News reports that Nvidia H200/B100 chip orders are strong, and TSMC's 3/4nm production capacity is close to full capacity
The AI annual conference, NVIDIA GTC, will be held on March 17th Western Time in the United States. The market estimates that H200 and B100 will be released in advance to grab the market. It is understood that the H200 and the new generation B100 will adopt TSMC's 4nm and 3nm processes respectively. The H200 will be launched in the second quarter, and it is rumored that the B100 adopts a Chiplet design architecture and has been ordered for production. The legal representative pointed out that Nvidia has strong orders, and TSMC's 3nm and 4nm production capacity is almost fully loaded, and the first quarter of operation is not weak.
Regarding the issue of Nvidia's new generation chip orders occupying TSMC's advanced processes, TSMC stated that the production capacity process will still follow the content stated in the previous legal statement and will not be further explained.
According to reports, the B100 of the Nvidia Blackwell series is seen by the market as the next generation Nvidia GPU weapon. In addition to being first built using TSMC's 3nm technology, it is also the first Nvidia product to be packaged in Chiplet and CoWoS-L formats, solving high power consumption and heat dissipation problems, single card efficiency, and crystal tube density. It is estimated to surpass AMD's MI300 series launched in the first quarter.
Server manufacturer Dell has revealed Nvidia's upcoming Artificial Intelligence (AI) GPU Blackwell, which has a power consumption of up to 1000W, a 40% increase from the previous generation of chips, and requires Dell to use its innovative engineering to cool these GPUs.
According to current market news, the Nvidia B200 has more powerful computing performance than the current H100 product, but its power consumption is also more astonishing, expected to reach up to 1000W, an increase of over 40% compared to the H100. Nvidia's H200 chip is considered the most powerful AI computing chip in the industry due to its Hopper architecture and HBM3e high bandwidth memory. It is estimated that due to the computing power of B100 chip being at least twice that of H200, which is four times that of H100, the computing performance of B200 will be even more powerful.
TSMC's advanced processes continue to be fully loaded, with TSMC's capacity utilization rate exceeding 90% in February, and the demand for artificial intelligence (AI) remains unchanged. According to the supply chain, applications such as AI and high-performance computing (HPC) can produce only a quarter of the number of chips produced on a single wafer compared to consumer products, making production and manufacturing more difficult and complex; TSMC's ability to achieve stable mass production is crucial for the chip industry. TSMC currently accounts for 43% of its revenue from HPC/AI application platforms, which is on par with smartphones.
According to reports, the B100 of the Nvidia Blackwell series is seen by the market as the next generation Nvidia GPU weapon. In addition to being first built using TSMC's 3nm technology, it is also the first Nvidia product to be packaged in Chiplet and CoWoS-L formats, solving high power consumption and heat dissipation problems, single card efficiency, and crystal tube density. It is estimated to surpass AMD's MI300 series launched in the first quarter.
Server manufacturer Dell has revealed Nvidia's upcoming Artificial Intelligence (AI) GPU Blackwell, which has a power consumption of up to 1000W, a 40% increase from the previous generation of chips, and requires Dell to use its innovative engineering to cool these GPUs.
According to current market news, the Nvidia B200 has more powerful computing performance than the current H100 product, but its power consumption is also more astonishing, expected to reach up to 1000W, an increase of over 40% compared to the H100. Nvidia's H200 chip is considered the most powerful AI computing chip in the industry due to its Hopper architecture and HBM3e high bandwidth memory. It is estimated that due to the computing power of B100 chip being at least twice that of H200, which is four times that of H100, the computing performance of B200 will be even more powerful.
TSMC's advanced processes continue to be fully loaded, with TSMC's capacity utilization rate exceeding 90% in February, and the demand for artificial intelligence (AI) remains unchanged. According to the supply chain, applications such as AI and high-performance computing (HPC) can produce only a quarter of the number of chips produced on a single wafer compared to consumer products, making production and manufacturing more difficult and complex; TSMC's ability to achieve stable mass production is crucial for the chip industry. TSMC currently accounts for 43% of its revenue from HPC/AI application platforms, which is on par with smartphones.