News Nvidia is reportedly close to adopting Intel Foundry's 18A process node for gaming GPUs

[Admin]

According to an analyst, Broadcom and Nvidia are reportedly close to adopting Intel Foundry as their manufacturing partner.

Nvidia is reportedly close to adopting Intel Foundry's 18A process node for gaming GPUs : Read more

 

[Flemkopf]

I'm glad to hear there's some interest in Intel's foundry services. I don't think anyone outside of TSMC is happy about the current bleeding-edge manufacturing situation. Samsung's 30% yield on its 2nm is not great, although that article should be taken with a grain of salt, and some quick Google-foo suggests that its more mature 5nm is closer to 88%. GlobalFoundries never made it past about 12nm.

Seriously, just having options outside of a single company is going to be critical for everyone from both a security and a market perspective. I don't think anyone in the global market wants to be relying on SMIC and the Chinese government, and reducing single points of failure in the event of any widespread disruption will be huge.

 

[logainofhades]

Having Nvidia as a customer would definitely give Intel some much needed cash right now.

 

[gg83]

Fingers are crossed for 18A!!!! Intel needs a win.

 

[Gururu]

AMD is screwed.

 

[logainofhades]

AMD is screwed.

How is AMD screwed exactly? Nvidia moving to Intel just gives TSMC more capacity for AMD chips. The TSMC Arizona fab will be making AMD chips as well.

 

[Notton]

This should be a win-win for PC gaming if 18A works out.
Nvidia would no longer have to split 1% of their allocation at TSMC for consumer GPUs and they won't have to trickle it out.

Nvidia tried this in the past with Ampere, but Samsung 8N was... yeah... 🙄
And if 18A works for Geforce GPUs, there's no reason it won't work with Radeon or Ryzen.

(I'm not saying the GPU prices will come down, but at least you might be able to buy one)

 

[TerryLaze]

How is AMD screwed exactly? Nvidia moving to Intel just gives TSMC more capacity for AMD chips. The TSMC Arizona fab will be making AMD chips as well.

TSMC is booked out for a couple of years, nvidia making some chips at intel isn't going to change that, it's not moving capacity from tsmc to intel, it's ADDITIONAL.
Also AMD doesn't have an issue with capacity but with paying for it, they are not exactly swimming in money with only their datacenter segment doing well.

 

[renz496]

How is AMD screwed exactly? Nvidia moving to Intel just gives TSMC more capacity for AMD chips. The TSMC Arizona fab will be making AMD chips as well.

Nvidia capacity will not change. Using intel fab only means that they can devote all their TSMC capacity for AI chip. Also depending on the situation nvidia top gaming gpu probably will still going to be made by TSMC.

 

[bit_user]

Intel is prepping its performance-enhanced 18A-P fabrication technology, which promises to either increase performance at the same power, or decrease power at the same performance.

You could just say it improves energy efficiency (i.e. perf/W). The two examples you gave are just points on the perf/W curve typically used to give a rough sense of how the improvement is shaped. However, with no specific numbers given, it makes more sense just to say it's more efficient.

 

[bit_user]

GlobalFoundries never made it past about 12nm.

They almost had 7 nm, but killed it in 2018, because they were running behind schedule. I'll bet they were kicking themselves for that, back in 2021.

More recently, I thought I saw something about them licensing someone else's node, but now I can't find it.

 

[bit_user]

This should be a win-win for PC gaming if 18A works out.

Depends on pricing. Somehow, I have a hard time believing 18A will be cheaper per transistor than TSMC N4P. At least, not for a while.

That's okay, if what you're making on it is either small CPU chiplets or big, expensive server CPUs. But, gaming GPUs need to be both big and cheap. That's why they're rarely on a very new node.

 

[bit_user]

depending on the situation nvidia top gaming gpu probably will still going to be made by TSMC.

That's something I think we've never seen. Because their gaming GPUs all share the same architecture, it's more economical for them to make them all on the same node. If they spit, the way that makes sense is to put the HPC/AI chips or embedded chips on a different node than the gaming chips. Otherwise, they have to do custom layout for their gaming GPU logic twice, which is very costly.

Also, if they were really going to split nodes, I doubt they'd put the lower end gaming GPUs on such a new & expensive node as 18A. I think they'd use an older, cheaper node.

 

[TerryLaze]

Depends on pricing. Somehow, I have a hard time believing 18A will be cheaper per transistor than TSMC N4P. At least, not for a while.

That's okay, if what you're making on it is either small CPU chiplets or big, expensive server CPUs. But, gaming GPUs need to be both big and cheap. That's why they're rarely on a very new node.

Being able to use 18A for prime products will free up a lot of capacity on their older nodes, they could make less efficient but powerful and relatively cheap GPUs, and in masses.
And I don't know if nvidia will be too cheap to make multiple designs for multiple nodes, they are having a tough time.... /s , but intel will most definitely use all their nodes as best as they can.