News Intel 4Q25 Earnings

[DrMrLordX]

bleak margin looking forward

Oh so you're looking at the Q1 guidance.

 

[Joe NYC]

So somehow they have no foresight of any sort. I don't know if this is a) they are all stupid b) it's too chaotic c) even malice/intentional.

The decision making is probably just too slow. By the time it is finally agreed on, it was obsolete.

 

[511]

So somehow they have no foresight of any sort. I don't know if this is a) they are all stupid b) it's too chaotic c) even malice/intentional.

Intel is always either too early or too late to the boat for a while

Means they leech off tightly gatekept IP/patent portfolios.

Every Megacorp in nutshell tbh

 

[mikegg]

You know, they say hindsight is 20/20, but now I call theml, Intel "bad timing" Corporation.

They make a decent chip with Pantherlake, at a time when the prices in memory and SSD will likely limit it's impact. And they got their dGPU out months after crypto boom stopped. And if Xe4 rumors are accurate, then perhaps after AI market crashes they would be significantly behind competition, to the likes of B390 vs 890M.

Had they continued their Optane lineup, they could have benefitted from the AI push for their datacenter lineup. Heck, they could have got enthusiast Optane DIMMs out for PCs, and have a slow memory + fast memory tier. The DIMMs with 100-300nS latency is actually fast enough to be RAM for vast majority of use cases.

Here’s another one they will miss:

Right now, local LLMs are a relatively small market. But at some point, local models will be good enough and interest in this market will skyrocket.

Guess what? No unified memory consumer products with a big GPU from Intel.

Apple and AMD will benefit the most. AMD with Strix Halo. Apple with Pro and Max series Macs.

And by the time Intel comes out with something, market boom will probably be over.

They have truly missed every single trend in the last 10 years. Every single one. Call it unlucky or poor management or both.

 

[adroc_thurston]

But at some point, local models will be good enough and interest in this market will skyrocket.

Uh, no. Scaling laws still hold true.
Anything remotely capable will run on a big chungus rack stashed in some data center.

 

[mikegg]

Uh, no. Scaling laws still hold true.
Anything remotely capable will run on a big chungus rack stashed in some data center.

Uh, yes. Local models will never be as good as cloud models. It won’t change. What will change is local models will get good enough to be very useful and local hardware will also be the same.

Perfect example is M5 series adding matmul acceleration into GPUs. Now suddenly prompt processing is 4x faster than M4 series. Run something like GLM 4.7 flash locally and you can do something very useful and private.

Don’t underestimate people wanting local waifus and generating porn offline.

 

[511]

Don’t underestimate people wanting local waifus and generating porn offline.

Yeah this can't be underestimated

Perfect example is M5 series adding matmul acceleration into GPUs. Now suddenly prompt processing is 4x faster than M4 series. Run something like GLM 4.7 flash locally and you can do something very useful and private.

We had Matmul acceleration on local hardware before Apple is way late in the game

 

[poke01]

Intels got bigger problems than getting people to buy their SoCs for local LLMs.

 

[adroc_thurston]

What will change is local models will get good enough to be very useful and local hardware will also be the same.

They won't. Scaling laws still hold true.
Overfitting a 7B model to do well in benchmarks won't make it more capable.

Don’t underestimate people wanting local waifus and generating porn offline.

The market for stonecold truecel jackoff sessions is smaller than you think, and they all roll gacha anyway.

 

[Fjodor2001]

Right now, local LLMs are a relatively small market. But at some point, local models will be good enough and interest in this market will skyrocket.

Guess what? No unified memory consumer products with a big GPU from Intel.

NPU is what will be used for this if anything on regular PCs.

 

[poke01]

We had Matmul acceleration on local hardware before Apple is way late in the game

Yep Nvidia did it in 2018, AMD in 2023 and Intel in 2022
*consumer hardware

 

[511]

Intels got bigger problems than getting people to buy their SoCs for local LLMs.

Intel's main issue is execution in DCAI and IFS to some extent at least IFS is making progress can't say that about DCAI

 

[adroc_thurston]

Intel's main issue is execution in DCAI and IFS to some extent at least IFS is making progress can't say that about DCAI

Intel issue is that their CPU IP plainly sucks.
The rest is collateral stuff.

 

[511]

Intel issue is that their CPU IP plainly sucks.
The rest is collateral stuff.

P core sure not for E core as for collateral i don't think IFS is collateral Intel was and is a manufacturer

 

[adroc_thurston]

not for E core

Atrocious power and they know it and they're doing nothing about it.

i don't think IFS is collateral Intel was and is a manufacturer

18A yield being dookie is an issue fixable in a year.
CPU IP is a thing they can fix maybe with UC.

 

[511]

Atrocious power and they know it and they're doing nothing about it.

they are?

18A yield being dookie is an issue fixable in a year.

Maybe IN 2-3 qtr year is a stretch

CPU IP is a thing they can fix maybe with UC.

should have done sooner was royal was having skill issue why did Royal even got funding smh

 

[adroc_thurston]

they are?

Given UC targets, yea.

Maybe IN 2-3 qtr year is a stretch

Year's normal given how bad they're having it now.

should have done sooner was royal was having skill issue why did Royal even got funding smh

I mean it looked like a cool moonshot. why not?

 

[511]

I mean it looked like a cool moonshot. why not?

Given UC targets, yea.

there is your answer should have handed the E core team the reign after seeing ARL ES and Royal Missing the target

 

[adroc_thurston]

there is your answer should have handed the E core team the reign after seeing ARL ES and Royal Missing the target

I mean it would've accelerated UC by like a year maybe. meh.

 

[511]

or two

 

[adroc_thurston]

or two

no way, it would've been stuck in pre-def until LNC faceplanted itself in retail anyway.

 

[511]

no way, it would've been stuck in pre-def until LNC faceplanted itself in retail anyway.

Intel had performance projection before the ES they knew what was going to happen with the P cores

 

[Win2012R2]

So 18A yields are still crap?

13 months ago -

9 December 2024: "Gelsinger fires back at recent stories about 18A's poor yields, schools social media commenters on defect densities and yields"

https://www.tomshardware.com/tech-industry/semiconductors/gelsinger-fires-back-at-recent-stories-about-18as-poor-yields-schools-social-media-commenters-on-defect-densities-and-yields

 

[511]

So 18A yields are still crap?

13 months ago -

"Gelsinger fires back at recent stories about 18A's poor yields, schools social media commenters on defect densities and yields"

uhh this was before the node was in HVM ........ as for the yields they are not where LBT want it to be

 

[Win2012R2]

uhh this was before the node was in HVM ........

So yields were excellent before HVM but became crap at HVM? 🙄