Nvidia alleged to be crafting AI chip tailored for China, adhering to export restrictions; potential incorporation of NVLink for constructing robust clusters.
The Lowdown on Nvidia's New B30 AI Chip
In the wake of the export ban on its H20 chips to China, Nvidia is said to be cookin' up a new Blackwell-based alternative called the B30. According to The Information, this baby features multi-GPU scaling capabilities, and it looks like it's gonna be a game-changer!
There's been a lotta' buzz about different names for this bad boy, from the RTX Pro 6000D to the B40, but now it's officially the B30. Y'all better take notes!
While some speculate that the "multi-GPU scaling" feature refers to NVLink, Nvidia ain't been including that in their consumer-grade GPUs since the last generation. So, let's keep an open mind, shall we?
At Computex, Nvidia showed off their RTX Pro Blackwell servers that can pack up to eight RTX Pro 6000 GPUs, all wrapped up in ConnectX-8 SuperNICs with built-in PCIe 6.0 switches for hittin' on some high-speed GPU-to-GPU action. When you smush two DGX Spark superchips together, you get the same vibrant show. And it looks like the B30's gonna be a fascinating combination of that!
Rumor has it that Nvidia might've fiddled around with those GB202 chips found on an RTX 5090, slippin' in some NVLink support for the B30. But, that's just in the realm of speculation for now!
Jensen himself spilled the beans and said Nvidia ain't makin' no more H20 alternatives based on Hopper. Instead, they're movin' on to the Blackwell architecture for their China-bound chips. The U.S. government threw a wrench in the works with that H20 ban, targetin' the memory bandwidth and interconnect bandwidth for bein' too hot to handle. The reasoning? These AI accelerators can whip up supercomputers in China for military purposes.
Nvidia's been dodgin' bullets with the U.S. government on this export business, and it ain't been all sunshine and roses for competitors. AMD's been in a pickle recently with potential losses of up to $800 million following the ban on their MI308 chips. Ouch!
Jensen ain't sugarcoatin' it. He's callin' out the U.S. government for puttin' 'em in a tight spot, 'cause if Chinese rivals like Huawei start pumpin' out hardware that can go toe-to-toe with Nvidia, they just might set the stage for the next global standards.
Stay plugged in by followin' Tom's Hardware on Google News!
Stay in the Game: Sign Up for Tom's Hardware Newsletter
Want the latest news, analysis, and reviews? Look no further than the Tom's Hardware newsletter. Subscribe now!
Behind the Scenes:
Nvidia's B30 AI chip is reemerging to serve the Chinese market with features tailored to U.S. export controls in mind. Here are some insights that bring the B30 and its predecessors into sharper focus:
Taking a Closer Look at the Nvidia B30
- Scalability: The B30 chip offers multi-GPU scaling, which enables multiple chips to join forces for an impressive show of firepower as a single AI computing cluster[1][2]. It's unclear whether this scalability is powered by NVLink or ConnectX-8.
- Design Philosophy: Given GDDR7 memory now holds the reins, the B30 is expected to offer improved performance and efficiency over the H20, which used HBM[5].
- Partnerships: The B30's fabrication is likely entrusted to TSMC, paving the way for stronger partnerships with Taiwanese companies and Chinese cloud giants[5].
The B30: Apochrypha or Reality?
- RTX Pro 6000D: Originally a high-performance GPU aimed at catering to general computing and gaming needs, this beastie doesn't seem to have anything to do with multi-GPU scaling for AI applications, like the B30.
- B40: The buzz surrounding the B40 as Nvidia's next big thing in China has simmered down. All eyes are on the B30 to rise and shine!
In a nutshell, the B30 ain't no joke - it's a specialized AI chip for China's market, equipped with multi-GPU scaling capabilities that leaves its predecessors, the RTX Pro 6000D and B40, in the dust. Keep your eyes peeled for more details!
The B30 AI chip is designed for the Chinese market, offering multi-GPU scaling capabilities that enable multiple chips to collaborate as a single AI computing cluster, potentially revolutionizing data-and-cloud-computing in the industry. With its focus on scalability and efficiency, the B30 could reshape the finance and business landscapes by providing an edge in AI-related tasks. This specialized chip, different from the RTX Pro 6000D initially intended for general computing and gaming, represents a significant leap forward in technology, especially within the realm of artificial intelligence.
