NVIDIA's castration chip issue has sparked a domestic debate.
There is news that due to the fierce competition in China's AI chip market, NVIDIA has been forced to downgrade, and there are also some news reports that due to the scramble of Chinese manufacturers to buy NVIDIA AI chips, nvidia's ** businessmen have begun to silently raise **, is this NVIDIA reduced**, or raised**?
In the second half of 2022, NVIDIA launched a weakened version of the A800 and H800 for the Chinese market, which are favored by Chinese companies because these two chips have not been cut so hard, and some Chinese high-tech companies have also signed a large order of $5 billion in the second half of 2023, which shows the importance that local companies attach to NVIDIA.
However, Raimondo of the US Department of Commerce believes that the A800 and H800 are too powerful, so they strongly advise NVIDIA not to sell them in China, which made NVIDIA have to reduce its performance, and released the H20 of the H20, which is said to be 80% weaker, but it's not reduced.
At this time, a Chinese artificial intelligence company released a new artificial intelligence chip, which is said to have better performance than H20, so many companies chose this product and got 5,000 orders within a few months, with a total contract value of more than 500 million, which also pushed NVIDIA to keep the H20 price to the minimum.
When a company in Huaxia asked nvidia's ** business, they told reporters that H20's ** is only about 1 4 of foreign manufacturers, but in terms of performance, it has not been affected too much, so it will be used in a large number of markets, because the sales are very good, so the seller intends to raise the price, which is completely different from H20's**.
Why is there such a contradiction? The author speculates that both may be correct, but the goal of the comparison is different. The reason for this reduction is that compared with A800 and H800, the former is priced at more than 100,000, the latter is above 260,000, and the latter is only 80,000.
The reason why I say this is because the price of domestic AI chips is not low, because the price of domestic AI chips is between 110,000-120,000, and compared to H20, their price is much higher, and their performance will not be much better than H20.
At present, domestic AI chips are still facing compatibility problems, because most of the previous AI software is built on NVIDIA's CUDA platform, so the use of domestic AI chips, there will be compatibility problems, coupled with the relatively backward technology of domestic AI chips, resulting in relatively high power consumption, which will bring great difficulties to domestic enterprises to use domestic chips.
In this case, many network companies that originally used NVIDIA chips still prefer to use H20, because the market demand is still very large.
In a sense, the reason why Chinese companies make this decision is because they are very unhappy with Chinese AI chips, because they think they are helping local companies, but when they spend more money to buy a domestic AI chip, they become very unhappy, so they will buy NVIDIA chips.