Thread regarding Intel Corp. layoffs

There was a slide that said that Intel 18a is only up to 15% perf/watt than Intel 3. If that's not a typo, 18a is DOA.

@zol — I guess it’s even worth mentioning WHY. Despite this being BK’s one decision that destroyed the company, he was fundamentally right about one thing. A mature SAQP process is a he-l of a lot cheaper to run than an EUV process

@zol — all three intermediate processes are unlikely to even run long enough to reach the yield maturity foundry customers crave

@zol — you are correct both that foundry customers prefer nodes at a mature state of the yield ramp and that the general IFS plan is to use foundry customers to run out node depreciation rather than Intel flagships. Unfortunately these truths do not add up to what you think they do. There is no plan to pick up foundry customers for the three intermediate nodes — ever. They’re all going to 18A and Intel 7.

Foundry customers prefer mature nodes. The existence of intermediate nodes is to attract foundry customers and to finance future nodes even if they only run intel flagship products for a short while. TSMC makes its money on mature nodes and uses them to finance future nodes which lose money until they are mature enough for the more risk adverse customers.

It's not 5 nodes, it's 5 products, every product slightly changes the process.
Problem is time, Intel is too slow to publish 5 products in 4 years

5N4Y is a total joke in the industry.
Everyone with a clue knows it’s just marketing nonsense.
PPAT (power, perf, area, thermals$ and COST is all that matters to customers.

Do you see any customer?
Well, there’s your answer on how effective those nodes are.

yes, real distinct nodes require major ROI and retooling. However, marketing names are far cheaper.

@yqb+1 it really seems if we were at least TRYING we could build at least one Lunar Lake tile on one of those things. Maybe IO/SOC

@xac+1 — it’s a very relevant question but I doubt anyone who knows the answer and values their job will ever tell you

Let's be very clear here.
Intel 7 = Intel 1276 = Intel 10+++, also referred one time to 10#
Intel 4 and Intel 3 = 1276 with a single vs. 2 libraries
Intel 20A and 18A = 1278, single vs. 2 libraries

1274 10++ was already in HVM when Pat came over; 10+++ or Intel 7 is not a major node at all. So, it boils down to P1276, single node with 2 variants
1278 is far from being ready

How many major customers has Intel actually reeled in for 18A? Genuine question.