 Pro, while Qwen3-Thinking secured a top score of 74.1 at LiveCodeBench, comfortably ahead of both Gemini-2.5 Pro and OpenAI's o4-mini, demonstrating its practical utility for developers and engineering teams.

Read more of this story at Slashdot.

EditorDavid

Huawei Shows Off 384-Chip AI Computing System That Rival Nvidia's Top Product

3 weeks 6 days ago
Long-time Slashdot reader hackingbear writes: China's Huawei Technologies showed off an AI computing system on Saturday that can rival Nvidia's most advanced offering, even though the company faces U.S. export restrictions. The CloudMatrix 384 system made its first public debut at the World Artificial Intelligence Conference (WAIC), a three-day event in Shanghai where companies showcase their latest AI innovations, drawing a large crowd to the company's booth. The CloudMatrix 384 incorporates 384 of Huawei's latest 910C chips, optically connected through an all-to-all topology, and outperforms Nvidia's GB200 NVL72 on some metrics, which uses 72 B200 chips, according to SemiAnalysis. A full CloudMatrix system can now deliver 300 PFLOPs of dense BF16 compute, almost double that of the GB200 NVL72. With more than 3.6x aggregate memory capacity and 2.1x more memory bandwidth, Huawei and China "now have AI system capabilities that can beat Nvidia's," according to a report by SemiAnalysis. The trade-off is that it takes 4.1x the power of a GB200 NVL72, with 2.5x worse power per FLOP, 1.9x worse power per TB/s memory bandwidth, and 1.2x worse power per TB HBM memory capacity, but SemiAnalysis noted that China has no power constraints only chip constraints. Nvidia had announced DGX H100 NVL256 "Ranger" Platform [with 256 GPUs], SemiAnalysis writes, but "decided to not bring it to production due to it being prohibitively expensive, power hungry, and unreliable due to all the optical transceivers required and the two tiers of network. The CloudMatrix Pod requires an incredible 6,912 400G LPO transceivers for networking, the vast majority of which are for the scaleup network." Also at this event, Chinese e-commerce giant Alibaba released a new flagship open-source reasoning model Qwen3-235B-A22B-Thinking-2507 which has "already topped key industry benchmarks, outperforming powerful proprietary systems from rivals like Google and OpenAI," according to industry reports. On the AIME25 benchmark, a test designed to evaluate sophisticated, multi-step problem-solving skills, Qwen3-Thinking-2507 achieved a remarkable score of 92.3. This places it ahead of some of the most powerful proprietary models, notably surpassing Google's Gemini-2.5 Pro, while Qwen3-Thinking secured a top score of 74.1 at LiveCodeBench, comfortably ahead of both Gemini-2.5 Pro and OpenAI's o4-mini, demonstrating its practical utility for developers and engineering teams.

Read more of this story at Slashdot.

EditorDavid

Researchers Quietly Planned a Test to Dim Sunlight Over 3,900 Square Miles

3 weeks 6 days ago
California researchers planned a multimillion-dollar test of salt water-spraying equipment that could one day be used to dim the sun's rays — over a 3,900-square mile are off the west coasts of North America, Chile or south-central Africa. E&E News calls it part of a "secretive" initiative backed by "wealthy philanthropists with ties to Wall Street and Silicon Valley" — and a piece of the "vast scope of research aimed at finding ways to counter the Earth's warming, work that has often occurred outside public view." "At such scales, meaningful changes in clouds will be readily detectable from space," said a 2023 research plan from the [University of Washington's] Marine Cloud Brightening Program. The massive experiment would have been contingent upon the successful completion of the thwarted pilot test on the carrier deck in Alameda, according to the plan.... Before the setback in Alameda, the team had received some federal funding and hoped to gain access to government ships and planes, the documents show. The university and its partners — a solar geoengineering research advocacy group called SilverLining and the scientific nonprofit SRI International — didn't respond to detailed questions about the status of the larger cloud experiment. But SilverLining's executive director, Kelly Wanser, said in an email that the Marine Cloud Brightening Program aimed to "fill gaps in the information" needed to determine if the technologies are safe and effective.âIn the initial experiment, the researchers appeared to have disregarded past lessons about building community support for studies related to altering the climate, and instead kept their plans from the public and lawmakers until the testing was underway, some solar geoengineering experts told E&E News. The experts also expressed surprise at the size of the planned second experiment.... The program does not "recommend, support or develop plans for the use of marine cloud brightening to alter weather or climate," Sarah Doherty, an atmospheric and climate science professor at the university who leads the program, said in a statement to E&E News. She emphasized that the program remains focused on researching the technology, not deploying it. There are no "plans for conducting large-scale studies that would alter weather or climate," she added. "More than 575 scientists have called for a ban on geoengineering development," according to the article, "because it 'cannot be governed globally in a fair, inclusive, and effective manner.'" But "Some scientists believe that the perils of climate change are too dire to not pursue the technology, which they say can be safely tested in well-designed experiments... " "If we really were serious about the idea that to do any controversial topic needs some kind of large-scale consensus before we can research the topic, I think that means we don't research topics," David Keith, a geophysical sciences professor at the University of Chicago, said at a think tank discussion last month... "The studies that the program is pursuing are scientifically sound and would be unlikely to alter weather patterns — even for the Puerto Rico-sized test, said Daniele Visioni, a professor of atmospheric sciences at Cornell University. Nearly 30 percent of the planet is already covered by clouds, he noted. Thanks to Slashdot reader fjo3 for sharing the news.

Read more of this story at Slashdot.

EditorDavid