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

[branch_suggestion]

Yeah NV went from "ARM is all you need, we're gonna literally buy the goddamn company" to this.
whatever that is (total complete surrender to x86?).

ARM seems to now be aligned against NV, which is absolutely hilarious.
Then again, they are kinda aligning against nearly everyone now unless you subscribe to their new paradigm and avoid the wrath of merchant ARM.

 

[adroc_thurston]

ARM seems to now be aligned against NV, which is absolutely hilarious.

yeah but NV has a custom core roadmap and a good license after the acq got severed.
They've spent years committing to ARM.
now this? what.

 

[Josh128]

roadmap changes don't happen overnight.

This is not a roadmap change, its simply a new GPU product application.

 

[Heartbreaker]

So is this "investment" or "takeover"? I'm not really sure what to think about this...

Prelude to takeover? If Intel become dependent on NVidia GPU and AI parts, Intel will be in even more trouble if the partnership ever ends.

 

[Josh128]

Now AMD must fight the three-headed hydra of US federal government and Intel and Nvidia.

Maybe next quarter Intel will find a way to expand the competition into the four horsemen.

You know they are talking with Tim Cook.

 

[adroc_thurston]

This is not a roadmap change

yeah it is, x86 server SOCs with NVL-C2C means a whole lot of things and none of them are good in context of the server Tegra roadmap.

 

[MS_AT]

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

You are one of the cool kids on the block that can NVLink with nVidia chip. What if nVidia says that it due to whatever it's NVLink or nothing? Then Intel is the only x86 vendor that can pair with nVidia. It basically cuts AMD from the market. Seems improbable but who knows. Unless AMD makes similar kind of deal. But then what will happen to its own GPU efforts.

 

[adroc_thurston]

Unless AMD makes similar kind of deal

why would they.

But then what will happen to its own GPU efforts.

they just gonna win.

It basically cuts AMD from the market

AMD never had much share in GPGPU box attach.
This is more relevant in context of server Tegra.

 

[Josh128]

yeah it is, x86 server SOCs with NVL-C2C means a whole lot of things and none of them are good in context of the server Tegra roadmap.

Thats going to be Intels SOCs, not Nvidias. Intel is probably just going to license NVs graphics IP and slap a "Powered by Nvidia AI" on it. Nvidia is not making and selling x86 chips, from anything I've read.

Its going to be a repeat of 2025 Strix Halo, but with Intel and Nvidia. Also probably some datacenter mega APUs like the 300A series Zen 4/CDNA 3 APUs.

 

[branch_suggestion]

yeah but NV has a custom core roadmap and a good license after the acq got severed.
They've spent years committing to ARM.
now this? what.

I do think NV are willing to bet on every horse in the race, but will they get away with winning anything back?
They are spreading themselves all over the place to see if anything sticks, like they are in a panic even though they are supremely dominant right now.
This isn't Grove style paranoid, this is just straight schizo.

 

[adroc_thurston]

Thats going to be Intels SOCs

Again, DC.

Intel is probably just going to license NVs graphics IP and slap a "Powered by Nvidia AI" on it

data center.

Nvidia is not making and selling x86 chips, from anything I've read.

they effectively would make x86 the default choice for GPGPU ecosystem again.
why?

 

[adroc_thurston]

I do think NV are willing to bet on every horse in the race, but will they get away with winning anything back?

I mean they've spent years moving CUDA to decent success to aa64, only to basically backtrack all of that and let x86 be the default choice again?
like, what.

 

[511]

I mean they've spent years moving CUDA to decent success to aa64, only to basically backtrack all of that and let x86 be the default choice again?
like, what.

Maybe they know something we don't know

 

[adroc_thurston]

Maybe they know something we don't know

I know Tegra roadmap is stuck in a never-ending struggle session, but I never ever expected an outright surrender.

 

[Josh128]

Again, DC.

data center.

they effectively would make x86 the default choice for GPGPU ecosystem again.
why?

Not really. Allowing an x86 option that they can offer their datacenter GPUs on doesnt kill the established ARM ecosystem whatsoever. It is a way to try and stifle AMDs x86 datacenter progress though, with the added benefit of whichever CPU tech wins out (or even if they just remain two different segments), Nvidia is a big part of it.

 

[MoistOintment]

Nvidia is not making and selling x86 chips, from anything I've read.

Tomshardware's article says they are:

Intel will fabricate custom x86 data center CPUs for Nvidia, which Nvidia will then sell as its own products to enterprise and data center customers. However, the entirety and extent of the modification are currently unknown.

 

[adroc_thurston]

Allowing an x86 option that they can offer their datacenter GPUs on doesnt kill the established ARM ecosystem whatsoever

YES IT DOES.
why choose aa64 mysterymeat when x86 is just available again?

It is a way to try and stifle AMDs x86 datacenter progress though

obviously, AMD's the 666 pound gorilla en route to kill everyone.
But again, NV spent years cultivating aa64 for CUDA. Now this will be poof.

 

[Josh128]

I mean they've spent years moving CUDA to decent success to aa64, only to basically backtrack all of that and let x86 be the default choice again?
like, what.

That is precisely why that this is not whats happening. You are reading way too much into this "joint venture", which is simply a highly visible way to appease the Trump administration with the added benefit of a potential option outside of TSMC for fabrication.

 

[511]

YES IT DOES.
why choose aa64 mysterymeat when x86 is just available again?

And at discounted rate

 

[adroc_thurston]

That is precisely why that this is not whats happening.

it *is* what's happening, you'll get Xeons with NVL-C2C for NV GPUs.

And at discounted rate

with full NVL-C2C integration too.

 

[Josh128]

YES IT DOES.
why choose aa64 mysterymeat when x86 is just available again?

obviously, AMD's the 666 pound gorilla en route to kill everyone.
But again, NV spent years cultivating aa64 for CUDA. Now this will be poof.

Because its not mysterymeat, its the defacto AI datacenter tech, and will likely remain that way.

 

[adroc_thurston]

Because its not mysterymeat

From a developer POV, it is.

its the defacto AI datacenter tech, and will likely remain that way.

x86 is the de-facto DC tech and it's about to become the default for GPGPU again.
There is no escape. No one's ever truly free.

 

[jpiniero]

YES IT DOES.
why choose aa64 mysterymeat when x86 is just available again?

nVidia has been offering x86 as an option.

Really, this is most likely just Donald Trump forcing a Pump & Dump... and no products will ever end up coming out.

 

[Josh128]

it *is* what's happening, you'll get Xeons with NVL-C2C for NV GPUs.

with full NVL-C2C integration too.

Then explain why they are investing a paltry 5B into Intel instead of 100B+. Riddle me that.

 

[adroc_thurston]

nVidia has been offering x86 as an option.

WRONG.
Not for parts that matter (rackscale stuff).

Then explain why they are investing a paltry 5B into Intel instead of 100B+. Riddle me that.

you don't need 100B dollar to hijack an existing roadmap.