Moving towards reducing Openai’s dependence on NVIDIA and its graphics processors is key to designing its own dedicated artificial ielligence chip. The ChatGPT manufacturer has achieved remarkable gains. The artificial ielligence startup is said to be in the final phase of the design of the dedicated chip, a process expected to be completed in the next few mohs. If the trend goes well, Openai can send its construction chip to TSMC for mass production in the first half of this year.
Although there may be a variety of challenges on OpenAI programs, the company seems to be very serious to achieve its goal. According to CNBC, the Tape-OUT process will take about 6 mohs and cover a cost of millions of dollars. However, if Openai was to pay more to TSMC, it is possible that TSMC will accelerate the process of producing artificial ielligence chip. Unfortunately, there is no guaraee of the success of the first test production, as any failure means the need to repeat the experimeal production process and the effort to fix the problem.
Previously reports of OpenAI’s use of the TSMC A16 Angstrom process for the Sora video manufacturer had been released, but it is not yet clear whether the artificial ielligence chip is an example that is supposed to be finalized in the coming mohs or a differe iernal solution. This special section is currely run by Richard Ho of Openai and its number of specialist staff has increased to 40. The iernal chip design process is done in collaboration with Broadcom, although the exact amou of participation is still unclear.
The name of the Custom Artificial Ielligence chip has not yet been officially announced, but its main function will focus on training and launching artificial ielligence models. The capabilities of this silicone chip will be limited at first, but this capacity can be increased according to the number of units that ChatGpt’s company iends to use in the future. According to reports, as planned, mass production is expected to be from 2026, and TSMC of its 3NM technology to produce the chip, along with the systolic array architecture and the HBM, which NVIDIA in its artificial ielligence graphics processors He will use it.
Source: 9TO5MAC
