News Firm Estimates Intel's GPU Unit Losses at $3.5 Billion, Suggests Sell Off

[InvalidError]

My goodness, an analyst who has no idea that investing in a new product requires spending money, then releasing the product, and only then making money form the product?

The most retarded part is analysts not considering the fact that over 60% of PCs out there use Intel IGPs. No graphics, no IGP, goodbye 40-50% of Intel's CCG income going straight into AMD's arms. That is many more billion dollars than claimed graphics losses.

The graphics division "losses" are basically a tax cheat or "creative accounting" thing.

 

[digitalgriffin]

Intel's long term survival depends on two things:

1. Fixing their foundry issues and staying in step with TSMC

2. Moving into data server AI. x86 architecture in data servers is losing relevance. Servers are where a lot of profits come from.


So Intel has to invest in data server AI acceleration hardware if they like it or not; user GPU's with low margins, not so much.

 

[TerryLaze]

I don't know, but they weren't wrong about the economics. The key point you're glossing over is IFS, which is the first step towards Intel spinning off its fabs, like AMD did almost 15 years ago. I'm not saying it's 100% going to happen, but the fabs are indeed a boat anchor around Intel's neck.

If intel wanted to sell off their FABs why would they be building even more? It's an expansion to their business just like ARK is, if they wanted to get rid of the FABs they would downsize that department, make it more viable for a potential buyer.

 

[zipspyder]

I'm hearing rumors there might be issues on the hardware side as well. It crops up in the minimum framerate tests. First gen always has teething pains...hope they don't cancel.

 

[TerryLaze]

Intel's long term survival depends on two things:

1. Fixing their foundry issues and staying in step with TSMC

2. Moving into data server AI. x86 architecture in data servers is losing relevance. Servers are where a lot of profits come from.


So Intel has to invest in data server AI acceleration hardware if they like it or not; user GPU's with low margins, not so much.

1. Why?! Because otherwise all the people that need x86 PCs are going to stop needing them? Even if all of TSMCs output of relevant archs would be used for making AMD CPUs (which would never happen) ,how much of the market would they be able to supply?
2. Intel has AI on their iGPUs for like 5 years now, I don't know about datacenter but it stands to reason that they would have it there as well.

Low margin stuff pays for the high end stuff, all the celerons and pentiums make it worth it to make CPUs, optane failed because the customer level stuff wasn't selling enough.
For the GPU department maybe the iGPUs will be enough to keep the high end stuff going but cheap dGPUs that sell heaps will help even more.

 

[digitalgriffin]

1. Why?! Because otherwise all the people that need x86 PCs are going to stop needing them? Even if all of TSMCs output of relevant archs would be used for making AMD CPUs (which would never happen) ,how much of the market would they be able to supply?
2. Intel has AI on their iGPUs for like 5 years now, I don't know about datacenter but it stands to reason that they would have it there as well.

Low margin stuff pays for the high end stuff, all the celerons and pentiums make it worth it to make CPUs, optane failed because the customer level stuff wasn't selling enough.
For the GPU department maybe the iGPUs will be enough to keep the high end stuff going but cheap dGPUs that sell heaps will help even more.

1. It's all about the TCO. If competitors turn to other solutions it's because Intel is not delivering a cost effective solution. After you lose a certain percentage of market share, your influence over said market wanes. And intel is losing server market share. You know this, I know this. Why bury your head like an Ostrich? Ignoring the issue will only make it worse. One of the ways to maintain this is by maintaining high performance/$ for threaded loads, and highest performance possible for non threaded loads. This means smooth technological and economic advancements at the foundry level. Remember what the margins were like on the Bulldozer fiasco? Abysmal! Sure you can sell them at a reduced rate, but what will it cost you if the node and design is not competitive? GloFo was part of the problems holding AMD back.

2. Intel just doesn't need AI. They need PERFORMANT & COST COMPETITIVE AI. The most costly part of AI is the training. Google is now working on systems with hundreds of millions of inputs. They are aiming for over 1 Billion inputs soon. The more data a training set has, the more intelligent it becomes for all sorts of AI workloads, from image recognition/inference/path generation/nature based and more.

I suggest you read "The tipping point" It talks about how trends and companies can easily shift dominance.

 

[TerryLaze]

1. It's all about the TCO. If competitors turn to other solutions it's because Intel is not delivering a cost effective solution. After you lose a certain percentage of market share, your influence over said market wanes. And intel is losing server market share. You know this, I know this. Why bury your head like an Ostrich? Ignoring the issue will only make it worse. One of the ways to maintain this is by maintaining high performance/$ for threaded loads, and highest performance possible for non threaded loads. This means smooth technological and economic advancements at the foundry level. Remember what the margins were like on the Bulldozer fiasco? Abysmal! Sure you can sell them at a reduced rate, but what will it cost you if the node and design is not competitive? GloFo was part of the problems holding AMD back.

2015 2016 Data Center Group 3,573 3,718
2017 2018 Data Center Group 4,026 5,100
2019 2020 Data Center Group 4,553 6,181
2021 2022 Data Center Group 5,547 4,649
These are all q2 numbers since q2 2022 was a terrible quarter for intel.
Since you brought up bulldozer, in 2015 and 2016 intel made below 4bil in data center.
In 2017 when zen came out intel made 4bil, in 2018 since people waited to buy to see what zen would bring but finally where disappointed by the overhyped zen intel made 5bil, then in 2019, still before covid it dropped back to a more normal 4,5bil, then in 2020 it exploded to 6,2bil in 2021, back to 5,5, and now in 2022 it's at 4,6bil the same point it was in 2019 right before covid.
So explain to me like I'm five, how exactly did intel lose market share and where to did it lose it?
Excluding the two years of covid pumped up numbers intel is at the same point it was before.

If AMD is gaining market share then that's great but if they only sell to businesses that wouldn't buy intel anyway then how is intel losing market share?

Intel isn't losing market share they are losing potential growth market and even then it's still a question if intel want's those markets.

 

[digitalgriffin]

2015 2016 Data Center Group 3,573 3,718
2017 2018 Data Center Group 4,026 5,100
2019 2020 Data Center Group 4,553 6,181
2021 2022 Data Center Group 5,547 4,649
These are all q2 numbers since q2 2022 was a terrible quarter for intel.
Since you brought up bulldozer, in 2015 and 2016 intel made below 4bil in data center.
In 2017 when zen came out intel made 4bil, in 2018 since people waited to buy to see what zen would bring but finally where disappointed by the overhyped zen intel made 5bil, then in 2019, still before covid it dropped back to a more normal 4,5bil, then in 2020 it exploded to 6,2bil in 2021, back to 5,5, and now in 2022 it's at 4,6bil the same point it was in 2019 right before covid.
So explain to me like I'm five, how exactly did intel lose market share and where to did it lose it?
Excluding the two years of covid pumped up numbers intel is at the same point it was before.

If AMD is gaining market share then that's great but if they only sell to businesses that wouldn't buy intel anyway then how is intel losing market share?

Intel isn't losing market share they are losing potential growth market and even then it's still a question if intel want's those markets.

Raw numbers are fine but they aren't the whole picture. The fact is they are losing % market share. And this past year is looking to be a disaster. They are getting replaced by custom ARM CPU designs for big corporate customers like Amazon, NVIDIA for AI, and AMD for legacy x86 EPYC. Apple has abandoned them as well. They are losing market share % despite the raw numbers. EPYC data servers have doubled their market penetration. While the number is still small, that's still a pretty significant penetration. AMD just offers a lower TCO with more cores on one package, and an AMAZING number of PCIe lanes which is significant for data storage access.

If you work for Intel, I can see why they are failing.

 

[-Fran-]

2015 2016 Data Center Group 3,573 3,718
2017 2018 Data Center Group 4,026 5,100
2019 2020 Data Center Group 4,553 6,181
2021 2022 Data Center Group 5,547 4,649
These are all q2 numbers since q2 2022 was a terrible quarter for intel.
Since you brought up bulldozer, in 2015 and 2016 intel made below 4bil in data center.
In 2017 when zen came out intel made 4bil, in 2018 since people waited to buy to see what zen would bring but finally where disappointed by the overhyped zen intel made 5bil, then in 2019, still before covid it dropped back to a more normal 4,5bil, then in 2020 it exploded to 6,2bil in 2021, back to 5,5, and now in 2022 it's at 4,6bil the same point it was in 2019 right before covid.
So explain to me like I'm five, how exactly did intel lose market share and where to did it lose it?
Excluding the two years of covid pumped up numbers intel is at the same point it was before.

If AMD is gaining market share then that's great but if they only sell to businesses that wouldn't buy intel anyway then how is intel losing market share?

Intel isn't losing market share they are losing potential growth market and even then it's still a question if intel want's those markets.

I'll point out the obvious though: the "market" is not fixed in size. The same Data Center can double it's server farm capacity via expansion and your 100% supply could be reduced to 50% if they went with another vendor instead just by counting CPUs without reducing your amount sold. There's also the details of margin and inflation to take into account. 4.5B now is clearly "less" than 4.5B in 2019. Much like the market, money is not "fixed".

Regards.

 

[bigdragon]

Got to love those Wall Street analysts. They'll do or say anything that will make a short-term buck. Yes, Intel selling off its graphics division would stop the bleeding and make a quick cash infusion, but it would also do unrecoverable long-term damage while putting money into the hands of competitors. The downsides are far too negative. Intel should keep spending money to make money (in the future). Nvidia, ATi, 3DFX, and others didn't get things perfect the first several products either.

Reading JPR's recommendations reminds me of business parasites from grad school. Get into a company, strip its assets, sell it off, and get a big fat parachute. Rinse and repeat. I prefer it when "leaders" build something instead of stripping it for components.

 

[TJ Hooker]

You're firing the coach of a sports team because the star player decided to have a terrible season. You can be the greatest strategist in the world, but in the end it is up to the players/engineers to actually do the work successfully.

Poor analogy (for the point you're trying to make), because firing a coach after a (really) bad season isn't unusual. Because if you have a team full of good players that still somehow completely fails to perform, that points to a leadership issue.

 

[Lorien Silmaril]

Except the endurance turned out to be not that incredible. That meant it could't be naively used as a DRAM substitute. Instead, you had to use it as a lower memory tier, which wasn't well supported by operating systems or software and didn't offer quite the same benefits in random access performance for massive datasets.


I'm sure that's not what killed it.


If the tech had untapped potential ever to be viable, they would've found a buyer for the IP.

I am reminded of how they screwed over Micron, during the breakup. They basically kept Micron from retaining some key IP and personnel to ever make it viable. Not sure if that was just Intel's legal department going rogue or what, but it seemed in bad faith.

doubt it's just their legal. Intel's playbook never changed: monopolize the market or sabotage the competition - even potential ones who were former businesses partners.

 

[overlord]

Would've been cheaper to buy Imagination Technologies Group than follow Raja's delusional money pit. This has a whiff of S3's dire Savage 2000 💩 S3 never recovered from the mess, wasn't long after Raja bailed on them joining ATI, who later became Radeon Technology Group! 😱

 

[Lorien Silmaril]

I actually asked you what the direct impact of a Senior VP has on driver development. I don't know the answer to that, and the answer you gave doesn't convince me he deserves 100% of the blame and the only person being held responsible for Intel's crappy GPU drivers. He doesn't deserve 0%, but the higher you get up in management above the actual developers, the less blame I would put on the person. Like I said above, no amount of strategy and cheer leading and synergy meetings is going to make up for incompetency by the actual coders.

Leaders are always the one responsible fire projects that go wrong. Always.

If your low level employees (coders in this case) are bad, it is a leader's job to replace them. If there are organizational issues causing project delays, you fix it, or at least draft a plan and delegate.

True, a VP wouldn't manage the execution of low level details - nobody expects Raja to know any details about how actual code is written at Intel. But it is his job to hire good middle level managers that do know such details, and to keep projects on track. How the coding teams execute is up to the lower level managers, but a good VP maintains enough oversight to at least know if projects are on time and in good quality.

It sounds like Intel has an entrenched culture for lack of transparency at every level, so everyone lies about the state of things and the higher-ups either don't pay attention, or lie themselves to cover their own ass.

 

[shady28]

Which is missing the point that it doesn't make sense to lay the blame for this mess at his feet. What exact effect does a Senior VP have on driver quality? <snip>

Raja was part of the outsourcing of it to India. He's responsible for the results.

 

[shady28]

Leaders are always the one responsible fire projects that go wrong. Always.

If your low level employees (coders in this case) are bad, it is a leader's job to replace them. If there are organizational issues causing project delays, you fix it, or at least draft a plan and delegate.

True, a VP wouldn't manage the execution of low level details - nobody expects Raja to know any details about how actual code is written at Intel. But it is his job to hire good middle level managers that do know such details, and to keep projects on track. How the coding teams execute is up to the lower level managers, but a good VP maintains enough oversight to at least know if projects are on time and in good quality.

It sounds like Intel has an entrenched culture for lack of transparency at every level, so everyone lies about the state of things and the higher-ups either don't pay attention, or lie themselves to cover their own ass.

Nailed it.

I don't know where the mentality that leaders are not responsible for the performance of their organization is coming from, but it's bunk.

The leaders at that level can fire the people below them if they are not performing well and get new people. They can outsource the operation if they decide that's best. They get a pass for two or three years living with their predecessor's decisions, but then they own the results. Kodura has been with Intel for 5 years.

Same goes for Gelsinger, who is 18 months in. If Kodura is performing poorly, fire him before the new CEO grace period is over, or get fired.

 

[TerryLaze]

doubt it's just their legal. Intel's playbook never changed: monopolize the market or sabotage the competition - even potential ones who were former businesses partners.

AMD also partnered with intel for a laptop cpu with amd graphics, should amd have given intel their gpu IP just because of that?!
No company ever gives away their IP unless they are forced to, you never know if it will become relevant in the future again.

Leaders are always the one responsible fire projects that go wrong. Always.

If your low level employees (coders in this case) are bad, it is a leader's job to replace them. If there are organizational issues causing project delays, you fix it, or at least draft a plan and delegate.

True, a VP wouldn't manage the execution of low level details - nobody expects Raja to know any details about how actual code is written at Intel. But it is his job to hire good middle level managers that do know such details, and to keep projects on track. How the coding teams execute is up to the lower level managers, but a good VP maintains enough oversight to at least know if projects are on time and in good quality.

It sounds like Intel has an entrenched culture for lack of transparency at every level, so everyone lies about the state of things and the higher-ups either don't pay attention, or lie themselves to cover their own ass.

You are going with the assumption the gaming GPUs are a high priority for intel right now or that 3.5bil is a lot of money for intel...
Intel is a huge software house with a lot of projects that are far more important than gaming GPUs and they are an even larger hardware house with many projects that are far more important than gaming GPUs.
Just as an perspective, since intel is using tsmc for their gaming GPUs, amd made 3,530 bil in revenue in the last 6 month from their gaming department and AMD is getting more wafers than intel, I have no idea how it gets split between cpu and gpu though.
The point is that intel is going to sell their GPUs at a lower price so lower margin than amd is and AMD only made 545 mil from gaming in the last 6 months, that's not an amount of money that intel would do any extra work for.

Intel is not going to take away resources from more lucrative projects for the gaming gpus.

Sure they could have delt with the whole matter of announcing and delaying it in a much more delicate way instead of the terrible job they did but that's a different argument.

 

[-Fran-]

Leaders are always the one responsible fire projects that go wrong. Always.

If your low level employees (coders in this case) are bad, it is a leader's job to replace them. If there are organizational issues causing project delays, you fix it, or at least draft a plan and delegate.

True, a VP wouldn't manage the execution of low level details - nobody expects Raja to know any details about how actual code is written at Intel. But it is his job to hire good middle level managers that do know such details, and to keep projects on track. How the coding teams execute is up to the lower level managers, but a good VP maintains enough oversight to at least know if projects are on time and in good quality.

It sounds like Intel has an entrenched culture for lack of transparency at every level, so everyone lies about the state of things and the higher-ups either don't pay attention, or lie themselves to cover their own ass.

Nailed it.

I don't know where the mentality that leaders are not responsible for the performance of their organization is coming from, but it's bunk.

The leaders at that level can fire the people below them if they are not performing well and get new people. They can outsource the operation if they decide that's best. They get a pass for two or three years living with their predecessor's decisions, but then they own the results. Kodura has been with Intel for 5 years.

Same goes for Gelsinger, who is 18 months in. If Kodura is performing poorly, fire him before the new CEO grace period is over, or get fired.

The VP in charge of drivers was never Raja, but Lisa Pearce. Raja, no matter how much you guys hate AMD or him, is not responsible for the driver mess. He can be responsible for many things, but not that one. He can't also freely point fingers because when you're under the same umbrella, throwing shade to the side is always a bad look from the outside.

I am pretty sure there's some tough conversations happening inside of Intel around this, but this is not Raja's exclusive mess, or even Pat's or Lisa's. This is one of the things, I'm pretty sure, Pat Gelsinger is trying really hard to change.

Regards.

 

[shady28]

The VP in charge of drivers was never Raja, but Lisa Pearce. Raja, no matter how much you guys hate AMD or him, is not responsible for the driver mess. He can be responsible for many things, but not that one. He can't also freely point fingers because when you're under the same umbrella, throwing shade to the side is always a bad look from the outside.

I am pretty sure there's some tough conversations happening inside of Intel around this, but this is not Raja's exclusive mess, or even Pat's or Lisa's. This is one of the things, I'm pretty sure, Pat Gelsinger is trying really hard to change.

Regards.

"Raja M. Koduri[1] ...is currently the chief architect and senior vice president of Intel's architecture, graphics and software (IAGS) division. "

Pearce works for Koduri. If she can't do the job he should have fired her 3-4 years ago.

So what's next. Just fire the programmers, after all they are the only ones that know the code? Give the boss a break, right.

Nonsense.

 

[mavroxur]

If Intel ever sells their graphics division I will never buy another one of their products. We absolutely need a third player in the market. The first round never goes well, and I have faith they’ll get it right the second time just like AMD RDNA 2.

Man, that'll really teach Nvidia and AMD a lesson, abandoning the new player and going back to them

LOL

 

[-Fran-]

"Raja M. Koduri[1] ...is currently the chief architect and senior vice president of Intel's architecture, graphics and software (IAGS) division. "

Pearce works for Koduri. If she can't do the job he should have fired her 3-4 years ago.

So what's next. Just fire the programmers, after all they are the only ones that know the code? Give the boss a break, right.

Nonsense.

No, because that's not how it works going up in the reporting lines. While I do agree that the head does take the responsability at the end, it's not clear cut when things get delayed, don't work or just aren't reported correctly. It could be they'll find that their internal processes just don't work well enough for the driver writers, or the middle management is too fat and doesn't communicate upstream well enough or that Raja doesn't look over the shoulders of his subordinates enough to know things aren't working.

This is not even me trying to defend the guy, but just trying to make everyone here blaming all of this kerfuffle on Raja (or Pat) because he's from AMD that it's a stupid oversimplification of a clearly more complicated issue. Firing Raja and putting Jensen Huang in that position won't magically make the engineers deliver faster or develop better drivers without reestructuring the whole division (which Pat is doing, as I understand it) and that would take years. So, as a kneejerk reaction, blaming anyone particular at this point is kind of moot and a very bad idea to fire them. They need to sort out the problems first, fix them and then based on what their internal audits show, then take action. And, as I said, I'm pretty sure Raja is not the only one that should get a good slap. Lisa taking ownership of the drivers development is no secret and it was very public. Raja even tweeted about it wishing her good luck. Also, Lisa has been with Intel for 25 years, just saying.

Regards.

 

[Darkbreeze]

He doesn't design the hardware at Intel, either. He hired in as a Senior VP, and was recently promoted to Exec VP. The people who write drivers work for him.

Which, kind of makes him responsible for everything that happens below him, including hiring or continuing to employ those who are clearly not getting the job done. NFL coaches and personnel directors don't actually score touchdowns either, but if they aren't bringing the right people in to get that done, they don't usually last too long. I think it's appropriate to apply that here as well. Especially since the product lines he WAS a part of at AMD were some of the least competitive, hottest operating products AMD every produced. To me he simply is a half baked con man. Smart, sure, but competent enough to ensure a successful product development, doesn't seem like it.

But, as you say, it might be a while before anything he had direct involvement with comes to fruition, so really it's hard to condemn or pardon too soon. Still, I have serious doubts when it comes to Koduri and bringing something successful to fruition. Then again, I'm no engineer, but you have to point the finger at somebody.

 

[spongiemaster]

It’s called Managment 101. The boss is always to blame.

Sure. Just like in sports, you can't fire the whole team so you fire the coach to "change the culture in the locker room," and typically that has zero long term affects on the team performance. Michael Jordan will make any basketball coach look like a genius, and a crappy team will make any coach look like an incompetent idiot. The boss gets blamed because it is easier to blame everything on one person and get rid of them, not necessarily because it makes any sense.

 

[Darkbreeze]

If you can't bring the right people in after being given several (Five) years to do so, then in tech or in sports, you deserve to get a green slip.

 

[spongiemaster]

If you can't bring the right people in after being given several (Five) years to do so, then in tech or in sports, you deserve to get a green slip.

What's a green slip? A bigger paycheck? You mean pink slip?

You think Raja personally hires every employee that works below him? Somehow, I doubt that's the case. Also, just because you know you need people with specific skill sets, doesn't mean there's a pool of all stars sitting around looking for jobs. Ati/AMD has had awful drivers for 20+ years. If it was a simple case of hiring the right people, why has it taken them so long to get to their current workable position? It's amazing how easy everything is to people on the internet.

No one here really knows what's going on behind the scenes. I would put money on Intel having a better idea what the root issues are with the driver development team than the message board CEO's and management 101 graduates here. Pat Gelsinger promoted Raja a few months ago. By most accounts, Gelsinger is no idiot. Or maybe he should be fired, since he truly is the man at the top and responsible for everything that happens at Intel. Looks like he needs to do a better job hiring driver developers.