According to The Information, design defects will delay The release of Avida's next-generation Blackwell GPU architecture by three months or more. Blackwell is the successor of the current generation of Hopper in the market. It was originally scheduled to be released in the fourth quarter of this year, but the report said Avida has informed its main GPU customer Microsoft, its products will be delayed for at least another three months, that is, the release time will be postponed to the first quarter of next year.

If Microsoft is affected, other major customers such as Meta(Facebook), Google and Oracle will also be affected. If they are affected, other enterprises do not need to consider their own orders, because Avida will give priority to meeting the needs of these ultra-large-scale cloud customers.

According to a report by SemiAnalysis, a semiconductor research institution, the problem lies in the manufacturing process, especially in chip packaging. & ldquo; The core issue affecting shipment volume is directly related to the design of Avida Blackwell architecture. Due to the packaging problems in the design of TSMC and Avida, the supply of original Blackwell packaging is limited, & rdquo;SemiAnalysis wrote.

The SemiAnalysis report details the problem & mdash;& mdash; Which is far beyond the technical understanding scope of ordinary people. Obviously, Blackwell is an extremely complex silicon chip, which has a large number of possibilities of errors, but in fact it does. This problem is both the responsibility of TSMC and Nvidia. Nvidia issued the following statement: & ldquo; As we have introduced before, the demand for Hopper is very large, the wide application of Blackwell has begun, and production will increase as planned in the second half of the year. In addition, we will not comment on rumors. & rdquo; However, as an analyst said, Nvidia should disclose the product if there is indeed a delay in product delivery and the delay may have a significant impact on revenue. As Nvidia gives priority to supplying large enterprises that invest billions of dollars in orders at a time, these customers are most likely to feel pressure. Other enterprises are adopting a slower speed, and they may have just begun to deploy Hopper-based systems. Therefore, if ordinary enterprises order some Hopper-based systems, then there is no need to worry too much. Earlier this year, Nvidia announced that it would launch a new microarchitecture every year instead of every two years as before. At that time, people thought that this update frequency was full of risks and might put Nvidia in trouble, because such a fast release frequency meant that there was no buffer space for making mistakes. This will require perfect execution, but no enterprise is perfect, even Nvidia is no exception.

Now it seems that the worry really happened. The frustrating thing for Nvidia is that the problem is beyond its control. This is the manufacturing problem of TSMC. Yes, the two companies do cooperate in product packaging, but the main responsibility lies in TSMC. AMD should be cautious because it also announced plans to launch a new GPU architecture for its increasingly popular Instinct series every year. If there is the most understanding of the consequences of mistakes, AMD is the only one. In 2006, AMD fought against Intel with a highly competitive product, which was smooth sailing at the beginning, but & ldquo; Barcelona disaster)& rdquo; Gave it a hit & mdash;& mdash; This is a new architecture that tries to do too much work at the same time, but the delivery is delayed and the performance is poor. AMD lost all the strong momentum and hovered at a low level in the next ten years, almost completely defeated by Intel until the revival of Zen architecture. AMD certainly does not want this to happen again.

Source: Open-Source fans IT selected news