Nvidia has reportedly signed an agreement with TSMC to allow its next generation of graphics cards to be engraved in 3nm. The same engraving process is currently used by Apple to manufacture its latest A17 Pro processors.
According to information shared by the Taiwanese economic media DigiTimes, Nvidia has reached an agreement with TSMC to engrave its next generation of graphics cards in 3nm, which will likely be codenamed “Blackwell”.
As WCCFTech points out, this engraving finesse should initially benefit the “GB100” GPU, intended mainly for servers and supercomputers. The latter would constitute the first batch of GPUs entirely based on AI to be designed by the chameleon brand. This chip would also benefit from a design in chipletsupposed to bring “ breakthrough improvements in performance per watt ”, we read. Previous rumors, however, suggested that due to its “ structural design » much more complex, this GPU could only be manufactured piecemeal, from the month to the start of its production period.
The objective? Full focus on AI to remove AMD and Intel from this market
“ For Nvidia, which monopolizes more than 80% of the AI GPU market, the next generation B100 will use TSMC’s 3nm engraving and is expected to appear in the fourth quarter of 2024. It will seize commercial opportunities for AI deployment while the iron is hot and will supplant AMD, Intel and other competitors », Advances DigiTimes.
For context, Nvidia already benefits from a predominant position in the market for GPUs dedicated to AI. And the firm does not intend to stop there. We know, for example, that it aims to generate $300 billion in turnover thanks to AI by 2027. The idea will then be to leave as little room as possible for competition.
The transition to 3 nm and the arrival of new “Blackwell” GPUs, from the end of next year, is therefore crucial for Nvidia if it wishes to maintain its technical supremacy over Intel and AMD.
Note that in this case, the brand seems to be on the right track to achieve this. According to the latest rumors, AMD will launch its EPYC Instinct MI300 chips later this year, but only in 5nm. Intel, for its part, would deploy its Gaudi 3 AI accelerator next year, but again in 5 nm. The two brands would therefore not be able to move to 3nm GPUs before 2025, or even 2026… leaving Nvidia a nice head start.
Want to join a community of enthusiasts? Our Discord welcomes you, it is a place of mutual help and passion around tech.