The "improved" performance of the chip fell by 80%, and Nvidia blocked the domestic AI chip
The "improved" version of Nvidia has reduced the efficiency by eighty percent, is this going against domestic artificial intelligence?
Following the agreement between the United States, Japan and the Netherlands, Japan also quickly announced a plan for the production of 23 chips, reducing China's chip production capacity to below 45nm, and the United States had to drag the Netherlands in, but anyone can see that they are forced to helpless.
Unexpectedly, American companies "forced the palace", but it became a shield for Biden's camp, and the United States won a respite for American companies, "repaired" the economic and trade cooperation between China and the United States, canceled the permanent tax exemption policy for South Korean companies, and then, Chinese factories will no longer export high-end equipment.
The Netherlands also seized this opportunity to postpone ASML's order to January 1 next year and ship all DUV lithography machines, which temporarily resolved the contradiction between Chinese companies, but even so, the United States is secretly engaged in fierce competition.
The reappearance of Huawei's Kirin 9000 processor proves that China already has the strength to produce high-end chips, and Biden's ** had to abandon its original plan and focus on AI and AI instead, but now, Nvidia has taken the lead.
Originally, Nvidia could stand on the "peak", but Biden's team "blocked" them, and Huang Jenxun launched two models of chips A800 and H800 for the development of China, but because of the latest restrictions, these two models of products are not on the market.
According to data, the market value of China's AI market has reached more than 7 billion US dollars, and NVIDIA occupies 90% of the market, so NVIDIA has released three new products, namely HGXH20, L20PCLE and L2PCLE.
These three "customized versions" of the chips, namely training, inference and edge applications, are all built on the hopper and adalovelace frameworks, and will not be released until November 16 at the earliest, and the real production will not be until February of the following year, but some time ago there were rumors that these three "customized versions" of processors have dropped by more than 80%, are Chinese companies so easy to fool?
The end of Nvidia.
Judging from the data of **, these three artificial intelligence chips are not "improved version", but "shrunken version", in actual experiments, hgxh20 chips have been greatly restricted in bandwidth and computing speed, and can only reach about 20% of h100, that is, its performance has decreased by 80%.
You must know that the current artificial intelligence model, each artificial intelligence costs tens of millions of dollars, and the H100 chip alone has exceeded 10,000, and the GXH20 is used for AI training, which is like a contest between "adults" and "children", so even if you get an order, it will not be of much use.
Now China has basically achieved localization in the low-end field, whether it is a CPU for digital products or an artificial intelligence chip tailored for big data, as long as American companies cannot provide them with high-performance processors, it is only a matter of time before they lose China's ordersAfter European and American companies implement the "domestic" strategy, it will become very difficult to return to China.
Now Huawei's self-developed artificial intelligence chip, rising 910B, its computing power and Nvidia's A100 chip, compared with Nvidia's A100 chip, there is a lot of room for improvement, but this is only a feasible stage, not perfect enough, but also to improve computing power, power consumption and heat dissipation capacity, it is possible to catch up with NVIDIA.
On October 17, the United States issued a regulation on NVIDIA's AI chips, starting from October 23, NVIDIA's A800, H800, L40S and other high-end products will be prohibited from shipping without the corresponding license.
Judging from the current situation, the United States seems to have relaxed its suppression of China's semiconductor industry, but this is only a means, the purpose of the United States is to contain China's high technology, and AI and AI will dominate, and at that time, all the hard work of Nvidia will be in vain.
Huang Jenxun tried to "fool" Chinese companies with AI chips that were reduced by 80%, this idea obviously does not work, and domestic AI chips alone can overpower him, and there is no room for maneuver at all.
Bill Gates once said: "Nothing can stop China from obtaining high-quality chips!".Now, all of this is about to become a reality, and Intel has reacted: "What's the point of the U.S. being a global factory without orders from China?"”
In this clear position, Chinese enterprises have to work harder, don't expect European and American companies to have any "conscience", everything must start from profits, only to achieve complete localization, in order to reverse this situation, what do you think?