4 ms·

3x price increase for a 20x transistor increase = 85% cheaper per transistor, not to mention that the new transistors use far less power and thus generate less

by mouse_ 2y ago

3x price increase for a 20x transistor increase = 85% cheaper per transistor, not to mention that the new transistors use far less power and thus generate less heat/offer better battery life TSMC's results are always very impressive. Wonder how long they can keep this momentum.

ksec 2y ago
>TSMC's results are always very impressive. Wonder how long they can keep this momentum. For as long as the market is willing to pay for it. The oldest comment I wrote on HN was 2017, [1] but I made similar prediction on other forums in 2015. My original estimate was that by 3nm, which was suppose to come with GAA in 2022 or 2023 we will have a slow down simply because no company could afford it. That was at the time no single company reached a trillion market cap. Little did I know we have multiple multi trillion dollar market cap today, and AI continue to push the hardware spending forward. We have enough capital to push for at least 2nm, 1.4nm and 1nm till 2030. TSMC has roadmap till 0.8nm ( or A8 ). So unless the AI craze suddenly die off in the next two year, and despite a person not believing in AGI any time soon, I seriously doubt the craze will die off. As it could become the greatest jackpot in computing history since Smartphone. And Google, Facebook, Microsoft, Apple, Amazon, AMD, Intel, Broadcom, and many others together could be $500B annual spending on AI hardware alone. The market has enough money to push this forward. That is ignoring the government money as well! ( Every time I look into Nvidia it is more crazy number that I could imagine ) Not to mention ASML are working on much more cost effective method for their EUV TwinScan. This has a potential of $12K or even sub $10K per N3 wafer post 2030. The future is exciting. [1] https://news.ycombinator.com/item?id=14837701 https://news.ycombinator.com/item?id=14837701

by mouse_ 2y ago

ksec 2y ago

>TSMC's results are always very impressive. Wonder how long they can keep this momentum. For as long as the market is willing to pay for it. The oldest comment I wrote on HN was 2017, [1] but I made similar prediction on other forums in 2015. My original estimate was that by 3nm, which was suppose to come with GAA in 2022 or 2023 we will have a slow down simply because no company could afford it. That was at the time no single company reached a trillion market cap. Little did I know we have multiple multi trillion dollar market cap today, and AI continue to push the hardware spending forward. We have enough capital to push for at least 2nm, 1.4nm and 1nm till 2030. TSMC has roadmap till 0.8nm ( or A8 ). So unless the AI craze suddenly die off in the next two year, and despite a person not believing in AGI any time soon, I seriously doubt the craze will die off. As it could become the greatest jackpot in computing history since Smartphone. And Google, Facebook, Microsoft, Apple, Amazon, AMD, Intel, Broadcom, and many others together could be $500B annual spending on AI hardware alone. The market has enough money to push this forward. That is ignoring the government money as well! ( Every time I look into Nvidia it is more crazy number that I could imagine ) Not to mention ASML are working on much more cost effective method for their EUV TwinScan. This has a potential of $12K or even sub $10K per N3 wafer post 2030. The future is exciting. [1] https://news.ycombinator.com/item?id=14837701 https://news.ycombinator.com/item?id=14837701