News NVIDIA and Intel to Develop AI Infrastructure and Personal Computing Products

[jpiniero]

I think Intel and NVidia missed the boat on that, since the deal for official Xbox gear (using AMD) has been signed.

There's going to be other XBoxPC products too. The console is just a hedge/transition device.

 

[511]

There is still not late for handheld though

 

[Joe NYC]

There is still not late for handheld though

That would be just about the most brain dead move, for MSFT to go with RDNA 5 for full console and NVidia for handheld.

Being brain dead makes it more likely that Microsoft will do just that.

 

[marees]

I wonder how this is going to work. Will RTX graphics replace the Intel graphics chiplet? Or will they slap a dedicated GPU on the same package like the i7-8809G?

I am guessing like kaby lake g

 

[marees]

That would be just about the most brain dead move, for MSFT to go with RDNA 5 for full console and NVidia for handheld.

Being brain dead makes it more likely that Microsoft will do just that.

Nah. They are dead already. Killed by valve no more first party handhelds from Microsoft / surface teams

 

[Josh128]

If they sell today.

They almost certainly cant sell today. Likely a pre-determined lock out period.

 

[marees]

So now nv can design their own version of AT3 & AT4 chiplets ??

 

[Josh128]

This is mildly annoying as I owned around $7500 of INTC stock until about 2 months ago and sold it to more or less break even around $22 mark. That said, I fully expect the stock to return to the 20's as 5 billion will not magically make Intel have a viable process with good yields. Unless they attain positive EPS rather soon, this euphoria will likely fade unless more investments are announced in the next few months.

 

[Josh128]

So now nv can design their own version of AT3 & AT4 chiplets ??

You mean Medusa style GPUIOD chiplets? Its possible, but more likely probably that they license their GPU IP to Intel and Intel designs and produces the chiplets on 18A and 14A.

 

[jpiniero]

So now nv can design their own version of AT3 & AT4 chiplets ??

They are doing that with Mediatek's product, IIRC. Connected via NVLink.

 

[marees]

You mean Medusa style GPUIOD chiplets? Its possible, but more likely probably that they license their GPU IP to Intel and Intel designs and produces the chiplets on 18A and 14A.

I am skeptical of gpu tiles on Intel fabs

 

[Josh128]

I am guessing like kaby lake g

Im guessing some type of Strix Halo-esque AI marketed APU AI with "Nvidia Graphics". If Intel is actually replacing their own Core Ultra iGPUs with Nvidia, that would shock the hell out of me as Nvidia is a competitor in that space right now.

 

[Josh128]

I am skeptical of gpu tiles on Intel fabs

Im skeptical of everything on Intel fabs until they actually produce some compute tiles themselves that perform well instead of relying on TSMC.

 

[jpiniero]

Im guessing some type of Strix Halo-esque AI marketed APU AI with "Nvidia Graphics". If Intel is actually replacing their own Core Ultra iGPUs with Nvidia, that would shock the hell out of me as Nvidia is a competitor in that space right now.

It would let Intel cut IGP R&D back down to where they were before Arc.

 

[branch_suggestion]

I can't wait to see how this all goes south.
Remember Intel still holds a solid majority of DC and client CPU, NV holds a virtual monopoly of GPU in general.
EU/Asian regulators are not going to like this one iota.

Intel could've made a similar deal with AMD, they have all the necessary IP but I'm guessing they wanted nothing to do with Foundry or INTC stock.
And well AMD is the underdog, Intel wants share and stability, AMD would be the one getting share in such a deal.

Oh well, it is gonna be like 3 years until any products come out anyway, and there are a lot more players out there.

 

[Joe NYC]

I can't wait to see how this all goes south.
Remember Intel still holds a solid majority of DC and client CPU, NV holds a virtual monopoly of GPU in general.
EU/Asian regulators are not going to like this one iota.

Chinese regulators are already (saying they are) looking into NVidia violating the terms of China's approval of Melanox merger. That was one of the side bars when the stories of China banning NVidia GPUs emerged.

So this deal with Intel, creating a virtual monopoly (without NVidia officially announcing it is acquiring all of Intel) defying China's authority to prevent such a monopoly.

 

[MS_AT]

For desktop isn't this basically about integrating NVLink into Intel CPUs?

 

[Saylick]

Big news. Wow!

It will be interesting to see how this plays out in the long run, but the details of the collaboration say a few things up front:
1) This is purely a collaboration on products, not to fab Nvidia GPUs on IFS (although I’m guessing the final packaging may be at Intel). You take an Intel chiplet with integrated NVLink and join it with an Nvidia GPU with NVLink, not unlike GB10.

2) This gives Nvidia another avenue to push AI PC to the market outside of the late to market GB10. I’m guessing Jensen is not happy at how their WOA effort has stagnated.

3) These products will take years before they hit the market, and $5B is a paltry amount over the span of a few years. It’s not hard to imagine that the cash infusion is mainly to curry favor with the Trump administration given the administration’s now desire to prop up Intel (the US Government owning Intel shares sure does make the interests intertwined).

4) Will Intel give up on their own GPU ambitions? Does Nvidia think their GPU tech is so desirable that the consumer would rather choose an Intel CPU with an attached Nvidia GPU over an Intel CPU+GPU SoC? I don’t see massive volumes of this collaborative product, which leads me to feel like this deal is mostly performative.

 

[Antey]

This could be a stealth acquisition.

 

[511]

Packing will be 100% IFS regardless of Chip let Process remember TSMC doesn't do Chiplet for external fabs only Intel does.

 

[LightningZ71]

Effectively, China banning NVidia GPUs, or at least a portion of them, put a big chunk of their volume at risk. NVidia needed to still have access to that market as it's just about the only growth/volume market left that has optimism going forward.

I don't think that this will run afoul of any anti-trust overview. AMD has sufficient volume in DC for AI chips, server chips, desktop chips with gpu IP and GPUs themselves that they represent de-facto competition in most markets. This is win/win for Intel and Nvidia in that Intel needs a competitive edge vs. AMD in AI, and NVidia needs an all-up x86 solution as that still represents a massive share of the market. They can't get an x86 license of their own, so, this is the next best thing. It also keeps NVidia away from something that they are still institutionally weak at, building CPUs from the ground up. Yes, they have a few, but, aside from their tight integration with their GPUs, they aren't exactly leading the performance race while still being sufficient in the targeted environment.

I do wonder if this will expand into Nvidia making an investment in Intel's A14 process tech. It would be handy for Nvidia to make the chiplets at Intel if they are going to be integrated into Intel CPUs.

 

[Joe NYC]

For desktop isn't this basically about integrating NVLink into Intel CPUs?

How exactly would that help Intel to depend on foreign IP inside its CPU?

 

[511]

How exactly would that help Intel to depend on foreign IP inside its CPU?

This is why i think this is for Halo or High end products not cheap ass products

 

[branch_suggestion]

Does Nvidia think their GPU tech is so desirable that the consumer would rather choose an Intel CPU with an attached Nvidia GPU over an Intel CPU+GPU SoC?

Change Intel to AMD and you have your answer.
Honestly I would love to know if any consideration of this deal was based on projected AMD competitiveness or if this is purely realpolitik around short term political malarkey?
I mean all Intel gets for certain out of this deal is some packaging volume, the market decides the rest.

 

[Joe NYC]

This could be a stealth acquisition.

As I said above:

Page 2 - News - NVIDIA and Intel to Develop AI Infrastructure and Personal Computing Products

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com