Nvidia recently announced a cut of its high-end A100 GPUs as a way of getting around US government-imposed restrictions on how it can sell AI training equipment in countries like China and Russia. -Down model has been launched. The card, often referred to as the A800, must have the same Mass computing capabilities as native However, a computing cluster can only talk with other playing cards at 400 gigabytes per second instead of 600Gbps. according to wall street journalWhich means the A800 should be just as good for simple tasks, but significantly less efficient when it’s part of something like a supercomputer.
An unnamed Nvidia spokesperson Told Reuters That Cardboard is “an alternative product to the Nvidia A100 GPU for customers in China” that “meets the US government’s explicit test for reduced export controls and cannot be programmed to exceed.” Although the exact specifications of the US exams are not public, WSJ Say they contain interconnect speeds other than “other performance thresholds”.
Nvidia’s notice about not being able to be programmed to exceed limits is required. Back when GPUs were in short supply, the company introduced that its new cards would have limitations that could not be hackedAdmittedly, playing cards are much less attractive for crypto mining. Familially, people were not able to get around these limitations.
The US has been imposing heavy restrictions on what kind of chips American companies can sell to China. In October, the Biden administration introduced new restrictions on what types of chips and chip manufacturing gear could very well be sold to the country, requiring producers to obtain export licenses. In September, Nvidia and AMD each warned traders that their companies could also be hit by new restrictions on the export of their top-tier wares.
The government has said the borders are an attempt to limit China’s technical and military capabilities. Some technologists, such as former Google CEO Eric Schmidt, have warned that the US has not been able to maintain, although various researchers have pushed back against the framing of the arms race, However, it is possible that the chips, like the A100, could be used for military or other nefarious purposes. As my colleague James Vincent identified, they could be used to create “automated targeting systems for drones or the facial recognition systems used by China to survey its population.”
Whether or not the A800’s slow interlink speeds will be enough to impede that work or significantly speed it up, Nvidia thinks it’s good enough to meet federal government requirements. For corporate, at least, that’s good news — it’s estimated that the restrictions could impact up to $400 million in quarterly gross sales “if customers don’t want to buy the company’s alternative product offerings or [US Government] Doesn’t give license on time.”