Intel - Don't call it a come back

[Brian_B]

‘Intel is back:’ New CEO’s plan to make chips for other companies excites investors

Intel said it would double down on manufacturing, and invest $20 billion in two new chip factories in Arizona. "

www.cnbc.com

At least not yet. They have a plan - and that's good. But plans are only as good as their execution.

There was also an article over at Tom's stating that Intel had "fixed" their 7nm process. Which is good. They expect to be starting commercial production in 2023. But remember when it was delayed and called broken - they expected to get it online by 2022/23... so it really isn't any improvement in any timelines - although investors seem to be taking it that way, and Intel seems to be spinning it as positively as they can. And I can remember when Intel "fixed" their 10nm process, but it's still broken today, so ... the proof will be in the pudding.

Color me cautiously optimistic.

 

[Stoly]

‘Intel is back:’ New CEO’s plan to make chips for other companies excites investors

Intel said it would double down on manufacturing, and invest $20 billion in two new chip factories in Arizona. "

www.cnbc.com

At least not yet. They have a plan - and that's good. But plans are only as good as their execution.

There was also an article over at Tom's stating that Intel had "fixed" their 7nm process. Which is good. They expect to be starting commercial production in 2023. But remember when it was delayed and called broken - they expected to get it online by 2022/23... so it really isn't any improvement in any timelines - although investors seem to be taking it that way, and Intel seems to be spinning it as positively as they can. And I can remember when Intel "fixed" their 10nm process, but it's still broken today, so ... the proof will be in the pudding.

Color me cautiously optimistic.

Not to mention that there are no current intel products based on 7nm.
Seeing is believing, so I'll give them the benefit of the doubt.

Gotta say that nowadays it may be much more difficult to execute a "tic toc" strategy given how much more complex is migrating to smaller nodes.
BTW I didn't see any mention of going massive cores to go head to head with threadripper.

 

[LazyGamer]

I see a new CEO reading the talking points. To investors this at least looks like business as usual, and for Intel, that's a positive indicator if only a shallow one.

And I can remember when Intel "fixed" their 10nm process, but it's still broken today, so ... the proof will be in the pudding.

I was actually just issued a laptop from work with an Intel 10nm CPU... which is proof that they're shipping stuff and not much more. The bigger issue is that Intel isn't making these with more than four cores so far as I can tell and that limits their reach!

Not to mention that there are no current intel products based on 7nm.
Seeing is believing, so I'll give them the benefit of the doubt.

This is going to get rough. I've let a lot of the noise on the Apple M1 release die down and have just started to take a look, and realistically speaking, I think I'd almost be happier with one of those for 'just a laptop' than anything Intel is capable of producing today, and Apple ain't slowing down to wait. By 2023 I'd expect anything mobile I have to be Apple the way things are going.

Desktops and servers are another matter of course, but wouldn't you know, my work laptop replaced my work desktop. In years past I'd have been somewhat upset about that.

I'd never have guessed that I'd be sitting here saying this, but here I am, and realistically for productivity and for content creation (within consumer limits), what Intel does (or AMD) matters less and less!

Gotta say that nowadays it may be much more difficult to execute a "tic toc" strategy given how much more complex is migrating to smaller nodes.

As I understand it, there's a major 'hump' in terms of technology to get to the next wave of 'refinement' nodes. It might be EUV if I'm remembering correctly, but regardless, 7nm and smaller are opened up by implementing and then perfecting it.

Intel's problem seems to have been that they bet the farm on 'refining' to 7nm rather than investing and developing the technology that would take them directly there and beyond.

BTW I didn't see any mention of going massive cores to go head to head with threadripper.

Threadripper isn't special technologically speaking; neither is Zen, really. What's unique is that they exist and can actually be bought.

Outside of AMD proving that there's a market for such configurations there's nothing stopping Intel (...except Intel...) from releasing a product line to challenge Threadripper or even exceed it.

Of course, Threadripper can be compared to Xeons as well, at least in terms of products in the same broad segment of functionality. Cost and performance would almost certainly be in AMDs favor I expect and would only be mitigated if Intel felt that that market segment would be worth addressing. Realistically they may simply be content to let AMD do the work to supply that market segment given how small it is (cross-section of folks that need many more cores and connectivity but don't need the reliability, support, and SLAs that come with enterprise gear).

 

[Stoly]

I see a new CEO reading the talking points. To investors this at least looks like business as usual, and for Intel, that's a positive indicator if only a shallow one.


I was actually just issued a laptop from work with an Intel 10nm CPU... which is proof that they're shipping stuff and not much more. The bigger issue is that Intel isn't making these with more than four cores so far as I can tell and that limits their reach!

This is going to get rough. I've let a lot of the noise on the Apple M1 release die down and have just started to take a look, and realistically speaking, I think I'd almost be happier with one of those for 'just a laptop' than anything Intel is capable of producing today, and Apple ain't slowing down to wait. By 2023 I'd expect anything mobile I have to be Apple the way things are going.

Desktops and servers are another matter of course, but wouldn't you know, my work laptop replaced my work desktop. In years past I'd have been somewhat upset about that.

I'd never have guessed that I'd be sitting here saying this, but here I am, and realistically for productivity and for content creation (within consumer limits), what Intel does (or AMD) matters less and less!

As I understand it, there's a major 'hump' in terms of technology to get to the next wave of 'refinement' nodes. It might be EUV if I'm remembering correctly, but regardless, 7nm and smaller are opened up by implementing and then perfecting it.

Intel's problem seems to have been that they bet the farm on 'refining' to 7nm rather than investing and developing the technology that would take them directly there and beyond.

Threadripper isn't special technologically speaking; neither is Zen, really. What's unique is that they exist and can actually be bought.

Outside of AMD proving that there's a market for such configurations there's nothing stopping Intel (...except Intel...) from releasing a product line to challenge Threadripper or even exceed it.

Of course, Threadripper can be compared to Xeons as well, at least in terms of products in the same broad segment of functionality. Cost and performance would almost certainly be in AMDs favor I expect and would only be mitigated if Intel felt that that market segment would be worth addressing. Realistically they may simply be content to let AMD do the work to supply that market segment given how small it is (cross-section of folks that need many more cores and connectivity but don't need the reliability, support, and SLAs that come with enterprise gear).

My comment was more that intel is nowhere near core count vs Threadripper or EPYC, when AMD moves to 5nm, they will release 4th gen Epyc with at least 96 cores which is twice the highest core Xeon. How is intel going to counter that?

 

[LazyGamer]

My comment was more that intel is nowhere near core count vs Threadripper or EPYC, when AMD moves to 5nm, they will release 4th gen Epyc with at least 96 cores which is twice the highest core Xeon. How is intel going to counter that?

First they'll have to determine whether that's actually something worth producing, and supposing Intel gets smaller nodes figured one way or the other, that's really the last thing too. Obviously such a part isn't really feasible on their 14nm node. Not impossible, just not very feasible.

If Intel decided that there was enough of a market for them to go ahead with such a monstrosity then they'd find a way, and we'd be foolish to expect otherwise. Whether they do or not is more related to the differences in priorities, i.e., AMD has market segments to break into (and in some cases invent), whereas Intel is far more beholden to the bottom line.

If the market for such products were too small and the margins too thin, the ROI might even be negative, and while AMD can use all of the attention they can get, Intel might just let AMD eat the costs. At that point the production budget might be more correctly assigned to marketing for accounting purposes!

 

[Brian_B]

I was actually just issued a laptop from work with an Intel 10nm CPU... which is proof that they're shipping stuff and not much more. The bigger issue is that Intel isn't making these with more than four cores so far as I can tell and that limits their reach!

Yeah... 10nm "works" but that doesn't mean it isn't broken. Pumping out a few low yield, low core count, small die area, low power CPUs isn't the same thing as putting out high volume, high performance flagship products. The process isn't reliable enough for them to be able to get good yields on anything that requires significant die size. I'm not a semiconductor engineer, so I don't know specifics, but just looking at what's out there the problem they have is pretty public, just no one at Intel is saying it out lout in public.

If Intel decided that there was enough of a market for them to go ahead with such a monstrosity then they'd find a way, and we'd be foolish to expect otherwise. Whether they do or not is more related to the differences in priorities, i.e., AMD has market segments to break into (and in some cases invent), whereas Intel is far more beholden to the bottom line.

Well, right now the problem is that, without outsourcing, Intel can't. They are already pushing the limit of what they can do with their own 14nm process. Trying to cram more cores on 14nm is going to be dies that are so large, power hungry, and/or clock slow that it just isn't going to be competitive. To get up there where they can compete with AMD on core counts, Intel has to be able to transition large, high performance dies over to a smaller process node. So, I guess I will backtrack a bit - when you say they will find a way - yeah... that is true - they could outsource and make it happen. But they won't do it vertically integrated until they fix their foundries.

 

[David_Schroth]

My comment was more that intel is nowhere near core count vs Threadripper or EPYC, when AMD moves to 5nm, they will release 4th gen Epyc with at least 96 cores which is twice the highest core Xeon. How is intel going to counter that?

They can strongly encourage enterprise software makers to price by the core rather than the socket....

 

[Grimlakin]

They can strongly encourage enterprise software makers to price by the core rather than the socket....

They do.. as I said in the 96 core count epyc thread. Licensing 96 cores for sql enterprise is just a dumb amount of money. But if you're cloud hosting and the cost for memory density isn't an issue.. moar cores all day long. This is going to make that density of 1024 threads in a 2u rack space seem lazy when they add 512 more threads to the same amount of rack space. (Not including water cooling).

The real tell will be can the memory chips and cups maintain long term reliability in a always on always used space. Right now 32 gig ecc memory modules are throwing predictive failures pretty often in our esxi hosts. We have the support contracts to take care of them but still a pain.

 

[LazyGamer]

Right now 32 gig ecc memory modules are throwing predictive failures pretty often in our esxi hosts.

Yikes... I don't know that we've replaced a memory module in a delivered system in the last five years. Plenty of RAID controllers (thanks HP) though. A few spinners because somehow the solid-state revolution hasn't made it to our main customer yet.

All Intel and not current-gen HP servers because the baseline remains static, and they're not pushed very hard in the field, so maybe that's part of it?

 

[Grimlakin]

Yikes... I don't know that we've replaced a memory module in a delivered system in the last five years. Plenty of RAID controllers (thanks HP) though. A few spinners because somehow the solid-state revolution hasn't made it to our main customer yet.

All Intel and not current-gen HP servers because the baseline remains static, and they're not pushed very hard in the field, so maybe that's part of it?

Yea the only systems I've seen predictive failures on are on esxi hosts oddly enough. Of course they have a good chunk of memory 512gb each. And I've seen more than a handfull of HBA failures. It seems I'm getting 2 a year. But that covers all server types.

And these servers are getting long in the tooth off of the 3 year contract.. going for a 5 year lifecycle.

 

[LazyGamer]

Of course they have a good chunk of memory 512gb each.

That's pretty rare for us; but so is serious virtualization. Our data needs to be more in flight at low latencies and needs to be segregated pretty thoroughly; generally we have compute and memory to spare.

Sad part is that since there wasn't a real need, they're skipping us to containers and more or less skipping over the 'virtualization' phase. And those are going into clouds with only a fraction maintained locally for prototyping.

 

[Grimlakin]

That's pretty rare for us; but so is serious virtualization. Our data needs to be more in flight at low latencies and needs to be segregated pretty thoroughly; generally we have compute and memory to spare.

Sad part is that since there wasn't a real need, they're skipping us to containers and more or less skipping over the 'virtualization' phase. And those are going into clouds with only a fraction maintained locally for prototyping.

If you need low latency putting them in the cloud is going to be PAINFUL. Ask if they can guarantee aggressive latency thresholds. They can't and their 'always up' algorithms mess with it too.

 

[MacLeod]

Sounds good to me. I'm long and deep on Intel stock so for my retirement's sake I hope they are right.