3 ms·
I like that we are moving away from (fake) nanometers and giving names to process nodes instead; comparing different technologies by Xnm was completely broken.
by Aissen 3y ago
I like that we are moving away from (fake) nanometers and giving names to process nodes instead; comparing different technologies by Xnm was completely broken.
- amelius 3y agoAt least you could meaningfully compare using "<" and ">". You can't do that with codenames.
- Gasp0de 3y agoHow was it meaningful? What would the info "This one is called 5nm and this one is called 8nm" give you? Because it didn't really say anything about performance or watt or performance per watt. I have an old ATI Radeon HD4850 GPU, I bet it beats your Nvidia RTX 4090 since 4850>4090 ;)
- kevincox 3y agoThat is true, but just calling it the TSMC 2024 Process also allows comparing. But they don't like using dates because then if they stay on the process for a few years it is more obvious that it is "old".
- mastax 3y agoTechnically TSMC doesn't use nm. It's the N2 node, not the 2nm node, but tech press is the way it is.
- eis 3y agoUnfortunately TSMC does indeed use nm extensively https://www.tsmc.com/english/dedicatedFoundry/technology/logic https://www.tsmc.com/english/dedicatedFoundry/technology/log...
- dragonelite 3y agowell or else you will get intel naming scheme of using 10nm ++++++
- wmf 3y agoThe node names have numbers in them that are clearly intended to be compared. It's no coincidence that "Intel 3" and "N3" both have 3 in them.
- rbanffy 3y agoIdeally it could be a measurement of density - how many transistors can dance on the head of a pin.
- bee_rider 3y agoI wish we could get a measure like transistor density for some reference designs or something. (I mean it would be nice for us if every transistor was the same size, but just like we provide benchmarks for different instruction mixes, I guess we would want density benchmarks for low power/high chip flavors).