Intel - "PC sales are softening" / AMD - PC CPU sales rise 33% YoY

[blueone]

Here's what Abhinav Davuluri is saying to justify a bullish position ($65 FVE) on Intel stock
- Secular PC decline and relative lack of competition from AMD contributed to Intel delaying its 10-nm process ramp
- Intel had less incentive to rush out its new 10-nm process, as new processes are typically margin-dilutive while ramping
- Intel 4 with EUV will greatly simplify its manufacturing process flow
- Intel has an opportunity to regain manufacturing leadership if it can better deploy GAA
- Intel is the TSMC alternative for many products to avoid concentration risk
- Fab cost $20B plus creates significant barriers to entry
- Intel's only competition for leading-edge manufacturing are TSMC and Samsung
- Only Intel and AMD have x86 architecture providing differentiation
- ARM-based datacenter chips have had little success in displacing x86
- ARM-DC chips are used in less intensive workloads, a small niche application
- Revenue decline is due to divesting NAND business to SKHynix

...relative lack of competition from AMD contributed to Intel delaying its 10-nm process ramp

Seriously? If only that were true, and it isn't.

x86 architecture is not a differentiation for the fast-growing cloud computing providers (GOOG, AMZN, MSFT, ORCL, etc.). These companies own their software stacks down to the OS and even below. x86 is only a differentiator for enterprise servers and Windows PCs, in which Windows and Linux and applications are written and tuned for x86. AMD has made significant inroads into single and dual socket servers. Of course, AMD does not compete in the four socket and larger market.

Do you believe what Davuluri wrote?

 

[hist78]

Here's what Abhinav Davuluri is saying to justify a bullish position ($65 FVE) on Intel stock
- Secular PC decline and relative lack of competition from AMD contributed to Intel delaying its 10-nm process ramp
- Intel had less incentive to rush out its new 10-nm process, as new processes are typically margin-dilutive while ramping
- Intel 4 with EUV will greatly simplify its manufacturing process flow
- Intel has an opportunity to regain manufacturing leadership if it can better deploy GAA
- Intel is the TSMC alternative for many products to avoid concentration risk
- Fab cost $20B plus creates significant barriers to entry
- Intel's only competition for leading-edge manufacturing are TSMC and Samsung
- Only Intel and AMD have x86 architecture providing differentiation
- ARM-based datacenter chips have had little success in displacing x86
- ARM-DC chips are used in less intensive workloads, a small niche application
- Revenue decline is due to divesting NAND business to SKHynix

"Revenue decline is due to divesting NAND business to SKHynix"

The -1% Non-GAAP revenue decline has already excluded the impact from the sale of NAND memory division to SK Hynix. Although Intel did not mention the amount of revenue decline can be attributed to the divested NAND division.

 

[benb]

Seriously? If only that were true, and it isn't.

x86 architecture is not a differentiation for the fast-growing cloud computing providers (GOOG, AMZN, MSFT, ORCL, etc.). These companies own their software stacks down to the OS and even below. x86 is only a differentiator for enterprise servers and Windows PCs, in which Windows and Linux and applications are written and tuned for x86. AMD has made significant inroads into single and dual socket servers. Of course, AMD does not compete in the four socket and larger market.

Do you believe what Davuluri wrote?

I interjected Davuluri analysis because the analyst perspective on semiconductor stocks is often so different from insiders. I find myself triangulating (or multiangulating) with issues I'm close to, like semiconductors. There are elements of team sports fanship which I try to avoid.

What lands for me is the view that x86 is differentiated. It's not necessarily the highest end premium option, that would be IBM/Oracle, but there seems to be a careful evolution of x86, generation after generation, to address 80% of the compute cycles out there, and optimize the performance/value curve in a way that makes x86 attractive.

When the Apple M1 chip came out, there was an observation that the performance/value curve for Apple cores had crossed over x86, an epochal achievement. Apple could now provide more compute for the dollar, a basic competitive advantage. That is a real differentiation, and in due time we will see Apple's market penetration grow, as a consequence.

ARM cores are not crossing over x86 though, AFAIK. That "no change" is boring but it has consequences just as Apple core cross-over has consequences.

 

[Portland]

Intel has a lot of problems, but Intel also has a lot of great engineers, but even great engineers can be backed into a corner by a strategy that doesn't play out the way it's supposed to.

When Marvell and mellanox designed chiplets it was the beginning of the end of many engineers.

 

[Daniel Nenni]

Here's what Abhinav Davuluri is saying to justify a bullish position ($65 FVE) on Intel stock
- Secular PC decline and relative lack of competition from AMD contributed to Intel delaying its 10-nm process ramp
- Intel had less incentive to rush out its new 10-nm process, as new processes are typically margin-dilutive while ramping
- Intel 4 with EUV will greatly simplify its manufacturing process flow
- Intel has an opportunity to regain manufacturing leadership if it can better deploy GAA
- Intel is the TSMC alternative for many products to avoid concentration risk
- Fab cost $20B plus creates significant barriers to entry
- Intel's only competition for leading-edge manufacturing are TSMC and Samsung
- Only Intel and AMD have x86 architecture providing differentiation
- ARM-based datacenter chips have had little success in displacing x86
- ARM-DC chips are used in less intensive workloads, a small niche application
- Revenue decline is due to divesting NAND business to SKHynix

Here are the ones I did not laugh at:

- Intel has an opportunity to regain manufacturing leadership if it can better deploy GAA
- Fab cost $20B plus creates significant barriers to entry

Funny ones:

- Intel is the TSMC alternative for many products to avoid concentration risk.
Intel chose TSMC as an alternative for many products to avoid concentration risk.

- Intel's only competition for leading-edge manufacturing are TSMC and Samsung
Being the 3rd horse in a three horse race means you are in last place.

- ARM-based datacenter chips have had little success in displacing x86
Yet Google, Amazon, Alibaba, etc... are using ARM for their own datacenter chips.

Abhinav Davuluri spent two years at Intel so he SHOULD know what he is talking about. What is his agenda here?

 

[benb]

Here are the ones I did not laugh at:

- Intel has an opportunity to regain manufacturing leadership if it can better deploy GAA
- Fab cost $20B plus creates significant barriers to entry

Funny ones:

- Intel is the TSMC alternative for many products to avoid concentration risk.
Intel chose TSMC as an alternative for many products to avoid concentration risk.

- Intel's only competition for leading-edge manufacturing are TSMC and Samsung
Being the 3rd horse in a three horse race means you are in last place.

- ARM-based datacenter chips have had little success in displacing x86
Yet Google, Amazon, Alibaba, etc... are using ARM for their own datacenter chips.

Abhinav Davuluri spent two years at Intel so he SHOULD know what he is talking about. What is his agenda here?

I guess I'm convinced by Dan and Blueone, x86 on balance is a legacy cost not a differentiation.

Intel is a fab owner, they can raise prices while their fab cost is fixed, same as TSMC and probably Samsung is doing. Inflation/shortage is good for fixed assets.

On the downside, Intel had a tailwind from COVID lockdowns leading to unprecedented PC demand out of the blue. The disappearance of that demand may explain the mystery of the -1% sales growth (since it's not NAND); the YoY comparison may be unusually difficult due to unusual lockdown demand.

 

[blueone]

When Marvell and mellanox designed chiplets it was the beginning of the end of many engineers.

Why is this the case?

 

[blueone]

- ARM-based datacenter chips have had little success in displacing x86
Yet Google, Amazon, Alibaba, etc... are using ARM for their own datacenter chips.

Correct. In my view, the most revealing Arm architecture win has been at Oracle's cloud business by Ampere, which Oracle is also a very significant investor in. Larry Ellison is a smart guy. The Altra Max, with 128 single threaded Arm V8 cores, is very impressive. (Multiple hardware threads have inconsistent benefits for software. Some software benefits a lot, other software, not so much. Perhaps IBM Power CPUs are best at it.) Personally, if I were designing a cloud datacenter with custom software I'd probably choose Ampere's. The challenge for Ampere is that a lot of enterprise software is still written for and tuned for the x86 instruction set. Ampere also claims to be working on their own architectural variant of Arm cores, and I wonder if that's such a good idea. Apple succeeded with it in client CPUs, but for cloud computing, well, it should be interesting to see what Ampere comes up with.

 

[blueone]

When the Apple M1 chip came out, there was an observation that the performance/value curve for Apple cores had crossed over x86, an epochal achievement. Apple could now provide more compute for the dollar, a basic competitive advantage. That is a real differentiation, and in due time we will see Apple's market penetration grow, as a consequence.

ARM cores are not crossing over x86 though, AFAIK. That "no change" is boring but it has consequences just as Apple core cross-over has consequences.

Apple M1 cores use the Arm instruction set, and are based on one of the Arm 64bit CPU variants. Apple bought an "architecture license" from Arm, and added significant enhancements to instruction processing, but M1 cores are still Arm at the core. (Sorry for the silly pun, but I couldn't resist.)

Apple Announces The Apple Silicon M1: Ditching x86 - What to Expect, Based on A14

www.anandtech.com

 

[mgoldsmith1979]

Apple M1 cores use the Arm instruction set, and are based on one of the Arm 64bit CPU variants. Apple bought an "architecture license" from Arm, and added significant enhancements to instruction processing, but M1 cores are still Arm at the core. (Sorry for the silly pun, but I couldn't resist.)

To clarify, Apple's CPUs are ISA compliant to ARM v8 (and various extensions of that ISA). There is no HDL shared between what ARM designs on any product they license out to customers, and what Apple designs in-house. To say Apple CPUs "are based on one of the ARM 64b CPUs" would be akin to saying Intel cores are "based on one of the AMD x86-64 cores" because they support the same ISA - it is simply not true. Apple has designed their own CPU since the Swift (ARM v7A ISA) shipped in Apple A7 CPU in 2012; prior to that they licensed ARM Cortex CPUs for A4/A5/A6 generations. This is what the architecture license grants: the ability to design a from-scratch CPU that is compliant to the ISA being licensed; if the product were a derivative it would be using the "based on Cortex" license, which is a different beast (and, from what I can tell exclusive to Qualcomm).

 

[blueone]

To clarify, Apple's CPUs are ISA compliant to ARM v8 (and various extensions of that ISA). There is no HDL shared between what ARM designs on any product they license out to customers, and what Apple designs in-house. To say Apple CPUs "are based on one of the ARM 64b CPUs" would be akin to saying Intel cores are "based on one of the AMD x86-64 cores" because they support the same ISA - it is simply not true. Apple has designed their own CPU since the Swift (ARM v7A ISA) shipped in Apple A7 CPU in 2012; prior to that they licensed ARM Cortex CPUs for A4/A5/A6 generations. This is what the architecture license grants: the ability to design a from-scratch CPU that is compliant to the ISA being licensed; if the product were a derivative it would be using the "based on Cortex" license, which is a different beast (and, from what I can tell exclusive to Qualcomm).

I have no real basis for disagreeing with you, other than numerous industry articles, written by some people who should know better if you are correct, saying the M1 is "Arm-based". Arm-based, to me, goes beyond simply being instruction set compatible, it means one design is based on the other. I agree with your Intel-AMD analogy. Since I've never found any discussion of the topic from Apple insiders, I assume you either are/were an insider, or you have access to some.

 

[hist78]

I guess I'm convinced by Dan and Blueone, x86 on balance is a legacy cost not a differentiation.

Intel is a fab owner, they can raise prices while their fab cost is fixed, same as TSMC and probably Samsung is doing. Inflation/shortage is good for fixed assets.

On the downside, Intel had a tailwind from COVID lockdowns leading to unprecedented PC demand out of the blue. The disappearance of that demand may explain the mystery of the -1% sales growth (since it's not NAND); the YoY comparison may be unusually difficult due to unusual lockdown demand.

I believe the decreasing PC demand after the unusual surge triggered by COVID-19 lockdown might be a factor for Intel -1% revenue decline. But it doesn't explain the whole picture. I think the -1% revenue decline indicates Intel's problem is that they don't have the products the market needs and wants. Otherwise why all semiconductor companies (except Intel, ams OSRAM, and SkyWater Technology) can grow their business in Q1 2022?

1. For the Intel's stronghold x86 market, AMD is doing great in Q1. AMD ASP (Average Selling Price) for x86 product increased YoY and QoQ in Q1 2022.

2. For those non x86 market that Intel does have some presence, such as discrete graphic cards, AI, and embedded, Intel's competitors (Nvidia, AMD, etc.) are doing very well.

3. For those areas Intel doesn't have an offering, such as chips used in smartphones,
Qualcomm (41% YoY) and MediaTek (32.1% YoY) are doing amazingly.

Many those companies who achieved the great growth in Q1 2022 are either not using or not totally using leading edge node processes. And many of them don't even have a fab inhouse.

IMHO, Intel biggest problem is not in their manufacturing. Intel just doesn't have the right product to grab the market opportunity.

I've updated the semiconductor companies Q1 revenue growth list below:

Semiconductor Companies Q1 2022 Revenue Growth (Excluding equipment, software, and materials manufacturers)

May 4, 2022 Update: Added Wolfspeed earnings Wolfspeed, Inc. (NYSE: WOLF), formerly known as Cree, Inc., today announced revenue of $188.0 million for its third quarter of fiscal 2022, ended March 27, 2022. This represents a 37% increase compared to revenue from continuing operations of $137.3...

semiwiki.com

 

[mgoldsmith1979]

Arm-based, to me, goes beyond simply being instruction set compatible, it means one design is based on the other.

That is a common but incorrect interpretation, and various tech journalists don't do enough (IMO) to disambiguate. ARM has provided architecture licenses for versions of their ISA for several generations (at least as far back as v5/6A with X-Scale), but it always has meant a ground-up proprietary core design that is only ISA compliant (and ARM provide the tools to validate that ISA compliance). Qualcomm's Scorpion and Krait cores were ground-up designs compatible with the ISA, but they axed the design team after the SD810 or 820 and now "Kryo" cores are derived from Cortex family of off-the-shelf CPUs from ARM, which is based on a new tier of license model ARM introduced in 2016 (https://www.anandtech.com/show/10366/arm-built-on-cortex-license). Samsung's Mongoose lineage was a similar grounds-up and they also axed that design team after ~5th gen of that CPU.
Marvell had a team, Applied had a team, Broadcom had a team, NVidia had a team and AFAIK all gone now.

There have been a handful of other players in the market that took an existing machine and swapped the ISA (Broadcom Vulcan as example, which was converted from MIPS to ARMv8 ISA, but under the hood is mostly the same pipeline as their GC4400), so hard to tell if the machine itself is 'new' or only the use of ARM as the ISA vs MIPS /PowerPC /RISC.

Fujitsu has a custom ARM ISA core with the A64FX in Fugaku, and Nuvia had announced a core prior to their acquisition by Qualcomm. Ampere is said to be working on a custom core as well, but not clear which license model that will follow. HiSilicon as well had some custom or semi-custom cores (by name at least), but the changes there were limited to the support for China-specific crypto accelerators IIRC. Other start-ups designing custom CPUs appear to have migrated to RISC-V ISA now, and there is less interest in the mobile or network players to differentiate through custom ARM CPUs when the focus is on co-processors or accelerators instead (eg: better ISPs, neural network processors, etc).

 

[blueone]

That is a common but incorrect interpretation, and various tech journalists don't do enough (IMO) to disambiguate. ARM has provided architecture licenses for versions of their ISA for several generations (at least as far back as v5/6A with X-Scale), but it always has meant a ground-up proprietary core design that is only ISA compliant (and ARM provide the tools to validate that ISA compliance). Qualcomm's Scorpion and Krait cores were ground-up designs compatible with the ISA, but they axed the design team after the SD810 or 820 and now "Kryo" cores are derived from Cortex family of off-the-shelf CPUs from ARM, which is based on a new tier of license model ARM introduced in 2016 (https://www.anandtech.com/show/10366/arm-built-on-cortex-license). Samsung's Mongoose lineage was a similar grounds-up and they also axed that design team after ~5th gen of that CPU.
Marvell had a team, Applied had a team, Broadcom had a team, NVidia had a team and AFAIK all gone now.

There have been a handful of other players in the market that took an existing machine and swapped the ISA (Broadcom Vulcan as example, which was converted from MIPS to ARMv8 ISA, but under the hood is mostly the same pipeline as their GC4400), so hard to tell if the machine itself is 'new' or only the use of ARM as the ISA vs MIPS /PowerPC /RISC.

Fujitsu has a custom ARM ISA core with the A64FX in Fugaku, and Nuvia had announced a core prior to their acquisition by Qualcomm. Ampere is said to be working on a custom core as well, but not clear which license model that will follow. HiSilicon as well had some custom or semi-custom cores (by name at least), but the changes there were limited to the support for China-specific crypto accelerators IIRC. Other start-ups designing custom CPUs appear to have migrated to RISC-V ISA now, and there is less interest in the mobile or network players to differentiate through custom ARM CPUs when the focus is on co-processors or accelerators instead (eg: better ISPs, neural network processors, etc).

Thank you for the detailed response. Very informative.

 

[Xebec]

I believe the decreasing PC demand after the unusual surge triggered by COVID-19 lockdown might be a factor for Intel -1% revenue decline. But it doesn't explain the whole picture. I think the -1% revenue decline indicates Intel's problem is that they don't have the products the market needs and wants. Otherwise why all semiconductor companies (except Intel, ams OSRAM, and SkyWater Technology) can grow their business in Q1 2022?

1. For the Intel's stronghold x86 market, AMD is doing great in Q1. AMD ASP (Average Selling Price) for x86 product increased YoY and QoQ in Q1 2022.

2. For those non x86 market that Intel does have some presence, such as discrete graphic cards, AI, and embedded, Intel's competitors (Nvidia, AMD, etc.) are doing very well.

3. For those areas Intel doesn't have an offering, such as chips used in smartphones,
Qualcomm (41% YoY) and MediaTek (32.1% YoY) are doing amazingly.

Many those companies who achieved the great growth in Q1 2022 are either not using or not totally using leading edge node processes. And many of them don't even have a fab inhouse.

IMHO, Intel biggest problem is not in their manufacturing. Intel just doesn't have the right product to grab the market opportunity.

I've updated the semiconductor companies Q1 revenue growth list below:

Semiconductor Companies Q1 2022 Revenue Growth (Excluding equipment, software, and materials manufacturers)

May 4, 2022 Update: Added Wolfspeed earnings Wolfspeed, Inc. (NYSE: WOLF), formerly known as Cree, Inc., today announced revenue of $188.0 million for its third quarter of fiscal 2022, ended March 27, 2022. This represents a 37% increase compared to revenue from continuing operations of $137.3...

semiwiki.com

AMD Q2 ASPs are going to be very interesting. They’ve launched and are selling very premium next generation Epyc processors with vcache; but on the desktop they very recently got pushed down the stack. The top end chips are regularly selling for $150-200 USD off, and the mid range (formerly 5800X and 5600X) were more or less replaced by significantly cheaper alternatives (5600, 5700X) and also dropped in price themselves. Mobile is very hard to tell; the real mobile ASPs are in Corporate purchases and the big OEMs have been a bit slow to adopt AMD here unfortunately.

I suspect we’ll see both Intel and AMD drop ASPs in Q2..

 

[blueone]

AMD Q2 ASPs are going to be very interesting. They’ve launched and are selling very premium next generation Epyc processors with vcache; but on the desktop they very recently got pushed down the stack. The top end chips are regularly selling for $150-200 USD off, and the mid range (formerly 5800X and 5600X) were more or less replaced by significantly cheaper alternatives (5600, 5700X) and also dropped in price themselves. Mobile is very hard to tell; the real mobile ASPs are in Corporate purchases and the big OEMs have been a bit slow to adopt AMD here unfortunately.

I suspect we’ll see both Intel and AMD drop ASPs in Q2..

And I've been reading that Chromebook demand is projected to fall considerably in 2022. I suspect that hurts Intel more than AMD.