Re:TSMC and Graphcore Prepare for AI Acceleration on 3nm
by wetpaws on Thursday August 27, 2020 @01:24PM
(#24296053)
Gentle reminder that 3nm is a meaningless marketing number that does not represent neither the transistor size no element density, and it does not correlate in any way with all other Xnm production of other companies.
[ Parent
| on HN ]
Re:TSMC and Graphcore Prepare for AI Acceleration on 3nm
by phkahler on Thursday August 27, 2020 @01:37PM
(#24296235)
>> ... it does not correlate in any way with all other Xnm production of other companies.
True. But it should be significantly better/smaller than TSMC 5nm which is claimed to be 1.85x density over TSMC 7nm, which they are using now to make a chip with 59 Billion transistors.
The only other players to compare to are Intel and Samsung.
Re:TSMC and Graphcore Prepare for AI Acceleration on 3nm
by KaoruAoiShiho on Thursday August 27, 2020 @03:46PM
(#24297811)
It does correlate... generally the numbers respect actual end performance in respect to one another, nobody's having a smaller name that's actually worse than the bigger name of a different company.
Re:TSMC and Graphcore Prepare for AI Acceleration on 3nm
by doctoboggan on Thursday August 27, 2020 @01:32PM
(#24296177)
So what does it mean then?
Re:TSMC and Graphcore Prepare for AI Acceleration on 3nm
by Koshkin on Thursday August 27, 2020 @01:40PM
(#24296273)
According to Wikipedia [1], it refers to "the process' minimum feature size."
[1] https://en.wikipedia.org/wiki/Semiconductor_device_fabricati...
Re:TSMC and Graphcore Prepare for AI Acceleration on 3nm
by wetpaws on Thursday August 27, 2020 @01:44PM
(#24296318)
In many cases, like in Samsung case, it does not even refer to the feature size, and just represents a generation of the tech process or some arbitrary incremental improvements.
Re:TSMC and Graphcore Prepare for AI Acceleration on 3nm
by zionic on Thursday August 27, 2020 @01:38PM
(#24296247)
Literally nothing more than a marketing "group name" for a family of processes, none of which have any dimension approaching 3nm.