[Motley Fool] Apple dodged an Intel-shaped bullet

[swilli89]

I think Intel rapidly going from Tick - Tock to "Tick - Tock - Optimization - Optimization ++" is all the evidence anyone needs to prove they are having serious delays on the foundry process.

Maybe its possible their processes have seemed so good is because they are also able to spend Billions in chip r&d custom making their logic for the process (which is probably obvious).

 

[oak8292]

Here ya go...

https://newsroom.intel.com/newsroom...con-mesa-intel-tmg-china-event-fact-sheet.pdf

This is a very important qualifier;

'Intel will disclose more details and a timeline for product availability at a later date.'

The Stratix 10 from Altera was first announced in October 2013 and first silicon was October of 2016. Altera thought they were going to get at least a one year jump on Xilinx with 14nm FinFET. However, Xilinx had 16nm silicon from TSMC a year earlier than Altera. Now that Intel has finally released the Stratix 10 this next FPGA should not take as long but it is important to realize that the date of announcement has nothing to do with silicon availability. If Xilinx gets it's 7nm FPGA from TSMC out first in 2018 it will be a huge black eye on Intel. Xilinx is skipping the TSMC 10nm because they can't afford to re-engineer a low volume product that frequently (my assumption).

 

[IntelUser2000]

Maybe its possible their processes have seemed so good is because they are also able to spend Billions in chip r&d custom making their logic for the process (which is probably obvious).

I agree with this as well. Without the tight working teams of manufacturing and chip design, they would have quickly lost any advantage, even during the days when they actually had one. They were at its peak, a top-tier IDM. This is where the argument that if Intel were split into Foundry, and CPU design, they would turn out better. No, they'd do even worse than today.

 

[StrangerGuy]

I think I said this before, but Intel doesn't understand their fundamental business model. They are not in the business of pushing process node tech, they are in the business of selling their chips. The process node is just a means to that end, their customers doesn't care what Intel chips are made on as long as the chips performs well at the end of the day. All that efforts for an ultimately meaningless title is ironically subsidizing the R&D of Samsung and TSMC through tool vendors like ASML.

 

[krumme]

I think I said this before, but Intel doesn't understand their fundamental business model. They are not in the business of pushing process node tech, they are in the business of selling their chips. The process node is just a means to that end, their customers doesn't care what Intel chips are made on as long as the chips performs well at the end of the day. All that efforts for an ultimately meaningless title is ironically subsidizing the R&D of Samsung and TSMC through tool vendors like ASML.

Good point about funding samsung and tsmc indirectly!
I have said Intel was focusing their business wrong for years and taken more flak in this forum than a rusty flying tincan in bf1.
To me it seems there have been a cultural powerfight inside the company. A sign is eg Pat Gelsinger beeing sidelined by current CEO. The process and manufacturing people having the most power during most of the last decade.

A result of the Bod looking at where the money goes. Going for profit on short and medium term. Thinks who can make a difference for that and then vote for fab man. Thats how ceo is selected.

Ofcource hugely idiotic for anyone with just slight product insight and passion but its the norm.

We need people like steve jobs, jhh and heck to a certain degree even Lisa. They got business drive, passion and product knowledge. Not those mechanics. They are very much needed but not as ceo of a technology firm.

When ibm high freq 7nm hits zen 2 we will see what mess Intel is in with its blown out fixed cost. Its a more risky business than most know when margins get under pressure.

 

[maddie]

There was a time when pushing node advancement made the most sense for ROI vs architectural improvements. Physics being what it is means that the pendulum has swung to other aspects of the product as node advancements have hit harder barriers. The present problem with Intel, and this happens to all at some time, is that there is an inertia slowing change due to very human factors such as who gains power and who loses power in the company. In another word, politics.

 

[oak8292]

I think I said this before, but Intel doesn't understand their fundamental business model. They are not in the business of pushing process node tech, they are in the business of selling their chips. The process node is just a means to that end, their customers doesn't care what Intel chips are made on as long as the chips performs well at the end of the day. All that efforts for an ultimately meaningless title is ironically subsidizing the R&D of Samsung and TSMC through tool vendors like ASML.

Maddie is right that there is a shift from node to architecture as node shrinks become increasingly complex with multi-patterning. Nodes and the lowest cost transistor has always been the game at Intel for the last 30 years. Architecture was always secondary.

Talk by Robert Colwell a retired Intel architect at Hot Chips;

• Chip architect’s job was to remove obstacles from Moore’s Law silicon bounty
  
  – System perf = f(CPU clock, CPU uArch, sys)
  – From 1980 to 2010, clocks are 3,500x faster
  – What did Arch/uArch achieve beyond that? Maybe 50x?
  

http://www.hotchips.org/wp-content/...190-Keynote1-ChipDesignGame-Colwell-DARPA.pdf

The loss of Dennard's scaling started tick tock to focus every other year on architecture. Now the slowing of Moore's Law isn't quite existential but it requires another big shift.

P.S. Both Samsung and TSMC also provided capital to ASML for the development of EUV. TSMC has already cashed out their shares in ASML.

 

[IntelUser2000]

I think I said this before, but Intel doesn't understand their fundamental business model. They are not in the business of pushing process node tech, they are in the business of selling their chips.

It was commonly said many years ago that AMD would have a superior design, but Intel would edge it out with a newer and better process.

They basically pioneered scaling. It'd take a momentous amount of effort to align themselves to be a company that starts to focus on design. It just seems nowadays everyone is ahead of Intel in design.

 

[swilli89]

To be fair Intel still is able to clock past 5ghz with similar to slightly worse IPC than that of Ryzen processors. They will be in serious trouble if Ryzen hits that 5+ghz barrier with 7nm..

 

[krumme]

When apple and samsung introduces a new processor it will have 3 years with a nice depreciation profile. From brand new highend phone to midrange. Far better profit day one because people is obsessed with getting the new mobile stuff. But also after 3 years when it can still be used in midrange or as cheaper former top end model. While Intel will have to resort to using it in chipset whatnot or even selling the equipment at a huge loss.
It just flat out doesnt make sense these days to fund your process development from pc processors. Not sharing the cost with mobile.
Its a testament to how professional and experienced Intel is on this side they can still rage in those huge profit.
But its a blunder. And even if 7nm and 5nm is technically viable it looks like it does not economically make sense to go there even at the same time as tsmc and the others that is in a different ecosystem. It looks like a hard wall in practical sense.

Intel now takes a lot of the usual flak for beeing late. I mean its nonsense. There might be some true to Intel beeing mixed up in their own -Moores law runs eternally- history. Its so sad to see it still promoted by them. They have to let it go and stop lying to themselves and the surroundings. The seniors that build the prior processes is stil the same and still as competent and they do it just as well.