Taiwan's DigiTimes has published a report about the tidal wave of orders at TSMC, which it reckons will push the contract chipmaker's revenues, and equipment investments, to new heights. Huawei coming under US Sanctions in mid-September left a 14 per cent revenue hole in TSMC's books but thanks to pent up demand, TSMC has plugged that hole with an influx of orders and it is thought its previously stated annual revenue growth target of 20 per cent will be easily achieved.
The DigiTimes story, in Traditional Chinese, talks a little about orders with the likes of Apple and Sony before it adds into the mix some interesting insider info concerning TSMC's dealings with Nvidia. In brief, it asserts that Nvidia is looking to switch GPUs.
The Nvidia portion of the report relays that Nvidia signed up to product GeForce RTX 30 series GPUs using Samsung's 8nm production facilities as it received a "considerable contract discount," on this business. At least some of this business will be shifted to TSMC in 2021, asserts the source. This will account for a "large number of orders," for TSMC and help it maintain a very high capacity utilisation in 2021.
Consumer Nvidia Ampere currently on Samsung 8nm
Elsewhere in the report it is claimed that Nvidia previously planned to spread the production of its Ampere GPUs across both Samsung and TSMC for better risk management. If you remember back to mid-May when the Ampere architecture and the A100 GPU for data analytics, scientific computing and cloud graphics was launched, it was confirmed that this serious GPU was fabricated on TSMC 7nm. The latest DigiTimes report appears to says that Nvidia RTX 30 GPUs will be joining the pro-grade GPUs on TSMC 7nm.
Professional Nvidia Ampere GPUs currently use the TSMC N7 process
It will be interesting to see if the new TSMC made RTX 30 series will be marketed as 'Super' cards or form the foundation for a complete refresh. It is possible that mid-to-lower-end GeForce Ampere GPUs will still be fabricated by Samsung.
N.B. I found the Microsoft Bing translation of the DigiTimes segment about Nvidia to be clearer than the Google one.
