Jump to content

too soon. it is like how Qualcomm announced a new snapdragon chip half a year ago but phones with it are only shipping half a year later. you cant get your hands on these until the buyers of these chip begin shipping them. This means intel, AMD, qualcomm and what have you need to get their products launched and in market as well. 

Sudo make me a sandwich 

Link to comment
https://linustechtips.com/topic/1612215-node-shrink/#findComment-16729302
Share on other sites

Link to post
Share on other sites

8 minutes ago, Dehaka said:

TSMC already has 2nm, why is nvda and amd still running 5nm ancient history?

Because the GPUs you buy today were designed—and their design validated—a long time ago. You can't redesign them for a new node on the fly, that takes time. New nodes also tend to be a lot more expensive and initial production tends to have lower yield rates. Plus node capacities need to be booked well in advance.

 

So it makes sense to release what you have right now, rather than redesign and revalidate everything as soon as a new node comes out and delay the release of a new GPU architecture by a few years.

Remember to either quote or @mention others, so they are notified of your reply

Link to comment
https://linustechtips.com/topic/1612215-node-shrink/#findComment-16729305
Share on other sites

Link to post
Share on other sites

23 minutes ago, Dehaka said:

TSMC already has 2nm, why is nvda and amd still running 5nm ancient history? when is the release date for next node shrink'd gpus and cpus? can't get any noticeable performance gain without node shrink.

TSMC has N2 up and running?
News to me.

TSMC N3 did not shrink SRAM much, and AMD does use 3nm for some parts of their CPUs (like Zen 5c epyc chips (which uses less SRAM then Zen5)). 

They could have used N3 for blackwell or rdna4; it was fully inside their time table, its just that they made some cost benefit analysis and found it lacking for their use case. probably due to SRAM/Analog scaling not being enough for the cost increase of the node. 

We already know Nvidia ordered Intels 18A PDK information. Neither TSMC N2 or Intels 18A are ramped for production yet, but just because of this knowledge, I do suspect the next product from them will be using one of those two nodes. 

AMD is using N2 for Zen 6 which comes out in 2026. 

Link to comment
https://linustechtips.com/topic/1612215-node-shrink/#findComment-16729306
Share on other sites

Link to post
Share on other sites

1 hour ago, Dehaka said:

will they release 3nm node gpus early next year?

Will who?
Blackwell next comes out in q4 2026/q1 2027 and I strongly suspect TSMC N2 or Intel 18A, Skipping N3 all together.
UDNA I also strongly suspect will be skipping N3

I dont have any information on Celestial to even speculate if they will be using Intel 3/Intel 18A/TSMC N3/TSMC N2

Again, SRAM scaling on N3 is an issue, so the cost to move to that node is not likely worth it for architectures that have so much cache on chip. A chiplet/tile design may work out like RDNA 3, and have the SRAM on a different node. 

Link to comment
https://linustechtips.com/topic/1612215-node-shrink/#findComment-16732689
Share on other sites

Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now

×