Intel Nova Lake-S uses TSMC N2 process Tape-Out

[Xebec]

The plan was to move to outsourcing model closer to AMD. The product groups wanted this as well. Intel doesnt publicly comment on strategy plans under consideration .... But if you don't think I am correct, feel free to ask former Intel Execs/board members. Reminder I published notes on this and predicted the outcome between 2021 and 2023

Instead Intel went all in on foundry and Fabs with end goal of 18A being unquestioned leader. and the stock crashed.

If Intel did not have any foundry, and it had all its CCG, DCAI revenue. what would the stock price be? 80% More revenue than AMD..... Assume same product cost GM as AMD.

Thanks --

Two seperate topics;

For the stock price - I agree it would be a *little* better, but probably not signficantly as their products haven't been competitive. There could certainly have been a bump for the 'bold move', though some investors might have also seen that Intel had no path to better cost margins than AMD without the fabs. (Yes - counter argument on stronger scale.. but.. not inner circle with TSMC, etc.)

With or without fabs, there is a clear narrative that Intel has been consistently losing market share in high margin areas (workstation, server) for 5 years running now. Though they might have had enough cash to keep the dividend scheme a little longer..

...

For the fabs vs fabless I was just curious if anyone knows for sure what Swans plan was. Did Intel actually have an internal plan to stop fab development after a certain year or node (and what was that criteria). I know this is not likely ever to be public, but.. based on your comments I'm curious .

 

[siliconbruh999]

The plan was to move to outsourcing model closer to AMD. The product groups wanted this as well. Intel doesnt publicly comment on strategy plans under consideration .... But if you don't think I am correct, feel free to ask former Intel Execs/board members. Reminder I published notes on this and predicted the outcome between 2021 and 2023

Instead Intel went all in on foundry and Fabs with end goal of 18A being unquestioned leader. and the stock crashed.

If Intel did not have any foundry, and it had all its CCG, DCAI revenue. what would the stock price be? 80% More revenue than AMD..... Assume same product cost GM as AMD.\

the stock price would have been good but i can guarantee you the design will be same as now un-competitive their design needs to do better also regarding Borad they are the issue i have heard from ex-Intel people when Intel pulled back from Mobile it was due to pressure from the board also kranzich was let go after series of blackmailing between board and him why do you think they used the company policy to get rid of him and even than he was fired graciously for the mess he has created. Not to mention they wanted foundry to print only Intel chips when they had the best fabs in town
If Intel wants to start a new chapter they need to start a new they already have a new CEO find new products people and new Board no need to have Kranzich era BOD should just retire all of them

 

[MKWVentures]

the stock price would have been good but i can guarantee you the design will be same as now un-competitive their design needs to do better also regarding Borad they are the issue i have heard from ex-Intel people when Intel pulled back from Mobile it was due to pressure from the board also kranzich was let go after series of blackmailing between board and him why do you think they used the company policy to get rid of him and even than he was fired graciously for the mess he has created. Not to mention they wanted foundry to print only Intel chips when they had the best fabs in town
If Intel wants to start a new chapter they need to start a new they already have a new CEO find new products people and new Board no need to have Kranzich era BOD should just retire all of them

Agreed. What percentage of the current board is same as 2018?

 

[siliconbruh999]

Agreed. What percentage of the current board is same as 2018?

Frank D. Yeary(2009), Gregory D. Smith(2017) are the only 2 left out of 11 current board btw should get rid of few more that joined in 2019-20

 

[jimmymusk]

This is a strategic move by Intel, and here’s why it matters:

1. Dual-Process Strategy (18A + N2) Shows Pragmatism​

• Risk Mitigation: By taping out compute modules on TSMC N2, Intel ensures it can meet its late-2026 deadline even if 18A faces delays or yield issues.
• Capacity Hedge: TSMC’s N2 provides extra wafer supply if demand exceeds Intel’s internal 18A capacity (likely for a flagship SKU like this).

2. Heterogeneous Design = Performance Leap (But Challenges)​

• 52-core config (16P+32E+4LPE) suggests Intel is pushing core counts beyond AMD’s Zen 6 rumors.
• Xe3/Xe4 GPUs: Celestial (Xe3) for gaming + Druid (Xe4) for media could finally close the iGPU gap with RDNA 5.
• Testing Complexity: Powering on these chips will be grueling—expect months of validation for the 8,800MT/s memory controller and hybrid cores.

3. Timeline Realism​

• Q3 2026 launchseems achievable if:
  • TSMC N2 yields are stable by mid-2025 (N2P ramp expected then).
  • Intel’s Foveros packaging for the multi-tile design stays on track.

Open Questions:​

• Will Intel use N2 for the GPU tile and 18A for CPU? (Likely, given N2’s density advantage.)
• How will power efficiency compare to AMD’s Zen 6 on TSMC N2?

 

[jimmymusk]

Intel’s strategy of using both 18A and N2 processes is a smart hedge—just like how we approach AI-driven drug discovery at Ardigen.

✔ Dual-Path Validation

• Computational predictions (fast but noisy)
• Lab experiments (slow but definitive)
  → Cuts validation time by 50%

✔ Heterogeneous AI Models
Like Intel’s hybrid cores, we combine:

• Graph NNs (target discovery)
• Transformers (toxicity screening)
  → Better accuracy than single-model systems

Key Lesson:
Whether in chips or bioinformatics, redundancy + specialization = faster breakthroughs.

(For teams balancing speed/accuracy: We built a free AI workflow checker – might help!)

 

[Xebec]

Open Questions:

• Will Intel use N2 for the GPU tile and 18A for CPU? (Likely, given N2’s density advantage.)
• How will power efficiency compare to AMD’s Zen 6 on TSMC N2?

I suspect Intel will use an older node (N3/N4) for the GPU tile based on their choices with Arrow, Lunar, and Meteor lake, and also for capacity / cost reasons. They could probably use Intel 3 also but uneducated guess I'd think they would stick with TSMC.

Re Power efficiency - I think for multi threaded loads the e cores are winners..

 

[jimmymusk]

I suspect Intel will use an older node (N3/N4) for the GPU tile based on their choices with Arrow, Lunar, and Meteor lake, and also for capacity / cost reasons. They could probably use Intel 3 also but uneducated guess I'd think they would stick with TSMC.

Re Power efficiency - I think for multi threaded loads the e cores are winners..

Yeah man, I think you’re spot on. Intel’s probably gonna stick with TSMC for the GPU tile—makes sense if you look at what they’ve been doing with Arrow, Lunar, and Meteor Lake. Cost and capacity matter big time, and TSMC’s been a safe bet for them so far. Intel 3 might happen down the line, but honestly, I’d be surprised if they don’t just keep rolling with what already works.


And yup, couldn’t agree more about the E-cores. For multi-threaded stuff, they’ve become real workhorses. Power-efficient and solid performance—Intel actually did a good job there.

I suspect Intel will use an older node (N3/N4) for the GPU tile based on their choices with Arrow, Lunar, and Meteor lake, and also for capacity / cost reasons. They could probably use Intel 3 also but uneducated guess I'd think they would stick with TSMC.

Re Power efficiency - I think for multi threaded loads the e cores are winners..

 

[siliconbruh999]

I suspect Intel will use an older node (N3/N4) for the GPU tile based on their choices with Arrow, Lunar, and Meteor lake, and also for capacity / cost reasons. They could probably use Intel 3 also but uneducated guess I'd think they would stick with TSMC.

Re Power efficiency - I think for multi threaded loads the e cores are winners..

They are likely reusing the GPU tile from Panther on Intel 3 or graphics tile on 18AP