NVIDIA Could Use Intel's Foundry Service for Chip Manufacturing

[AleksandarK]

Yesterday, NVIDIA announced its next-generation Hopper architecture designed for data center applications and workloads. There is always a question of availability, as the previous period showed everyone that the supply chain is overbooked and semiconductors are in very high demand. During the Q&A press session today, NVIDIA's CEO, Jensen Huang, tried to answer as many questions as possible. However, an exciting topic arose regarding the potential collaboration with Intel. As a part of Intel's IDM 2.0 strategy, the company plans to offer its chip manufacturing capabilities to the third-party companies willing to make efforts and port their designs to Intel's semiconductor nodes. NVIDIA, one of the largest TSMC customers, could be a new Intel customer. Below, we compiled a few quotes that highlight Jensen Huang's opinions, taking the quotes from Tom's Hardware.

Our strategy is to expand our supply base with diversity and redundancy at every single layer. At the chip layer, at the substrate layer, the system layer, at every single layer. We've diversified the number of nodes, we've diversified the number of foundries, and Intel is an excellent partner of ours[…]. They're interested in us using their foundries, and we're very interested in exploring it. [...] I am encouraged by the work that is done at Intel, I think this is a direction they have to go, and we're interested in looking at their process technology. Our relationship with Intel is quite long; we work with them across a whole lot of different areas, every single PC, every single laptop, every single PC, supercomputer, we collaborate. [...] We have been working closely with Intel, sharing with them our roadmap long before we share it with the public, for years. Intel has known our secrets for years. AMD has known our secrets for years. We are sophisticated and mature enough to realize that we have to collaborate.[...] We share roadmaps, of course, under confidentiality and a very selective channel of communications. The industry has just learned how to work in that way.

In addition, he also noted that talking about collaborating with different foundries is not as easy, as separate plans and business models have to align to get a successful contract/deal. They can take a lot of time to get on the proper grounds; however, they can get it going if there is a desire to work.

View at TechPowerUp Main Site | Source

 

[Punkenjoy]

Yeah so Intel would make their GPU at TMSC so that Nvidia can make theirs at Intel.

I think it would be a shame if Intel made Nvidia GPU kick TSMC made Intel GPU...

 

[phanbuey]

Yeah so Intel would make their GPU at TMSC so that Nvidia can make theirs at Intel.

I think it would be a shame if Intel made Nvidia GPU kick TSMC made Intel GPU...

Probably the support chips / super cheap OEM gpus - not high end.

 

[trsttte]

Yeah so Intel would make their GPU at TMSC so that Nvidia can make theirs at Intel.

I think it would be a shame if Intel made Nvidia GPU kick TSMC made Intel GPU...

Whenever Intel has enough leading edge node capacity to be able to sell some they won't be using TSMC for GPUs either

 

[Wirko]

After Intel+AMD (Kaby Lake G) and Nvidia+AMD (Ampere DGX), this would be the third and final example of incest in the computing world.

 

[jeremyshaw]

After Intel+AMD (Kaby Lake G) and Nvidia+AMD (Ampere DGX), this would be the third and final example of incest in the computing world.

Little too late for that Nvidia used to make MB chipsets for Intel and AMD platforms (Comparison of Nvidia nForce chipsets - Wikipedia) before that ended with a bunch of lawsuits that saw Nvidia temporarily gaining an x86 license by the FTC, then trading it for Intel's promise to keep the PCIe bus in their SoCs and CPUs (and for Intel to not cut off VIA's x86 license, since VIA had no leverage against Intel, vs AMD with AMD64 becoming the defacto 64bit x86 instruction set).

 

[persondb]

Probably the support chips / super cheap OEM gpus - not high end.

They could also do SoCs and stuff like that there, not their highest-end but they have big markets in stuff like hobbyist(e.g. Jetson Nano), automobile, AI and etc.
Nvidia uses a lot of Samsung 8nm, which does seem to be considerably below the competing nodes from Intel(10nm/Intel 7) and TSMC and they could try using Intel nodes for something like their Orin SoC.

 

[looniam]

RTX 5090
RTX 5090+
RTX 5090++

 

[watzupken]

It will be interesting if that happens. Both Nvidia and Intel are equally as stuck up, so I do wonder how will this uneasy partnership work and how long it will last. Historically, Nvidia is always looking out for cheapest competitive fab to cut cost, which is why they ditched TSMC for Samsung‘s 8nm for their consumer GPUs.
At the end of the day, I do wonder if these aggressive foundry expansion will work in the longer run, bearing in mind the resources required by the foundries are still going to be limited to the same players. The latest news about limitation with ASML to produce enough lithography machines is one of the many examples.

 

[Vayra86]

RTX 5090
RTX 5090+
RTX 5090++

That would be Super!

 

[stimpy88]

So nGreedia wants their chips fabbed on a 10nm++ process now... Mind you, it has to be better than that awful Samsung node they currently use.

 

[trsttte]

At the end of the day, I do wonder if these aggressive foundry expansion will work in the longer run, bearing in mind the resources required by the foundries are still going to be limited to the same players. The latest news about limitation with ASML to produce enough lithography machines is one of the many examples.

People tend to focus on lithography, but there are many other steps until you get a finished chip. Intel fabs have lagged behind in the lithography as of late and are trying to catch up but they're still historically very good at everything else involved in the process (packaging for example).

I think a different problem that may happen is over capacity which will drive prices down and kill a lot of the margins manufacturers are currently enjoying - it's great in the short term but as much as we want cheap stuff it's not really sustainable

 

[MxPhenom 216]

So nGreedia wants their chips fabbed on a 10nm++ process now... Mind you, it has to be better than that awful Samsung node they currently use.

The Samsung 8nm isnt a terrible node, and theres no information on what node Nvidia would start using from Intel. By the time this is in full swing Intel could be onto their next node.

 

[Punkenjoy]

A key thing to remember is it's not because Intel is lagging behind right now that it will always do. Sometime you learn way more in failure than in success.

The same thing was true when intel was in the lead, it's was not something that would last forever. Intel could be ahead of TSMC in the future. Or maybe just be really competitive.

Time will tell.