AMD in a hurry? RTX 40 uses 4nm, Su Ma hurried to TSMC to discuss new process

time:2022-11-27 03:02:33 source:scripttoolbox.com author:Loudspeaker box
AMD in a hurry? RTX 40 uses 4nm, Su Ma hurried to TSMC to discuss new process

Although many companies have adopted TSMC's 4nm process, such as Qualcomm, MediaTek, etc., but we did not expect NVIDIA to go so radically this time, directly jumping from Samsung's 8nm process used by the previous generation RTX 30 to TSMC. 4nm process. You must know that NVIDIA announced the GPU with the core code name of Hooper in the first half of the year, and it still uses TSMC's 5nm process, so we think that the Ada Lovelace core used for games in the second half of the year should also use 5nm chips, so that AMD and NVIDIA will use 5nm chips in graphics chips. The craftsmanship is still at a level. I just didn't expect NVIDIA to skip 5nm on the game graphics card and directly use TSMC's 4nm, which is somewhat irrelevant! Of course, NVIDIA's approach is not incomprehensible. The core of this generation of Ada Lovelace architecture has more transistors and higher frequency. It also needs to take care of heat and power consumption. Using TSMC's latest 4nm is indeed the optimal solution. Moreover, TSMC's 4nm process was originally optimized from 5nm, so there is no pressure to switch from 5nm to 4nm. It's just that the cost has increased, which has also increased the price of the RTX 40 to a certain extent. However, this is obviously a bit unexpected by AMD. You must know that AMD's RNDA 3 architecture core has previously disclosed a mixed package of 5nm and 6nm, in which the traditional GPU core will be composed of multiple small chips, using TSMC's 5nm process , and AMD also designed an MCD module in the chip, which is equivalent to the IOD module of the CPU, and this module uses the TSMC 6nm process, which is optimized from the TSMC 7nm process. AMD has been competing with NVIDIA and Intel before, vigorously promoting its technological advantages in chips. But now that NVIDIA's single-chip solution adopts TSMC's 4nm, AMD's advantage is gone. This is undoubtedly a lot of pressure on AMD, so it is reported that AMD will visit TSMC in the near future, and the specifications are extremely high. The CEO Su Ma personally leads the team and hopes to discuss future cooperation with TSMC, especially the 3nm and 2nm processes. You must know that AMD's success in recent years is inseparable from TSMC's advanced chip technology. Therefore, in order to maintain its advantage in the chip part, AMD must strengthen its cooperation with TSMC in the future, especially to obtain more capacity allocation at 3nm and 2nm. Judging from AMD's current development blueprint, this year's graphics cards and CPUs will continue to use TSMC's 5nm technology, and may transition to TSMC's 4nm technology next year. However, both Zen 5 architecture processors and RDNA 4 architecture graphics cards should use 3nm technology, and TSMC will launch 2nm technology in 2025, so AMD should indeed discuss future advanced technology cooperation with TSMC at this time. In addition to advanced chip technology, AMD also hopes to develop more high-tech cooperation with semiconductor companies in Taiwan, especially 3D packaging. At present, due to AMD's efforts to develop Chiplet multi-chip packaging technology, and TSMC's technology accumulation in this part is also in a leading position, the two companies will also discuss cooperation in the 3D packaging part. At present, AMD is already using TSMC's 3D SoIC (system-on-a-chip) platform, including packaging technology and packaging solutions, and this is also an advantage that AMD currently has on the product side, so in addition to chip technology, AMD also needs to cooperate with TSMC discussed the price and distribution of this part of the technology. After all, more companies will definitely use TSMC's 3D packaging solution in the future. Speaking of this generation of graphics card technology, although NVIDIA uses TSMC's 4nm process and AMD uses 5nm process, but fortunately AMD's GPU architecture uses a multi-chip packaging solution, so although the process is slightly inferior to NVIDIA, However, power consumption and heat generation will still have obvious advantages. From the perspective of performance estimates, AMD's flagship products may lag behind NVIDIA by about 10% in performance, but in terms of energy consumption ratio, AMD undoubtedly has a clear advantage. But what needs to be reminded is that the price of NVIDIA's RTX 4090 has soared to 12,999 yuan, and the price of AMD's generation of graphics cards will not be much cheaper. We estimate that the price of RX 7900 XT is still expected to exceed 10,000 yuan. Although AMD will officially release the new generation of RX 7000 graphics cards on November 3, it is basically a trend that is more expensive than the previous generation of graphics cards. Be prepared for it! It can only be said that the graphics card is really unaffordable...

(Responsible editor:CPU)

Related content