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

Page 12 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Saylick

Diamond Member
Sep 10, 2012
4,018
9,432
136
GB10, for reference:
1758423135154.png

It's not hard to imagine an Intel/Nvidia version where the CPUs are x86 Intel cores.

For server, just imagine Grace-Blackwell except instead of Grace, it will be a custom Intel package with all the usual Intel IP except with an added Nvidia NVL C2C PHY.

These collaborative parts aren't hard to understand.
 

adroc_thurston

Diamond Member
Jul 2, 2023
6,801
9,526
106
It's not hard to imagine an Intel/Nvidia version where the CPUs are x86 Intel cores.

For server, just imagine Grace-Blackwell except instead of Grace, it will be a custom Intel package with all the usual Intel IP except with an added Nvidia NVL C2C PHY.
yes but where's the evil schizo masterplan where Jensen laughs and makes Intel into slaves.
These collaborative parts aren't hard to understand.
Yup, it's the most basic anti-AMD play ever.
 

Saylick

Diamond Member
Sep 10, 2012
4,018
9,432
136
yes but where's the evil schizo masterplan where Jensen laughs and makes Intel into slaves.
Yeah, I can see JHH use Nvidia's huge leverage on the industry to pull Intel where he wants them to go. Essentially, telling them to play ball or lose even more x86 market share to AMD in the AI datacenter space. Have Intel fab their collaboration CPUs in-house while Nvidia pays a good price for each since they know IFS has to be cheaper than TSMC.
Yup, it's the most basic anti-AMD play ever.
Sounds more and more like it.
 

adroc_thurston

Diamond Member
Jul 2, 2023
6,801
9,526
106
Yeah, I can see JHH use Nvidia's huge leverage on the industry to pull Intel where he wants them to go. Essentially, telling them to play ball or lose even more x86 market share to AMD in the AI datacenter space
True but he also really needs x86.
It's a mutual-interests alliance, really.
Have Intel fab their collaboration CPUs in-house while Nvidia pays a good price for each since they know IFS has to be cheaper than TSMC.
NV makes GPUs are Intel nodes so far are not good options for GPUs.
And CPUs are low margin adds for rackscale GPU solutions.
Sounds more and more like it.
Real competition hours indeed.
 

Darkmont

Member
Jul 7, 2023
83
272
106
What do you mean by Kitchen Sink the only thing I care is wether Zen 6 and NVL will be neck to neck with each other.
Kitchen sink. They’re throwing as much money as they can into the damn thing.

- Leading edge node with tons of cell choices to switch between for maximum zoomzoom.
- Giga SRAM pile, not even stacked they just blew up the die.
- Dual CT to spam moar of everything for nT leadership.
- Increased PL2.
- APO+ to gank developer EXEs and swap them with optimized™ ones.

It’s a tour de force in pulling every lever you possibly can to improve bar charts at the cost of everything else. Very in line with the Intel philosophy.
 

Joe NYC

Diamond Member
Jun 26, 2021
3,541
5,112
136
WoA will become irrelevant as soon as Nova Lake comes out. This Nvidia deal with Intel, killed it even more

What makes you think there will be a NVidia chiplet for Nova Lake?

Nova Lake may be close to (or already did) tape out and the NVidia Chiplet just showed up on a Power Point less than a week ago.
 

branch_suggestion

Senior member
Aug 4, 2023
810
1,755
106
Kitchen sink. They’re throwing as much money as they can into the damn thing.

- Leading edge node with tons of cell choices to switch between for maximum zoomzoom.
- Giga SRAM pile, not even stacked they just blew up the die.
- Dual CT to spam moar of everything for nT leadership.
- Increased PL2.
- APO+ to gank developer EXEs and swap them with optimized™ ones.

It’s a tour de force in pulling every lever you possibly can to improve bar charts at the cost of everything else. Very in line with the Intel philosophy.
All that to get outsold in desktop 10 to 1, well done Intel.
Outside of using the best node, none of this helps in laptop which is the market they are not completely dead in.
They need to conjure another RPL-282 for that and just dump them in OEM warehouses.
 
  • Like
Reactions: Tlh97 and Joe NYC

511

Diamond Member
Jul 12, 2024
4,294
3,938
106
All that to get outsold in desktop 10 to 1, well done Intel.
Outside of using the best node, none of this helps in laptop which is the market they are not completely dead in.
They need to conjure another RPL-282 for that and just dump them in OEM warehouses.
NVL is not out yet so how can you forsee that?
 

511

Diamond Member
Jul 12, 2024
4,294
3,938
106
Leading edge node with tons of cell choices to switch between for maximum zoomzoom.
Everyone is doing this tbh in 2026 so it's normal
- Giga SRAM pile, not even stacked they just blew up the die.
This sucks with the fact that their Fabric/LLC team had been screwing around.
- Dual CT to spam moar of everything for nT leadership.
- Increased PL2.
Kind of expected
- APO+ to gank developer EXEs and swap them with optimized™ ones.
I thought this was not true but if it's that's crazy.
 

Saylick

Diamond Member
Sep 10, 2012
4,018
9,432
136
What makes you think there will be a NVidia chiplet for Nova Lake?

Nova Lake may be close to (or already did) tape out and the NVidia Chiplet just showed up on a Power Point less than a week ago.
Nova Lake presumably has good enough perf/W that it makes WoA irrelevant. The Intel/Nvidia consumer part won't be out until 2028/2029 at best, but when it does GB10 or its derivatives will be even more irrelevant.
 
  • Like
Reactions: Tlh97 and Joe NYC

Joe NYC

Diamond Member
Jun 26, 2021
3,541
5,112
136
Nova Lake presumably has good enough perf/W that it makes WoA irrelevant. The Intel/Nvidia consumer part won't be out until 2028/2029 at best, but when it does GB10 or its derivatives will be even more irrelevant.

I am struggling to see how this makes sense. GB10 is a niche product that does not even officially support Windows. A MiniPC that's a DGX box lookalike.

WoA, from NVidia / MediaTek will be a laptop. so it will not be competing with Nova Lake, which will be a desktop chip.
 
  • Like
Reactions: Tlh97

Joe NYC

Diamond Member
Jun 26, 2021
3,541
5,112
136
GB10 and N1x are the same silicon.

Seems like they are still tweaking it.

the what.
NVL's an entire stack.

Is that really the case? That would be a quite a setback for Intel foundry if TSMC based Nova Lake were to be a replacement for Intel foundry based Panther Lake.

I think, more likely, there will be some overlap in higher wattage SKUs, but lower power SKUs will continue to be Panther Lake, they will not be replaced with Nova Lake.
 

adroc_thurston

Diamond Member
Jul 2, 2023
6,801
9,526
106
Seems like they are still tweaking it.
less 'tweaking' and more like it doesn't work.
Is that really the case?
yea
That would be a quite a setback for Intel foundry if TSMC based Nova Lake were to be a replacement for Intel foundry based Panther Lake.
you do understand that only a fraction of the NVL lineup is at TSM?
I think, more likely, there will be some overlap in higher wattage SKUs, but lower power SKUs will continue to be Panther Lake, they will not be replaced with Nova Lake.
You don't need to 'think' when roadmaps are a thing.
 

Joe NYC

Diamond Member
Jun 26, 2021
3,541
5,112
136
less 'tweaking' and more like it doesn't work.

Is that the Windows part on it that is not working or the whole thing?

you do understand that only a fraction of the NVL lineup is at TSM?

All but one (the smallest) compute dies are TSNC N2: Budget (4+8), Mainstream to Premium (8+16).
Only 1 die, Entry level 4+0 is die is apparently on 18A. Who knows if that single die even survives, unlike Arrow Lake, where also only the lowest end die was to be 20A, and it was scrapped.

As opposed to Panther Lake where CPU dies are Intel 18A
 

511

Diamond Member
Jul 12, 2024
4,294
3,938
106
All but one (the smallest) compute dies are TSNC N2: Budget (4+8), Mainstream to Premium (8+16).
Only 1 die, Entry level 4+0 is die is apparently on 18A. Who knows if that single die even survives, unlike Arrow Lake, where also only the lowest end die was to be 20A, and it was scrapped.

As opposed to Panther Lake where CPU dies are Intel 18A
The SoC die is 18A the iGPU is 18AP as for cancellation of 4+0 unlikely cause it succeds 4+0+4 PTL SKUs.
 
  • Like
Reactions: Joe NYC