The performance of the NVIDIA special version chip has shrunk by 80%, and the price is **!
This article is original, please do not plagiarize and repost, offenders will be investigated!
The United States continues to change chip rules and restrict the export of Nvidia's high-end artificial intelligence chips to Chinese mainland. I think Nvidia is very depressed, it could have won billions or tens of billions of yuan in orders in Chinese mainland and made a lot of money.
But U.S. chip regulations forced Nvidia to change its strategy, continuously reducing chip performance and launching special edition chips. And the NVIDIA Special Edition chip, which is widely concerned in the industry, has finally changed its mind, and the performance has been reduced by 80%*.
NVIDIA Special Edition chip.
If you want to say that the hottest and most concerned technology field at the moment, it is artificial intelligence. Major tech companies are trying to get involved, training their own large models and creating top-of-the-line AI products. And NVIDIA (NVIDIA) is standing on the cusp of the industry, conquering 90% of the global artificial intelligence chip market with an absolute leading advantage.
Almost all of the main high-end AI chips on the market are Nvidia products, and although the products are good, they can't be sold if they want to. According to U.S. regulations, Nvidia's high-end AI chips such as A100, H100, A800, and H800 are all banned.
Things are getting bigger and bigger, is it time to give up the Chinese market? Apparently not, because in order to meet export standards, they again reduced the performance of the chip, and on the basis of the H100 design, the H20 series came into being.
According to Reuters, in early February, Nvidia began accepting pre-orders from dealers for new AI chips designed specifically for China,** on par with competitors from companies such as Huawei.
This time, Nvidia H20 benchmarked against Huawei's Ascend 910B, in terms of performance, H20 shrank by 80% compared with H100, and compared with Huawei's product FP32, the performance is not as good as Huawei in some key areas. It remains to be seen how competitive such a product may be in the Chinese market, with the focus on H20's ***12,000 yuan.
And through channels such as dealers, ** will rise to 1At around 70,000 yuan, which is equivalent to 120,000 rupees, it belongs to the same price range as the Huawei RISE 910B.
What are the opportunities for Huawei?
Counting the H20 series, Nvidia is the second time to develop a special version of China's chips, and the United States has always been eyeing China's high-tech industry and setting export control rules for high-quality advanced chip products. The only chance for Nvidia to go on the market is to reduce the computing power of the chips and reduce the level of performance.
The more this is the case, the more alternative options there are for customers, after all, Nvidia can't come up with high-end products with core competitiveness, among which there are too many similar reference products that have been squeezed by performance, and customers don't necessarily have to do Nvidia's business.
Maybe it's Huawei's opportunity, and Jiang Taocai, vice president of KDDI, said that Huawei's Ascend 910B can basically be compared to the Nvidia A100.
According to public information, the AI computing power of the Ascend 910B is 256 Tflops, and the Atlas 300T product using this chip has 30 built-in AI cores. Looking at the Nvidia A100 again, the FP32 floating-point performance is 195 TFLOPS with a maximum power consumption of 400 W. The NVIDIA A100 is currently the most powerful AI chip in the world.
If Nvidia's cutting-edge AI chips can't be sold publicly, Chinese manufacturers like Huawei will obviously have more opportunities to ship AI chips. Even if Nvidia launches another H20 special edition chip, it won't help. What's more, the United States has repeatedly changed the rules, and who knows if Nvidia will be restricted again after the launch of H20 this time?
Instead of pinning your hopes on others, it is better to master the basic techniques yourself, which is the best way to solve the problem.
If you agree, please give a thumbs up and feel free to **, comment and share.