Support NeoGAF

[Unknown Soldier]

NVIDIA Reportedly The First & Only Customer For TSMC's Bleeding-Edge A16 Process Node, Utilizing For Next-Gen GPUs

NVIDIA is reportedly the only customer in line for TSMC's next-gen A16 process technology, eyeing it for future GPUs such as Feynman.

wccftech.com

For those who don't know, for probably the entire past decade Apple has been the first and often only customer for TSMC's newest and most advanced process

If this is true, it's a big change that it is now Nvidia in the driver seat as the first and only customer for TSMC's newest A16 process

The next Nvidia architecture after Blackwell is Feynman

Richard Feynman - Wikipedia

en.wikipedia.org

 

[Magic-DEI-Hire]

This isn't really news-worthy as much as you may think. Apple has been known for focusing on efficiency gains and power reduction with its internal chip designs (their most recent 3nm chips are on their third generation as of this year). I expect them to do the same for 2nm before going past that. Their chip design has been paying dividends (just look at their M series chips as an example of what they're capable of). Apple is making an entire system on a package (GPU, CPU, memory, AI cores) yet does so while maintaining both a high performance and a low power draw.

Apple M5 Chip: Why It's Sticking to 3nm Technology - TechWiser

Apple is working on the development of its next-generation M…

techwiser.com

Apple unleashes M5, the next big leap in AI performance for Apple silicon

Apple today announced M5, delivering advances to every aspect of the chip and the next big leap in AI.

www.apple.com

https://wccftech.com/apple-not-yet-in-talks-with-tsmc-over-a16-process/

For the time being, the Cupertino firm has reportedly secured more than half of TSMC's initial 2nm supply, with the technology being leveraged not just for chipsets that will fuel the iPhone 18 series, but also the M6 that is expected to be found in the new MacBook Pro models. After 2026, Apple may move to the slightly improved 2nm 'N2P' process for the A21 and A21 Pro […] there is also the possibility that Apple moves to TSMC's A14 node, or 1.4nm instead of A16, and we say this because the wafer manufacturer has already taken the first steps towards building production plants on its home turf, with an estimated investment of a whopping NT$1.5 trillion, or approximately $49 billion.

 

[RodriGames40]

Good news. Feynman/Rubin should be a sizeable upgrade for a change.

 

[Magic Carpet]

That sucks sour lemons.

 

[asdasdasdbb]

Can't wait to see what they charge for Feyman.

 

[Kuranghi]

Woohoo more overpriced GPUs, I'll PC game in native 4K again when I'm 60.

 

[kiphalfton]

Woohoo more overpriced GPUs, I'll PC game in native 4K again when I'm 60.

Hey look, we got an optimist in here.

 

[dgrdsv]

If this is true, it's a big change that it is now Nvidia in the driver seat as the first and only customer for TSMC's newest A16 process

It's kinda not.
Nvidia has always booked new processes in advance for future parts.
These may eventually end up not being used or used for some DC parts only.
Has been the case for about forever, even during Ampere times they've used TSMC's N7 for GA100.

The next Nvidia architecture after Blackwell is Feynman

Next one after Blackwell is Rubin, as shown in your own link.
Then it's Feynman.
But no one knows how these will map to gaming parts. Current expectation is for Rubin to also have gaming lineup but as for Feynman - this may end up being DC only.

 

[Unknown Soldier]

It's kinda not.
Nvidia has always booked new processes in advance for future parts.
These may eventually end up not being used or used for some DC parts only.
Has been the case for about forever, even during Ampere times they've used TSMC's N7 for GA100.

Up until recently, Apple was buying out the entire production of TSMC's most advanced process for the first year or so

That's not to say that Nvidia wasn't using TSMC because you are right that they were. Consumer Ampere was made on Samsung 8N tho

Next one after Blackwell is Rubin, as shown in your own link.
Then it's Feynman.
But no one knows how these will map to gaming parts. Current expectation is for Rubin to also have gaming lineup but as for Feynman - this may end up being DC only.

Ah all right. I admit I haven't been real clear on Nvidia's roadmap since they started having some parts only for datacenter and some only for consumer

 

[//DEVIL//]

What is A16 process node ?

and what is Feyman ? 6000 series ? or 7000 ?

 

[DeepEnigma]

And you'll still get 8GB of RAM.

 

[Wolzard]

What is A16 process node ?

and what is Feyman ? 6000 series ? or 7000 ?

A16 = 1.6 nm
Feyman = 7000

 

[winjer]

I bet this will be for AI chips. Only these chips are profitable to be made on a cutting edge node, exclusively for one company.
It will take a while, until gaming GPUs are made in such an advanced node.

 

[Griffon]

We will pay premium for the leftovers, probably will not see this node in consumer GPUs until the next gen after that one.

 

[asdasdasdbb]

What is A16 process node ?

A16 is just a renamed N2P with "Super Power Rail" (because presumably TSMC is going to charge more).

Everyone else is going to use A14, which is better even without SPR.

 

[The Cockatrice]

0 chances of these being for consumer GPU's.

 

[HJuggernaut]

we are now going with decimal point nm, interesting.

 

[Dr.D00p]

Well, at least it should deliver a genuine node shrinkage performance improvement over the faux 50xx GPUs, otherwise known as the 40xx Supers.

 

[StereoVsn]

I bet this will be for AI chips. Only these chips are profitable to be made on a cutting edge node, exclusively for one company.
It will take a while, until gaming GPUs are made in such an advanced node.

On the bright side it could free up some space on the other nodes.

 

[diffusionx]

This process is not going to be used for vidya GPUs, are you guys even paying attention to what is going on in this industry?

 

[asdasdasdbb]

This process is not going to be used for vidya GPUs, are you guys even paying attention to what is going on in this industry?

Course they will. They will just make the die sizes smaller and/or jack up the price.