Forget the White House Sideshow. Intel Must Decide What It Wants to Be.

[hist78]

Your career is tied to what is important amd it was x86. Unless the CEO and BOD believes than you are relegated to a career of irrelevance

Who decided that x86 would be Intel’s only path forward?

Isn’t Intel’s current crisis and difficulties clear evidence that Intel is on the wrong path? Intel is a product company and its mission should be to sell products that customers love and want. But if it goes the other way, with Intel demanding that customers must use x86-based products, it didn't work and it has proven to be a tragedy.

 

[DanX]

Who decided that x86 would be Intel’s only path forward?

Isn’t Intel’s current crisis and difficulties clear evidence that Intel is on the wrong path? Intel is a product company and its mission should be to sell products that customers love and want. But if it goes the other way, with Intel demanding that customers must use x86-based products, it didn't work and it has proven to be a tragedy.

“What happens if the customer wants a spoon, but you only have a fork?”

 

[user nl]

Here one can vote, let's see what the SemiWiki community votes?

Lessons for Intel From AMD’s Decisive GlobalFoundries Spin-Off: Could Going Fabless Solve Its Problems?

Intel is actually uncertain about whether it should spin out its foundry division, but there is a lesson from the past, coming from AMD.

wccftech.com

 

[Franco Cartolano]

Whether or not you spin off FAB, Intel has no future

I agree with you, Intel is attacked on every front, obviously by TSMC and Samsung on manufacturing, by Qualcomm and Huawei on the portable PC/tablets devices, by AMD on performance PC and servers. If NVIDIA launches its processors and the push for self-sufficiency silicon/Harmony OS continues in China, there won't be much space left for Intel. Apple has lead the way to a potentially X86-free world, which does not sound good for AMD either.

 

[tooLongInEDA]

Who decided that x86 would be Intel’s only path forward?

Isn’t Intel’s current crisis and difficulties clear evidence that Intel is on the wrong path? Intel is a product company and its mission should be to sell products that customers love and want. But if it goes the other way, with Intel demanding that customers must use x86-based products, it didn't work and it has proven to be a tragedy.

It is almost as though they grew up and prospered with the belief that the world revolved around Intel/x86 and have been too slow to adapt to the realisation that it actually revolves around something else. Much as people who believed that the universe revolved around the earth in the Middle Ages were reluctant to accept the distressing news that it actually did so around the sun.

I see the fundamental challenge here as more one of cultural adaptation than having the "wrong" [IDM] business model (while not discounting that as a factor).

BruceA's comment is also important. In large companies like Intel, you can only progress fast by joining the leading business unit (or attaching yourself to the leading customer account) - even if you were able to see the eventual headwinds, that still applies. Being aligned with perceived success usually matters far more than your direct personal contribution to any success.

 

[tomatoma]

I agree with you, Intel is attacked on every front, obviously by TSMC and Samsung on manufacturing, by Qualcomm and Huawei on the portable PC/tablets devices, by AMD on performance PC and servers. If NVIDIA launches its processors and the push for self-sufficiency silicon/Harmony OS continues in China, there won't be much space left for Intel. Apple has lead the way to a potentially X86-free world, which does not sound good for AMD either.

That's right, the rival of X86 is a rival for Intel, and a rival for AMD.
In a sense, Intel and AMD are based on mutual share and foundation.
Compete for market share there
If a disturbing person or a rival comes in there, both sides have to think about countermeasures.

By the way, I can have a favorable impression that you wrote x86 free without easily calling it ARM.
Consideration is given to the fact that RISC-V is also there.

 

[tomatoma]

That's right, the rival of X86 is a rival for Intel, and a rival for AMD.
In a sense, Intel and AMD are based on mutual share and foundation.
Compete for market share there
If a disturbing person or a rival comes in there, both sides have to think about countermeasures.

It is necessary to come up with some major countermeasures with x86 forces.
It doesn't matter if it's a mobile system, but if it's a computer system, it's pretty easy to do
x86 Starting x86 IP business or developing a better x86 microarchitecture...
Furthermore, it may be said that the method of opening the x86 will further increase the number of friends.

Another option is to adopt RISC-V and strengthen the ecosystem while developing it?
Converting x86 assets to RISC-V

 

[meo9725]

Guess what? The best people already left Intel, many of them my former colleagues. Intel is unable to attract top young talent which currently steers toward Nvidia, Apple, Amazon/Annapurna, Google, Microsoft, OpenAI and many other faster companies. Intel not one of them, Intel is wobbly, the opposite of a fast company. Unfortunately, at this point Intel is no longer fixable, certainly not by LBT who hasn't shown signs of true leadership so badly missing these days at Intel.

 

[DanX]

Guess what? The best people already left Intel, many of them my former colleagues. Intel is unable to attract top young talent which currently steers toward Nvidia, Apple, Amazon/Annapurna, Google, Microsoft, OpenAI and many other faster companies. Intel not one of them, Intel is wobbly, the opposite of a fast company. Unfortunately, at this point Intel is no longer fixable, certainly not by LBT who hasn't shown signs of true leadership so badly missing these days at Intel.

I disagree.
I have faith in LBT.
Top talents are overrated in fabs and design houses.
Plan and execution are the key factors. This is a characteristic of manufacturing.

 

[user nl]

Top talents are overrated in fabs and design houses.
Plan and execution are the key factors. This is a characteristic of manufacturing.

Perhaps some people like to disagree about the value of top talents. Read this story on one of TSMC's (and former INTEL's) top talents Wei-Jen Lo:

........................................
Lo’s team pushed ASML to reach an EUV source power of 250 watts—enabling a throughput of 125 wafers per hour, the threshold for commercial viability.

“At first, ASML might well have been stunned,” the veteran recalled. For years, the company’s EUV output power had stalled at 40 watts. The gap with TSMC’s target was so great that many saw little chance of success.

It wasn’t until one night in September 2014 that a breakthrough finally came: the pilot line at TSMC’s R&D headquarters hit 90 watts of output power for the first time.

Then-EUV program director Anthony Yen (嚴濤南) remembers Lo’s reaction: he immediately ordered 500 wafers a day to be produced on the EUV tool for an entire month, to prove it could sustain volume production.

The veteran noted that this was Lo’s signature strength: having once run Intel’s R&D fab, Lo could think through challenges from a manufacturing perspective as well. “In the global semiconductor industry, I doubt there’s anyone else with that capability,” he said.
..........................................

https://cwnewsroom.substack.com/p/two-pivotal-battles-that-cemented-tsmc-lead

 

[samwilde]

Intel started out making memory chips. They only went into microprocessors because a client asked for it. X86 is too important to ignore as a market but it did not need to be everything. Now the company has to make difficult choices, but even AMD still has the GPU guys.

IMHO they need to move all their wafers inside Intel and then we can discuss if they have enough scale to have their own fabs or not. I think they do if all the processors and GPUs are moved in.

 

[Paul2]

I don't doubt this if we use the correct setting but Microsoft Windows sucks big time. Lisa Su and Lip Bu tan should contact Satya and take some part of software about windows with themselves.

I can show you a 16 inch HX370 laptop with an OLED screen which consumes less than 2W on minimum brightness, and maximum powersaving settings (ASPM to max, every PCIE line except for SSD's powered down, USB ports power off, IRQs tuned, and all herded onto a single core, and all, but one core is explicitly powered down, all SoC peripherals except for display controller off, panel driven by EDP with PSR2, and the image is static)

And the last moment, is that it has to run under Linux...

 

[Fred Chen]

Perhaps some people like to disagree about the value of top talents. Read this story on one of TSMC's (and former INTEL's) top talents Wei-Jen Lo:

........................................
Lo’s team pushed ASML to reach an EUV source power of 250 watts—enabling a throughput of 125 wafers per hour, the threshold for commercial viability.

“At first, ASML might well have been stunned,” the veteran recalled. For years, the company’s EUV output power had stalled at 40 watts. The gap with TSMC’s target was so great that many saw little chance of success.

It wasn’t until one night in September 2014 that a breakthrough finally came: the pilot line at TSMC’s R&D headquarters hit 90 watts of output power for the first time.

Then-EUV program director Anthony Yen (嚴濤南) remembers Lo’s reaction: he immediately ordered 500 wafers a day to be produced on the EUV tool for an entire month, to prove it could sustain volume production.

The veteran noted that this was Lo’s signature strength: having once run Intel’s R&D fab, Lo could think through challenges from a manufacturing perspective as well. “In the global semiconductor industry, I doubt there’s anyone else with that capability,” he said.
..........................................

https://cwnewsroom.substack.com/p/two-pivotal-battles-that-cemented-tsmc-lead

Yeah, there's a big gap from R&D to manufacturing. Volume needs and breeds statistics. Let's not also forget N5 D0>1/cm2 (on average!) at risk production: https://web.archive.org/web/2020052...5nm-test-chip-yields-80-hvm-coming-in-h1-2020.

 

[Xebec]

It is almost as though they grew up and prospered with the belief that the world revolved around Intel/x86 and have been too slow to adapt to the realisation that it actually revolves around something else.

Unfortunately for Intel, their own experiences with trying to replace x86 reinforced this thinking:

iAPX 432 (early 80s)
i860 (late 80s/early 90s)
Itanium (~ 2000+)

And in the mid to late 1990s - x86 started to gradually dominate and kill off competing ISAs (both HPC and low power), teaching the (not always correct) lesson that software compatibility matters more than everything else. .

 

[Paul2]

lesson that software compatibility matters more than everything else

Only forward compatibility, not backward, and even that one is not given in X86 world. Executables with hardcoded modern SIMD instructions will not run on Pentium 4. A lot of obscure old instructions don't work as advertised on modern Intel CPUs. Kilometre sized erratas of previous CPUs also don't make the task of universal forward compatibility with Intel hardware a given. Microoptimisations guides change a lot with CPU generations.

Software superoptimised for NetBust was actually faster running on old Pentium 4s, than non-super-optimised code on newer, non-NetBust Pentiums.

 

[tomatoma]

Only forward compatibility, not backward, and even that one is not given in X86 world. Executables with hardcoded modern SIMD instructions will not run on Pentium 4. A lot of obscure old instructions don't work as advertised on modern Intel CPUs. Kilometre sized erratas of previous CPUs also don't make the task of universal forward compatibility with Intel hardware a given. Microoptimisations guides change a lot with CPU generations.

Software superoptimised for NetBust was actually faster running on old Pentium 4s, than non-super-optimised code on newer, non-NetBust Pentiums.

What are you talking about?
Do you understand the meaning of forward compatibility and backward compatibility?
It can be said that the X86 is also forward compatibility, but basically it is not possible to apply the features added in new generations to older products released earlier than that.
In the new generation, it is possible that the existing functions may be deleted, but I think it's possible to say that the x86 is generally backwards compatible.

Still, I think it's not wrong to say that it is compatible with both the front and the back.
The basic front refers to the old product being compatible with the new generation of products.
Backward compatibility refers to the new generation that includes or is compatible with the functionality of the old generation of products.

Software superoptimised for NetBust was actually faster running on old Pentium 4s, than non-super-optimised code on newer, non-NetBust Pentiums.

Of course, the code is highly optimized for netburst, so basically it only makes sense for NetBurst.
In a sense, it can't be applied to anything other than NetBurst, and when moving with a newer generation than NB, we can't make use of the new functions added in that new generation.

 

[Paul2]

It can be said that the X86 is also forward compatibility, but basically it is not possible to apply the features added in new generations to older products released earlier than that.

You can make use of SIMD instructions conditional on their support, and fall back to non-SIMD codepath

 

[tomatoma]

Unfortunately for Intel, their own experiences with trying to replace x86 reinforced this thinking:

iAPX 432 (early 80s)
i860 (late 80s/early 90s)
Itanium (~ 2000+)

And in the mid to late 1990s - x86 started to gradually dominate and kill off competing ISAs (both HPC and low power), teaching the (not always correct) lesson that software compatibility matters more than everything else. .

Yes, I agree, because I earned a lot of money with x86 in the first place and gained the advantage.
There is no stupid idea of throwing it away easily.
You don't have to abandon your favorable position and start from scratch yourself
Basically,
Choosing to throw away the existing foundation is stupid
I think it's not wrong for a company to make effective use of its advantages and its main products.

 

[Xebec]

Only forward compatibility, not backward, and even that one is not given in X86 world. Executables with hardcoded modern SIMD instructions will not run on Pentium 4. A lot of obscure old instructions don't work as advertised on modern Intel CPUs. Kilometre sized erratas of previous CPUs also don't make the task of universal forward compatibility with Intel hardware a given. Microoptimisations guides change a lot with CPU generations.

Software superoptimised for NetBust was actually faster running on old Pentium 4s, than non-super-optimised code on newer, non-NetBust Pentiums.

Fair - though this reticence to optimize code at all these days is partially why we're seeing small gains with new architectures and process nodes over time. Exactly at the time that we need software optimization because all low hanging fruits of CPU hardware are gone, we reverse into 'universal compatibility' code.

 

[BruceA]

Intel started out making memory chips. They only went into microprocessors because a client asked for it. X86 is too important to ignore as a market but it did not need to be everything. Now the company has to make difficult choices, but even AMD still has the GPU guys.

IMHO they need to move all their wafers inside Intel and then we can discuss if they have enough scale to have their own fabs or not. I think they do if all the processors and GPUs are moved in.

There is more to this Sam... remember Intel was a player in both DRAM and non-volatile EPROM and FLASH. Memory became a commodity and went from dozens of players to just the big cut throat.

x86 was a big decision to pivot the company. Oregon D1 used to do DRAM and moved them to logic while D2 in did FLASH and eventually logic as well.

Agree LBT should force all the silicon back inside! Why fund this at TSMC! Bring the revenue and fill up Ireland and Arizona, that is easily more than 600M units if you count the CPU, GPU and SOC chiplet dies. That gives the Intel Foundry a fighting chance.