Can Intel comeback ?

[podkall]

46 minutes ago, Satan_Prometheus said:

I don't think it really has anything to do with "fanboys" since AFAIK the DIY market is a pretty small percentage of the overall market for computer chips. I would think that corporate clients (datacenter clients and large prebuilt OEMs) are significantly more important, though if somebody has data suggesting otherwise that would be cool to see!

it's what makes you think these datacenters aren't fanboys? AMD can provide same core count with less wattage and more efficient CPUs yet people still choose Intel

[jordanbuilds1]

5 minutes ago, podkall said:

it's what makes you think these datacenters aren't fanboys? AMD can provide same core count with less wattage and more efficient CPUs yet people still choose Intel

what makes you think that companies care about that, they just choose the first option they see and keep it pushing.

[GuiltySpark_]

8 minutes ago, jordanbuilds1 said:

what makes you think that companies care about that, they just choose the first option they see and keep it pushing.

No, they buy what they know. 

 

"Nobody Gets Fired For Buying IBM Intel"

 

 

[Ha-Satan]

I'm definitely not well-enough informed to try to guess at the factors being considered by datacenter owners when they consider server CPU purchase, and I don't want to argue about the definition of a "fanboy" because I think it can kinda mean whatever you want it to mean.

 

But I'll point out that AMD's server market share has increased at a massive rate over the last 7ish years, well outpacing their market share growth in desktop and mobile:

 

https://www.tomshardware.com/pc-components/cpus/amd-records-its-highest-server-market-share-in-decades-but-intel-fights-back-in-client-pcs

 

Clearly, movers and shakers in the datacenter market are moving over to AMD at a much higher rate than in other markets.

[Beerzerker]

53 minutes ago, podkall said:

it's what makes you think these datacenters aren't fanboys? AMD can provide same core count with less wattage and more efficient CPUs yet people still choose Intel

Don't forget alot of that is based on contracts.
A company will get a quote from a builder/supplier for server setups based on both makes and obviously go with who's the cheaper of the two for what the company wants.

AMD has taken at least a decent chunk of the server market, no doubt about it and looks like they will continue gaining ground there.... That is once contracts expire and are redone.
Intel still holds a majority of that market ATM because they are still selling their chips cheap enough many will consider them yet and yes, business types have connections they use to leverage the absolute best deal they can get.

They don't care about brand loyalty, just the price of it all and if it works.

It's known as fact the server market is larger than any other when it comes down to it.
Gamers and lappy users, while what they buy still matters are not as large of a market for either as the server part is.

What's helping AMD right now is their efficiency because that figures into when those that makes these decisions on what to buy will look at every angle including costs just to run them (Power bill).
Between a lower cost to run and the fact they run cooler too matters due to the need for IT/server rooms to be air conditioned in most cases so they don't overheat and the less heat they have to deal with, the cheaper that cost will be.

I used to maintain the AC unit for my former employer's IT/Server room to know that.
They look at everything along lines of cost. 

Using my former employer as an example:
They consider costs of it to the point once they had chosed a fire suppression system to use, they had me personally to go in and prep the entire room by literally sealing the ceiling tiles and walls completely, so NO air in it could vent up and out due to fire suppression requirements of the gas-based fire extingushing system they chose for that room.
No, it wasn't a Halon based system - Halon hasn't been used for awhile for new system installs because better gases have been developed since Halon was the standard - IIRC, even the rules these days require a different gas type for brand new fire suppression setups.
The doors themselves had to seal up when closed to help hold the extinguishing gas if it was released in the room and yes, it was tested and work was done until the room passed the tests.

That took some time and testing to do for getting it right - Wasn't fun to do either.

There were certain fire safety rule requirements that had to be met and part of the cost to pass those tests were a part of their choice too, so it boils down purely to cost in the end.
 

[starsmine]

come back and ask again when 18A is publicly released. 

[Nat4Sail]

Intel isn't going anywhere, especially with aggressive tariffs likely on the horizon for USA. I've been buying their stock last few years, so maybe I'm drinking too much Kool-aid 

[Blue4130]

3 hours ago, jordanbuilds1 said:

what makes you think that companies care about that, they just choose the first option they see and keep it pushing.

It's been a while since I was in IT, but I can assure you that when my previous company was upgrading, we didn't just choose the first option. There were months of meetings and analysis before choosing the infrastructure. We were a cross (western mostly) canada company with 500 employees in many cities. The purchase was over $1,00,000 and higher ups wanted to be damn sure that we were getting the right gear for the job.

[starsmine]

6 minutes ago, Nat4Sail said:

Intel isn't going anywhere, especially with aggressive tariffs likely on the horizon for USA. I've been buying their stock last few years, so maybe I'm drinking too much Kool-aid 

that will mean a lot more IFF 18A works as intended. 

I do think it will, but its not not a sure thing. 

[GOTSpectrum]

4 hours ago, Satan_Prometheus said:

I don't think it really has anything to do with "fanboys" since AFAIK the DIY market is a pretty small percentage of the overall market for computer chips. I would think that corporate clients (datacenter clients and large prebuilt OEMs) are significantly more important, though if somebody has data suggesting otherwise that would be cool to see!

The biggest market for x86 is laptops and servers, desktops, even OEM desktops from the likes of Dell, are a rather small percentage of the market. 

But you are correct that OEMs dictate a lot of the market, both in desktops, especially in laptops and even in the HPC space. There were only limited options for EPYC first gen, so it only worked for part of the market. These days EPYC is not quite the default but in terms of dollars it's about 1/3 of the market, and in terms of the number of units it's 27%. This means that AMD are selling higher-priced chips on average. Compared to Intel, it seems Intel either has to offer much lower prices or is only covering the lower end of the market today. 

[GOTSpectrum]

7 minutes ago, starsmine said:

that will mean a lot more IFF 18A works as intended. 

I do think it will, but its not not a sure thing. 

I fear yields for 18A, ribbonFET and PowerVia are still new and will take some time to get right. The same for TSMCs GAAT node in development too 

[wasab]

Making chips in house has advantages and disadvantages. They are great and give them a leg up when they are technologically ahead. If they are behind, their fabs are nothing more than cash burning liability. 

[Monkey Dust]

On 11/21/2024 at 8:26 PM, GuiltySpark_ said:

No, they buy what they know. 

 

"Nobody Gets Fired For Buying IBM Intel"

 

 

Yep, never underestimate how risk-adverse the corporate world is. The most important aspect of any job is covering your own ass. If you make the wrong call on something, you really want (almost) everyone else, to have made the same call.

On 11/21/2024 at 11:32 PM, Nat4Sail said:

Intel isn't going anywhere, especially with aggressive tariffs likely on the horizon for USA. I've been buying their stock last few years, so maybe I'm drinking too much Kool-aid 

All the expensive chips are made in Taiwan, so I'd have thought AMD & Intel would be impacted equally?

 

I expect Intel will make a comeback at some point. They are still very dominant in the laptop market, which dwarfs the desktop market. AMD being either unable, or uninterested, in getting their chips in more laptops. They aren't as far behind as AMD was when they made their comeback.

 

The most significant risk to Intel isn't AMD, it's ARM, and maybe RISC V. If Window ARM compatibility improves, x86 could be rendered pretty niche.

[GreenLeaf]

with the no politics rule in mind, i live 3km from the intel fab Linus toured like 2 to 3 years ago in israel

i have to say, they have been under some heavy construction over there but recently they kind of stopped.

i also heard from a friend of my dad that a lot of "QC vendors" have left the company, idk if that's true or not but regardless i hope they come back cause no competition is always bad 

[ImorallySourcedElectrons]

On 11/21/2024 at 5:49 PM, SS451 said:

They make their own chips in house which might be what saves them but that is arguably what doomed them in the first place. 

How did it doom them? They're having the same rough ride as all the other electronics companies at the moment. AMD is somewhat insulated because they have a fabless model, but at the same time they're heavily exposed to the ups and downs of the manufacturing part of the industry, meanwhile Intel has some leverage there.

 

On 11/21/2024 at 5:56 PM, jordanbuilds1 said:

if battlemage sucks, its over for intel, but its looking like battle mage is gearing up to be a good release.

Just how? Intel's GPU developments are closely related to their "AI accelerator" cards for datacentre businesses (which is a worth while business for them), give them a stronger position on the contract-design SoC market, and their integrated graphics on their GPUs. GPU development made all the sense in the world for them given that they were already 90% of the way there.

 

On 11/21/2024 at 5:57 PM, jordanbuilds1 said:

which are now going to snapdragon

Qualcom and Intel aren't really competing in the same markets most of the time.

 

On 11/21/2024 at 5:59 PM, GOTSpectrum said:

or if they can finally get back to competing on node development in regards to power efficiency and density. 

A lot of the "Intel is behind" is being spun out by investment funds trying to get a big cash-out on Intel divesting its fab business, please stop spreading this narrative. Technically speaking, as electronics engineer, Intel has consistently been ahead of Samsung and TSMC at densities they can achieve for random logic, it's just that you almost never use this density in actual chips except for highly repetitive structures like memory. TSMC is "ahead" in cost models because they get a massive amount of support from the Taiwanese government, and Samsung represents a significant percentage of South Korea's GDP and basically has the spending and borrowing capacity of many countries. A lot of Intel's disadvantages are addressed in the CHIPS act, but the US government dragging its feet is hurting them.

 

On 11/21/2024 at 6:12 PM, ToboRobot said:

Intel certainly has a lot of talent, that could be used to create awesome products to return to it's former glory.

The economics and politics of this may be more important than the engineering.

The business will continue to exist, maybe a new name, maybe in different parts, but for the most part Intel will continue to exist as chips are still required.

One possibility that I haven't seen discussed much, is Intel could "buy" (to protect it's x86 IP) another company like Broadcom or Qualcomm to be more competitive.

Another possibility, could be nVidia wanting an x86 licence, and opening up CUDA cores for the other x86 licensees, so AMD and Intel would standardize GPU cores with the CUDA architecture, and nVidia would get an x86 license so they can build full stack AI products.  Would be interesting to see Intel and AMD CPUs compete with CUDA based iGPUs. 

Intel buying Broadcom would be bad, Broadcom is into a lot of highly specialized markets (Broadcom = Avago = Agilent = HP Semiconductor), they do really exotic things like manufacture high precision position encoders for the semiconductor industry, analog fiber optic modules for industrial high voltage isolation, etc. It'd require Intel to take on a massive product portfolio that would require them to keep on most of the staff Broadcom already has with little or no ability to optimize. Broadcom is a really weird company, and the entire Broadcom/Avago merger/buyout was one of the sketchiest thing that happened in the electronics industry over the last two decades.

 

On 11/21/2024 at 6:13 PM, jordanbuilds1 said:

intel lost alot with the release of alchemist, and with these core ultra chips i dont think it would truely help buisness now.

I wouldn't say they lost a lot with it, folks keep saying their GPUs are cancelled every month, and yet it looks like they were so unsuccessful that battlemage is getting released in the coming weeks/months.

[GOTSpectrum]

23 minutes ago, ImorallySourcedElectrons said:

A lot of the "Intel is behind" is being spun out by investment funds trying to get a big cash-out on Intel divesting its fab business, please stop spreading this narrative. Technically speaking, as electronics engineer, Intel has consistently been ahead of Samsung and TSMC at densities they can achieve for random logic, it's just that you almost never use this density in actual chips except for highly repetitive structures like memory. TSMC is "ahead" in cost models because they get a massive amount of support from the Taiwanese government, and Samsung represents a significant percentage of South Korea's GDP and basically has the spending and borrowing capacity of many countries. A lot of Intel's disadvantages are addressed in the CHIPS act, but the US government dragging its feet is hurting them.

This isn't even remotely true. 

Even Intel admitted they were behind, which is why the latest Intel chips have their cores fabbed by TSMC. Intel made a big deal about 5 nodes in the timespan of four years, advertising it as a "catch up and then take over" strategy. 

 

Intel has been behind TSMC on power efficiency and density, Intels 10nm(Intel 7) which is comparable to TSMCs N7 process was late, TSMC had started HVM in 2017. Intel didn't get 10nm(Intel 7) in full-scale production until 2019, a full two years later. Even then it was actually less power efficient than their own 14nm++ node. It was however denser, but yield rates were woeful and Intel decided to power ahead with Intel 7 being the main node their fabbed on for consumer hardware. Intel 7 is actually an optimised 10nm node, previously planned to be called 10ESF(10nm+ publicly). 

TSMC is the market-dominant power for a reason, and it isn't for price reasons. TSMC is now the most expensive Fab to contract, but people stick with them cause they concentrate on yields and power. Where intel concentrates on high power, high performance nodes. 

Intel is just behind, they are so behind in fact that they would rather TSMC fab their main fab instead of their own. Intel has ordered 14 billion of TSMC fabbed 3nm chips in 2024 alone, making them the second largest 3nm customer (after apple) and moving AMD into third 

[smcoakley]

1 hour ago, ImorallySourcedElectrons said:

Just how? Intel's GPU developments are closely related to their "AI accelerator" cards for datacentre businesses (which is a worth while business for them), give them a stronger position on the contract-design SoC market, and their integrated graphics on their GPUs. GPU development made all the sense in the world for them given that they were already 90% of the way there.

As someone who works on could software, "AI compute" hardware is huge right now, but Nvidia is the undisputed king in the market. There's very little competition from anyone else at the moment and I don't really have much hope for Intel in this market. But good on them if they want to try; I'd go after the cost angle, because the Nvidia stuff is X-pen-SIVE.

[ImorallySourcedElectrons]

2 minutes ago, GOTSpectrum said:

This isn't even remotely true. 

Even Intel admitted they were behind, which is why the latest Intel chips have their cores fabbed by TSMC. Intel made a big deal about 5 nodes in the timespan of four years, advertising it as a "catch up and then take over" strategy. 

 

Intel has been behind TSMC on power efficiency and density, Intels 10nm(Intel 7) which is comparable to TSMCs N7 process was late, TSMC had started HVM in 2017. Intel didn't get 10nm(Intel 7) in full-scale production until 2019, a full two years later. Even then it was actually less power efficient than their own 14nm++ node. It was however denser, but yield rates were woeful and Intel decided to power ahead with Intel 7 being the main node their fabbed on for consumer hardware. Intel 7 is actually an optimised 10nm node, previously planned to be called 10ESF(10nm+ publicly). 

Sorry, but you're just wrong on the density aspect, just plain wrong. You're mixing up namings with actual densities, this is a stupid marketing game TSMC started and I have no clue why folks are so willing to go along with it, go and check the PDKs. But Intel has consistently been able to push higher densities for random logic, for repeating patterns things get a bit more complicated and I think it's fair to say most of the major players (TSMC, Intel, Samsung, SK Hynix, and Micron) are hitting about the same targets if we're not considering functional devices. However, no one actually pushes those limits except for very specific things, in practice you're limited by thermal design constraints when you're designing an actual logic chip and which technology offers the best density at that sweet spot is very difficult to answer. And on power efficiency neither TSMC or Intel "leading logic process" are the leaders in that aspect, but that's because the general logic processes never really target that in the first place. My money for that one would be on an older TSMC or Samsung process node.  Furthermore, each process node has some peculiar advantages versus the other, like some of them are way better to implement linear electronics (RF, voltage regulators, amplifiers for temperature sensors, ...) than the other's. But I shall refer to an earlier post of mine: 

 

As to Intel's public statements, investor speak is very different from engineering speak. Technical conferences tend to have a very different tone.

 

7 minutes ago, GOTSpectrum said:

TSMC is the market-dominant power for a reason, and it isn't for price reasons. TSMC is now the most expensive Fab to contract, but people stick with them cause they concentrate on yields and power. Where intel concentrates on high power, high performance nodes. 

Intel is just behind, they are so behind in fact that they would rather TSMC fab their main fab instead of their own. Intel has ordered 14 billion of TSMC fabbed 3nm chips in 2024 alone, making them the second largest 3nm customer (after apple) and moving AMD into third 

No, it's extremely common in the electronics industry to go to someone else's fab. I don't get why folks get their panties in a twist over this as evidence of a company's failing. Can be related to production scheduling, could be for fab construction works or availability, could be that a specific feature simply worked better in another company's technology, etc. And TSMC's market dominance is due to their funding structure, if they had to compete fairly they wouldn't be doing nearly as well. There's a reason why TSMC is making a lot of virtue signaling moves to hold off antitrust action from both the EU and US.

 

And TSMC is not the most expensive fab to contract by a long run. If you want to see scary fab quotations, have a look at some of the boutique MEMS and integration fabs where the expected yield is less than 1%. And if you wonder how those survive: military and medical low volume orders.

[ImorallySourcedElectrons]

5 minutes ago, smcoakley said:

As someone who works on could software, "AI compute" hardware is huge right now, but Nvidia is the undisputed king in the market. There's very little competition from anyone else at the moment and I don't really have much hope for Intel in this market. But good on them if they want to try; I'd go after the cost angle, because the Nvidia stuff is X-pen-SIVE.

They are pushing it: https://www.intel.com/content/www/us/en/products/details/processors/ai-accelerators/gaudi.html 

 

I'd love to see the actual market share numbers, but given that companies tend to bulk buy these things when building server farms its probably hard to tell for industry analysts since there are probably no in-between suppliers/distributors.