Holy shit balls is Intel (INTC) Fucked?

[immamac 42,835]

Huge earnings miss on an already bad day. 

15% headcount reduction between now and end 2024. 

20% reduction in capex for 5 nodes in 4 years their main catch-up strategy. 

Killed quarterly dividend effective next quarter. 

Trading below book value. @ 24.76 after hours as of this writing. Puts their market cap at 104B

Their fabs are worth 99B. 

Can they pull out of this or is this the end of Intel as we know it?

 

[Captainant 27,625]

Their recent issues with 13th and 14th gen chips causing permanent damage to themselves ain't gonna help much either. Their datacenter sales have been getting eaten alive by ARM, it's gonna take more than a leadership change to correct their ship

Edited August 1 by Captainant

[Mother mopar 3,318]

May get more traction in this thread. https://www.surlyhorns.com/board/topic/177-markets-still-falling-like-whoa/&share_tid=177&share_fid=2053247&share_type=t&link_source=app

[immamac 42,835]

5 minutes ago, Captainant said:

Their recent issues with 13th and 14th gen chips causing permeant damage to themselves ain't gonna help much either. Their datacenter sales have been getting eaten alive by ARM, it's gonna take more than a leadership change to correct their ship

Their client computing group is very profitable. 

Data center is in freefall. 

5 minutes ago, Mother mopar said:

May get more traction in this thread. https://www.surlyhorns.com/board/topic/177-markets-still-falling-like-whoa/&share_tid=177&share_fid=2053247&share_type=t&link_source=app

This is Intel specifically. 

[Captainant 27,625]

45 minutes ago, immamac said:

Their client computing group is very profitable. 

Data center is in freefall. 

AMD has been eating into their client computing sales, Intel no longer is the go to option in the corporate laptop and workstation segment and AMD has them beat at a lower price and power consumption for most consumer uses.

On the other hand, this goober just yolo'd a 700k inheritance onto INTC this morning so I'm sure it'll all work out lol

[Incredulity 11,646]

1 hour ago, immamac said:

Trading below book value. @ 24.76 after hours as of this writing. Puts their market cap at 104B

They traded at approx 25 a few times in late 22 early 23.

Thats not to dispute their fuckedness or not.

[0xdeadbeef 1,645]

 

1 hour ago, immamac said:

15% headcount reduction between now and end 2024. 

That makes for a brutal work environment.   Waiting for the axe to swing kills productivity.

They still have a commanding lead, but the trend in CPUs seems like its reason enough for new leadership.

 

https://www.techpowerup.com/322317/amd-hits-highest-ever-x86-cpu-market-share-in-q1-2024-across-desktop-and-server

 

[Captainant 27,625]

Imagine paying a $20,000,000,000 mortgage on your own fabs, and then contracting 3rd party fabs anyway because yours don't work

[DalTxHornFan 3,097]

IDK the tech side of chips, but I've always understood them to be something other than a low-cost provider.  Tough to compete with the foundries in Asia.

[immamac 42,835]

11 minutes ago, Captainant said:

AMD has been eating into their client computing sales, Intel no longer is the go to option in the corporate laptop and workstation segment and AMD has them beat at a lower price and power consumption for most consumer uses.

On the other hand, this goober just yolo'd a 700k inheritance onto INTC this morning so I'm sure it'll all work out lol

They aren't, the tales of AMD eroding their CPU market share and ARM are a myth. Their incumbent position isn't eroding dramatically and they showed strength in the CCG. The big fuckup is in the GPU unit and the data center CPU unit. They are making too many CPU and not selling enough GPU so their margin mix is fucked up. Gaudi 3 is super late to the party and no one knows if the ramp is going to be there per Intel's own guidance. 

They are trying to unfuck their fab while still pumping out an incredible amount of product and it's a balancing act on product assortment. They bet that AI was going to use CPU for inference more than GPU for some things (which is still true) but they are just not great for transformer based models. 

You have to understand these bets are made 2 years in advance with a little bit of wiggle room in them. They likely have a drastically better product from a CPU perspective for inferencing workloads that are not transformer based, but no one gives a shit because computer can chat now. 

[immamac 42,835]

2 minutes ago, Captainant said:

Imagine paying a $20,000,000,000 mortgage on your own fabs, and then contracting 3rd party fabs anyway because yours don't work

Big oof. 

1 minute ago, DalTxHornFan said:

IDK the tech side of chips, but I've always understood them to be something other than a low-cost provider.  Tough to compete with the foundries in Asia.

Intel only fabs their own chips until last year when they announced they would foundry. 

Samsung, TSMC and Global Foundry make up the majority of the foundry market today. 

[Sam Lin 3,550]

45 minutes ago, immamac said:

You have to understand these bets are made 2 years in advance with a little bit of wiggle room in them. 

Oh no, these bets are 5-10 years in advance. 2yrs doesn't even begin to steer the ship in the industry.

[immamac 42,835]

Just now, Sam Lin said:

Oh no, these bets are 5-10 years in advance. 2yrs doesn't even begin to steer the ship in the industry.

Product assortment and mix is 8ish quarters. This was a tactical error not a strategic one of a capability one. 

[Dbeasy 15,040]

When we were picking GPU’s a few years ago for early video AI applications, we briefly entertained Intel’s solution. Also, my neighbor works on the GPU chips for AMD. Both our engineers and the AMD guy said the same thing. While the hardware is good enough to compete with nVidia, Intel is way too behind on the software toolsets to ever be a factor in the next five to ten years. 

But, will they go out of business or decline severely? Doubtful, the US government is doing and will do whatever it takes to keep them alive and healthy because the invasion of Taiwan is now in sight. 

[immamac 42,835]

1 minute ago, Dbeasy said:

When we were picking GPU’s a few years ago for early video AI applications, we briefly entertained Intel’s solution. Also, my neighbor works on the GPU chips for AMD. Both our engineers and the AMD guy said the same thing. While the hardware is good enough to compete with nVidia, Intel is way too behind on the software toolsets to ever be a factor in the next five to ten years. 

But, will they go out of business or decline severely? Doubtful, the US government is doing and will do whatever it takes to keep them alive and healthy because the invasion of Taiwan is now in sight. 

The weird thing about Intel is it's this gigantic mess of shit. It could be 6-8 companies. I fear that this will get them broken up more than they will just go away if that makes sense. 

[TwiceHorn 31,438]

Yeah, they've acquired a ton of inertia making basically the same product for as long as they have.  Hard to navigate that behemoth.

[elfenix 34,222]

2 hours ago, DalTxHornFan said:

IDK the tech side of chips, but I've always understood them to be something other than a low-cost provider.  Tough to compete with the foundries in Asia.

it really wasn't that long ago that intel was 2 nodes ahead of anything else in the world.  now it's like they're 2 nodes behind (what is a node has become fuzzier and fuzzier over the last few years so it's a bit harder to tell)

 

this chart is really weird to me considering just how compelling AMD's zen products have been.

 

Edited August 2 by elfenix

[MonkeyDoughnut 10,227]

Intel's going is about to have to recall a shit ton of chips and because they basically admitted they knew of issues they're going to be on the hook for a lot of damages to not just consumers, but partners as well. It's going to be another rough year or two all while Ryzen is getting nothing but praise.

[troph 24,781]

I’ve been an AMD fan since the early 2000s and love seeing them do well. Can’t trade them they are client and while I’m not an insider I’m also not a member of congress and I enjoy my freedom. But it’s great to see their people rewarded. I’ve built several PCs with Ryzen and either AMDs GPU or Nvidia and they have been amazing computers. 

Edited August 2 by troph

[immamac 42,835]

1 hour ago, MonkeyDoughnut said:

Intel's going is about to have to recall a shit ton of chips and because they basically admitted they knew of issues they're going to be on the hook for a lot of damages to not just consumers, but partners as well. It's going to be another rough year or two all while Ryzen is getting nothing but praise.

Their (AMD) gaming division is in complete freefall to irrelevance. They are fucked. Go look at their earnings call. Both gaming GPU and CPU are meh to bad outside the top skus. 

Intel has to recall the top binned chips, which account for around 7-9% of all chips from their respective generation. (13900k and 14900k) And that's only for the ones that had the voltage problem cook the chip. I'd say probably less than 30% of those. So in reality they are looking at a 3% RMA on the entire generation. Not irrelevant, but not the disaster you are claiming it is. 

Ryzen 9950X and the 9800X3D etc are not even shipping yet. The 7000 series was fucked and couldn't even support DDR5 past 32GB until about 4 months ago. The platform was released way prematurely and they have just finally worked through the kinks. 

The X3D chips are also horribly imbalanced because the L4 cache they added is only on one CCD so half the cores have a pretty big latency disadvantage, so much so that it's better to just disable those cores. The 9000 series is reportedly supposed to balance this back out. 

Also 14th Gen Intel was the last legacy architecture chip. Raptor lake refresh. 

The new core ultra is going to shit on it as they move to a true chiplet design and use a TSMC based process for some parts. Zen5 and Zen5c are going to be pushing the limits of what the zen architecture can bring according to many because the technical debt is so high (this is from Tim Keller who designed the zen architecture). They need to redo the architecture to continue scaling. The next Gen Intel is a brand new architecture and won't have this issue. Code named arrow lake. 

 

[Pato del Muerto 52,485]

Yeah I just can’t understand why this forum isn’t busting with hot opinions. 

[Gravy Train 974]

1 hour ago, immamac said:

They need to redo the architecture to continue scaling. The next Gen Intel is a brand new architecture and won't have this issue. Code named arrow lake. 

I'm excited for Arrow Lake and feel like it's the right move to mainstream desktop architecture to compete with Snapdragon (ARM) and TSMC/Apple.  It seems like Raptor Lake promised too much out of the gate and they had more time to get it right with Arrow Lake, despite the die incompatibility.  

[immamac 42,835]

27 minutes ago, Gravy Train said:

I'm excited for Arrow Lake and feel like it's the right move to mainstream desktop architecture to compete with Snapdragon (ARM) and TSMC/Apple.  It seems like Raptor Lake promised too much out of the gate and they had more time to get it right with Arrow Lake, despite the die incompatibility.  

Raptor Lake was kind of the last hoorah for the monolithic die. It came back to bite them because they wanted to push as much fucking power as possible through the chip to get insane clocks at a certain tdp. It was a dirty way to try to win category leadership with a very obviously inferior process node and general architecture. 

UCIe is one of the coolest fucking things the industry has coming and people really haven't started to grasp how much of a difference maker it's going to be. I have a meteor lake laptop and after the first few months of shitty bios firmware when they got it dialed in it's incredible. I can't imagine how fucking badass lunar lake with up to 32GB of LPDDR5X-8533 ON PACKAGE is going to be. 

That's bananas. 68.26GB/s bandwidth available to any core at incredibly low latency. 

[Captainant 27,625]

I think Intel has stepped on a bigger pile than they realize with the 13th and 14th Gen issues. They're burning their HEDT users and denying RMA and warranty requests on items that Intel acknowledged were manufacturer damaged parts

The above post is one anecdote, but I know a couple people who have had to completely rebuild their intel 13th gen machine and moved to AMD after constant crashes while using stock mobo profiles.

8 hours ago, immamac said:

UCIe is one of the coolest fucking things the industry has coming and people really haven't started to grasp how much of a difference maker it's going to be. I have a meteor lake laptop and after the first few months of shitty bios firmware when they got it dialed in it's incredible. I can't imagine how fucking badass lunar lake with up to 32GB of LPDDR5X-8533 ON PACKAGE is going to be. 

That's bananas. 68.26GB/s bandwidth available to any core at incredibly low latency. 

I think other API improvements like direct storage access over PCIe are gonna be a bigger deal for workload scaling than the CPU/sysmem interlink. Most enterprise and consumer software are plagued with single threaded bottle necks that more often than not make data transfer take as much time or longer than the compute work itself 

[immamac 42,835]

12 minutes ago, Captainant said:

I think Intel has stepped on a bigger pile than they realize with the 13th and 14th Gen issues. They're burning their HEDT users and denying RMA and warranty requests on items that Intel acknowledged were manufacturer damaged parts

The above post is one anecdote, but I know a couple people who have had to completely rebuild their intel 13th gen machine and moved to AMD after constant crashes while using stock mobo profiles.

I think other API improvements like direct storage access over PCIe are gonna be a bigger deal for workload scaling than the CPU/sysmem interlink. Most enterprise and consumer software are plagued with single threaded bottle necks that more often than not make data transfer take as much time or longer than the compute work itself 

SPDK has existed for nearly 8 years, and the software that implemented it is incredible, but software is gonna be bad forever because people don't actually care about performance if it means work. Direct storage isn't really good for anything besides peripheral to storage communication and that API is mainly only useful for gaming at this point. 

NVMe is too slow (both throughput and latency) to matter for direct storage in enterprise workloads. 

The reality is that Intel simply can't make that many 1X900K skus. It's a top bin, perfect functionality with no defects processor. It's the lowest percentage sku they make. The 1X700K and 1X800K both sell significantly more parts. The reality is that the HEDT market makes up less than 2% of all PC sales and the high end gaming market is sub 10%. 

It also used to matter more when your CPU was the most expensive part in your system by % of entire cost. Now the CPU dwarfs it. A 4090 or even 4080 costs 4x what a top bin CPU costs. Those caught on fire and literally blew up because of the power connectors and got denied RMA. Yet here we are with Nvidia still fucking trouncing AMD in the GPU side of things. It's the same story from Intel on the CPU side. 

I'm not sitting here defending Intel. I think they fucked up incredibly bad and handled the entire 13th and 14th Gen poorly. I personally believe the entire 14th Gen lineup was a mistake. It's nearly the exact same silicon with minor tweaks to microcode instead of actually new functionality or layout. 

I just want to make extra clear that I don't disagree that Intel fucked it royally here, I just also understand that it doesn't really matter in the grand scheme of things because of the overall picture. 

[Trey3216 15,343]

15 hours ago, immamac said:

The weird thing about Intel is it's this gigantic mess of shit. It could be 6-8 companies. I fear that this will get them broken up more than they will just go away if that makes sense. 

 

15 hours ago, TwiceHorn said:

Yeah, they've acquired a ton of inertia making basically the same product for as long as they have.  Hard to navigate that behemoth.

These two posts make me think that Intel is a lot like GE in a stock/company valuation sense.  Just a big, laborious behemoth that got caught behind the 8ball 

[immamac 42,835]

2 minutes ago, Trey3216 said:

 

These two posts make me think that Intel is a lot like GE in a stock/company valuation sense.  Just a big, laborious behemoth that got caught behind the 8ball 

It's comparable but it's a bit more nuanced than that as they only do semiconductor and really only focus on logic + a bit of networking which translates effectively. They exited the storage business both NAND and Optane. 

The issue is they have such incredible manufacturing scale that people (investors and customers) don't quite grasp how insane it is. 

ARM doesn't make a single chip. They have some reference designs, but they have never manufactured anything at scale. 

Intel chips are in 80%+ of all modern computers including the cloud. It's staggering. That's the issue though, the manufacturing side has never been held to a margin standard or accountable for earnings because that's the machine. That machine was a broken down piece of shit when gelsingher took over and now it's turning into a modern marvel again, but they have no customers besides themselves and their chips can't find that kind of foundry business anymore because the vertical cost structure advantage isn't there if you don't get to realize it over the 7-10 year cycle.

I don't envy them, they are staring down the barrel of a complete disaster. And if they fail we will all feel it because TSMC can't pick up the slack. 

[Dahobbs 17,522]

11 hours ago, immamac said:

Raptor Lake was kind of the last hoorah for the monolithic die. It came back to bite them because they wanted to push as much fucking power as possible through the chip to get insane clocks at a certain tdp. It was a dirty way to try to win category leadership with a very obviously inferior process node and general architecture. 

UCIe is one of the coolest fucking things the industry has coming and people really haven't started to grasp how much of a difference maker it's going to be. I have a meteor lake laptop and after the first few months of shitty bios firmware when they got it dialed in it's incredible. I can't imagine how fucking badass lunar lake with up to 32GB of LPDDR5X-8533 ON PACKAGE is going to be. 

That's bananas. 68.26GB/s bandwidth available to any core at incredibly low latency. 

Thoughts on Qualcomm's Snapdragon X Pro and X Elite?

I really hope Intel has finally gotten its manufacturing issues sorted. I guess we'll see over the next couple of years. We really need that domestic manufacturing base.  

[immamac 42,835]

1 minute ago, Dahobbs said:

Thoughts on Qualcomm's Snapdragon X Pro and X Elite?

I really hope Intel has finally gotten its manufacturing issues sorted. I guess we'll see over the next couple of years. We really need that domestic manufacturing base.  

Qualcomm has issues that may prevent the longevity of that product line associated with the licensing costs they owe to ARM. It could even go to product sale injunction. 

In an apples to apples comparison versus apple silicon snapdragon mostly holds it's own. It does a great job at what it's supposed to do and sips power and the thermals are awesome. 

It can't game for shit and the NPU is a bunch of fabricated nonsense. 

I'm not sold on anything on ARM just yet over the X86 instruction set, but for just fucking around and using web apps it's great. For more productivity focused workloads I think it's going to get trounced by the latest Gen X86 stuff from both AMD and Intel. 

[swraith 345]

19 hours ago, immamac said:

 

20% reduction in capex for 5 nodes in 4 years their main catch-up strategy. 

 

What does this mean in regard to nodes? I understand the capex cut but I don't understand the refence to nodes.

[immamac 42,835]

Just now, swraith said:

What does this mean in regard to nodes? I understand the capex cut but I don't understand the refence to nodes.

So Intel manufacturing has process nodes, these nodes are the core of their manufacturing capabilities. Typically you hear this mentioned in nanometers or angstroms etc. 

Intel 7 is actually an advanced 10nm process. Intel 5, Intel 3, Intel 18A are all the new ones. 

So part of it is shrinking the size of the transistors that are possible to make and part of it is using new technologies to actually grow/make/etch the chip transistors. 

Intel is going to have something called backside power delivery starting in their 18A node which basically makes x86 vs ARM comparisons in power draw a thing of the past. It's radically more efficient at delivering power to the transistors etc. 

These nodes were generally done in a tick/tock fashion at Intel. They would do a lot of work to get the node ramped and then improve the architecture of the chips on the 2nd generation of that node while they worked on the next node. 

5 nodes in 4 years is basically paying back all the technical debt and catch up to get to the equivalent 2nm process that TSMC is ramping right now and regain category leadership by 2028-2030ish. 

Each node gets harder and harder from a science/physics perspective at this point and we are starting to really get to the limits of what we think is possible using silicon as the semiconductor. We keep inventing new tech to push out the limit, but it's getting to be incredibly capital intensive. 

The majority of chips in cars, microwaves whatever are all 12nm+ in size and a lot of them are 20 something. The latest and greatest flagship products are all on Intel 5/4/3 or TSMC 4/3/2nm. 

There's only so much wafer capacity (how many discs you can make chips on) for these factories. The cool part about advanced nodes is you can make really small chips that can do the job of some older much larger in physical size chips for more $$ but less power and less size. 

The crazy part about 5 nodes in 4 years is you aren't going to even get close to paying back the nodes by the time you settle back into a cadence and you are effectively massively increasing your capacity in the hopes that you can make chips for other people outside of Intel. They haven't had anyone like Nvidia go to them for manufacturing the GPUs. That's still possible, but it's just unlikely. 

[TwiceHorn 31,438]

https://semiengineering.com/knowledge_centers/manufacturing/process/nodes/

Less jargony.

https://en.wikichip.org/wiki/technology_node

Edited August 2 by TwiceHorn

[Captainant 27,625]

57 minutes ago, immamac said:

I'm not sold on anything on ARM just yet over the X86 instruction set, but for just fucking around and using web apps it's great. For more productivity focused workloads I think it's going to get trounced by the latest Gen X86 stuff from both AMD and Intel. 

I disagree, ARM is so customizable that if customized to the right spec frequently beats x86 in raw throughput and processing capacity - not to mention the lower power consumption. That sort of thing probably is going to live more in the datacenter than in end-user devices, but every hyperscaler is releasing in-house designed ARM processors that are often a performance uplift (if your code is ARM compatible).

And in terms of client computing I think it's just a matter of time until Qualcomm and Apple build something that's good enough but cheaper than x86 that'll eat away at their base of sales. No doubt that HEDT will still be x86, but that's not really what ARM is meant for anyways

[Wiler77 1,566]

Couldn't happen to a more arrogant, entitled, and all around shithead of a company.  This has been building for years.

Sucks for the folks out of a job (I would never celebrate that), and Intel flat out failing would be a problem.

But the past 5 years is what happens when you take your eye off the ball on your crown jewel (manufacturing) to the point not even their bribes and threats against "partners" have much meaning anymore.  FAFO indeed.

Gelsinger is a smart dude and has the focus where it should be.  Will be interesting if he's given the rope he needs to truly right the ship.  It was a 10 year rebuild when he took over.

[immamac 42,835]

Just now, Captainant said:

I disagree, ARM is so customizable that if customized to the right spec frequently beats x86 in raw throughput and processing capacity - not to mention the lower power consumption. That sort of thing probably is going to live more in the datacenter than in end-user devices, but every hyperscaler is releasing in-house designed ARM processors that are often a performance uplift (if your code is ARM compatible).

And in terms of client computing I think it's just a matter of time until Qualcomm and Apple build something that's good enough but cheaper than x86 that'll eat away at their base of sales. No doubt that HEDT will still be x86, but that's not really what ARM is meant for anyways

ARM is slower and lower clocked. It's inappropriate to use for extremely intensive single core workloads where performance actually matters. This is proven in all the benchmarks and it lacks a lot of key accelerators for enterprise compute. CISC isn't a marketing gimmick. 

ARM shines in workloads where "good enough" reigns supreme and people have tons of small simple jobs or calculations to do. So yeah your run of the mill shitty SaaS startup or a bunch of assholes pretending to do computers are gonna think ARM is bitchin. It's perfect for the cloud where no one knows what the fuck they are running or doing. 

https://docs.databricks.com/en/compute/configure.html#aws-graviton-instance-types

Limitations section. "you basically can't use data bricks lol" 

[blacklab 11,926]

no one should use databricks. Never dealt with a bigger bunch of assholes in my life.

[52-80 19,143]

46 minutes ago, Wiler77 said:

But the past 5 years is what happens when you take your eye off the ball on your crown jewel (manufacturing) to the point not even their bribes and threats against "partners" have much meaning anymore. 

I’m just a tourist to the semi industry but isnt the core value of these companies in chip design, not chip manufacturing - which itself can easily be outsourced

[Texas Jeff 1,944]

I’m just a tourist to the semi industry but isnt the core value of these companies in chip design, not chip manufacturing - which itself can easily be outsourced

The United States needs a domestic chip manufacturing base. We rely too much on TSMC which is way too close to China.

It’s ok to separate chip design from chip manufacturing, but for our own defense we need someone … or a few someones … to make chips here with the latest technology.

Even if chip manufacturing is not profitable, we still need someone to do it here.

[52-80 19,143]

9 minutes ago, Texas Jeff said:

The United States needs a domestic chip manufacturing base. We rely too much on TSMC which is way too close to China.

It’s ok to separate chip design from chip manufacturing, but for our own defense we need someone … or a few someones … to make chips here with the latest technology.

Even if chip manufacturing is not profitable, we still need someone to do it here.

Sure, having a domestic foundry is value in and of itself. But as far as Intel getting trounced by AMD and NVDA in the last years, isnt that due to their inferior performance which is a function of planning and design, not a consequence of e.g. production inefficiency and high defect rate and factors like that

[immamac 42,835]

21 minutes ago, 52-80 said:

Sure, having a domestic foundry is value in and of itself. But as far as Intel getting trounced by AMD and NVDA in the last years, isnt that due to their inferior performance which is a function of planning and design, not a consequence of e.g. production inefficiency and high defect rate and factors like that

No, it was because Intel was vertically integrated and they had to design to their fab capabilities. Intel is not fabless they are the largest fab on earth by wafer capacity for X86. 

Step 3 of the plan was decoupling this and their first Gen of mixed tiles from TSMC and Intel is in market now (meteor lake). The ramp for meteor lake has been hindered by Intel's fab capacity basically being shit on yield. 

[swraith 345]

I am under the assumption that Intel is benefiting from the CHIPS Act to invest domestic chip production.

Intel and Biden Admin Announce up to $8.5 Billion in Direct Funding...

[immamac 42,835]

12 minutes ago, swraith said:

I am under the assumption that Intel is benefiting from the CHIPS Act to invest domestic chip production.

Intel and Biden Admin Announce up to $8.5 Billion in Direct Funding...

They are going to need to spend close to 10x that for leadership over the next 5-6 years is the issue. It's hard to grasp how incredibly capital intense it is to do semiconductor manufacturing now. 

I've questioned a lot of it in the last 2-3 years as to why? What are we getting with all these "better" semiconductors? 

Have phones become better? Have laptops? The answer is unequivocally no. From around 2018-2019 to now there are some power efficiency gains and theoretical max performance gains, but in the real world it translates to not that much more productivity and not that different of a workflow for the majority of consumer use cases. 

For enterprise usecase the need for more performance is fueled by more data needing to be processed, but that has also reached a point of "is this all just bullshit?" I've worked on and helped design some of the largest database footprints in the world, the data they are storing and processing is not really the core application or service data. It's a bunch of stuff that they collect so they can productize it and sell it to others. 

Uber is such a great tangible example because it's a consumer product. Have your rides really gotten better, faster, more efficient, etc? The answer is they have not perceptively changed. Now your app experience is loaded with other bullshit Uber wants to sell you and tell you to do and show you while you are on the ride. The data for your trip is relatively small compared to the engagement data they have on everything else you are doing in the app. If you stripped Uber down to the core service of ride matching and driver stuff it could run on 1/100th or less of the compute.

This is what Elon saw at Twitter when he joined and why the platform with removal of a ton of infrastructure changed very little. 

[Captainant 27,625]

12 minutes ago, immamac said:

This is what Elon saw at Twitter when he joined and why the platform with removal of a ton of infrastructure changed very little. 

aside from their uptime and error rates getting alot worse lol

[Wiler77 1,566]

1 hour ago, 52-80 said:

I’m just a tourist to the semi industry but isnt the core value of these companies in chip design, not chip manufacturing - which itself can easily be outsourced

Intel's core value was manufacturing.  They happened to sell a bunch of products their fabs built, but the crown jewel was the fabs.  They had a 1-2 node lead up until 2016 or 2017.  That completely flipped on it's head and they have been 1-2 nodes behind because they utterly shit the bed in a short amount of time.  Now they have a bunch of expensive fabs that aren't running at capacity because they've lost so much market share.

An improvement in the fab technology essentially means your products will be able to do more work for the same amount of power, or the same work for less power (or some combination of the 2).  It's a big deal.  It's why they are also using TSMC for some products.

[52-80 19,143]

30 minutes ago, immamac said:

I've questioned a lot of it in the last 2-3 years as to why? What are we getting with all these "better" semiconductors? 

Have phones become better? Have laptops? The answer is unequivocally no. From around 2018-2019 to now there are some power efficiency gains and theoretical max performance gains, but in the real world it translates to not that much more productivity and not that different of a workflow for the majority of consumer use cases.

If you approach this with a pessimistic lens you could have extended this question to the 10 years and you could have also argued no for an answer. 

From last few years, from my own personal use Apple’s M series has been incredible. Better processing allows better computational photography, like all the portrait and hdr tricks, super high resolution and framerate video capture and production on a cell phone alone, on-device live language translation, etc etc etc. 

Theres never been any tech innovation in a 3 year window that overnight makes a person twice as productive, but i bet you wouldnt commit yourself to this 3-year old tech for another 3 years, would you?

[Dahobbs 17,522]

2 hours ago, 52-80 said:

I’m just a tourist to the semi industry but isnt the core value of these companies in chip design, not chip manufacturing - which itself can easily be outsourced

It can't be easily outsourced. There is essentially one foundry right now that competes on bleeding edge of manufacturing (other than for simpler stuff like memory), TSMC. Having an entire industry subject to a single manufacturer is bad for all sorts of reasons (pricing, capacity, etc.), but its even worse when that company is subject to constant threats of invasion from China. 

Edited August 2 by Dahobbs

[Dahobbs 17,522]

2 hours ago, 52-80 said:

Sure, having a domestic foundry is value in and of itself. But as far as Intel getting trounced by AMD and NVDA in the last years, isnt that due to their inferior performance which is a function of planning and design, not a consequence of e.g. production inefficiency and high defect rate and factors like that

A big part of AMD's efficiency and performance gains have been driven by the node improvement rather than by design changes. As to NDVIDIA, Intel hasn't really competed in the same space as them until very recently (push for discrete GPU and AI). But again, a lot of NDVIDIA's gains have been driven by the node improvement, not merely design. TSMC is already essentially tapped out in terms of capacity until some more fabs become operational. And despite investment in a fab in Arizona, TSMC has been reluctant to actually bring its bleeding edge know-how outside of Taiwan. A lot of that is economic, but another big factor is the feeling that Taiwan is safer if the US absolutely has to protect it to keep access to its chip manufacturing.  

[immamac 42,835]

7 minutes ago, Dahobbs said:

A big part of AMD's efficiency and performance gains have been driven by the node improvement rather than by design changes. As to NDVIDIA, Intel hasn't really competed in the same space as them until very recently (push for discrete GPU and AI). But again, a lot of NDVIDIA's gains have been driven by the node improvement, not merely design. TSMC is already essentially tapped out in terms of capacity until some more fabs become operational. And despite investment in a fab in Arizona, TSMC has been reluctant to actually bring its bleeding edge know-how outside of Taiwan. A lot of that is economic, but another big factor is the feeling that Taiwan is safer if the US absolutely has to protect it to keep access to its chip manufacturing.  

Nvidias GPUs are arguably worse in some cases than AMD equivalent node. Their advantage is in their software ecosystem and developer friendly tooling and deployment not in superior silicon. 

The fun part is the big customer for Intel 18A that is unannounced is most certainly one of AMD or Nvidia. If they pull that off and leapfrog TSMC Intel gets the last laugh. 

[Dahobbs 17,522]

22 minutes ago, immamac said:

Nvidias GPUs are arguably worse in some cases than AMD equivalent node. Their advantage is in their software ecosystem and developer friendly tooling and deployment not in superior silicon. 

The fun part is the big customer for Intel 18A that is unannounced is most certainly one of AMD or Nvidia. If they pull that off and leapfrog TSMC Intel gets the last laugh. 

Yeah, hopefully Intel is taking a short term kick in the pants for a longer term gain. I know they've done a lot to try and right the manufacturing debacle that occurred with 14nm and 10nm (and the renamed 7nm). Betting on quad patterning over EUV ended up being a huge mistake. Whether someone likes or hates Intel, the truth is we need them as no one else domestically is in a position to manufacture volume chips on the bleeding edge.  

Fairly good write up on this for others interested (not you since I know you know this stuff better than I do): https://www.tomshardware.com/pc-components/cpus/intel-completes-assembly-of-first-commercial-high-na-euv-chipmaking-tool-as-it-preps-for-14a-process

[Wiler77 1,566]

1 hour ago, immamac said:

Nvidias GPUs are arguably worse in some cases than AMD equivalent node. Their advantage is in their software ecosystem and developer friendly tooling and deployment not in superior silicon. 

The fun part is the big customer for Intel 18A that is unannounced is most certainly one of AMD or Nvidia. If they pull that off and leapfrog TSMC Intel gets the last laugh. 

ding ding ding.  NVidia is a SW company at this point, and they openly admit it.  They have a Secretariat-sized lead over others.  That's not to say others can't be profitable, but NVidia is set to dominate for a LONG time.

Pretty sure the heavy rumors on 18A are that NVidia is that customer.  I can't imagine placing a bet like that on Intel after the struggles they've had.  Surely Jensen has a plan B.  Hell, they have enough money to buy capacity at multiple fabs just to keep the competition from getting it.  That would be an Intel-style move.