Analyst: “Five, Six, Seven Years” before Intel Can Fix Its Chipmaking Process

Megalith

Megalith

24-bit/48kHz
Staff member
Joined
Aug 20, 2006
Messages
13,000
Wall Street analyst Hans Mosesmann believes it could take as long as seven years for Intel to fix its business, which is reportedly suffering due to the company’s “broken” 10-nanometer process technology. While these chips may finally see the light of day at the tail end of 2019, the effect of being “behind by a year or two in terms of process technology” is becoming increasingly evident: Mosesmann points out that AMD’s stock is seeing an “unprecedented” rise and believes they could have 25 percent of the market in a few years.

Intel has teased the large-scale release of its next-generation 10 nanometer chips for years, promising they'd deliver better performance with lower power usage than chips built with the company's older generation, 14 nanometer technology. Intel said last month that its 10 nanometer chips will be released for holiday 2019. Samsung is already manufacturing 10 nanometer chips, and AMD plans to launch 7 nanometer products later this year. One nanometer equals one-billionth of a meter.
 
So is there something intrinsically different about a 10nm chip than, say, a 9nm chip?

Is this one of those deals where they are so invested in it that they are just going to keep dumping money and time at it?

When do they move on to the next, smaller process and write 10nm off?
 
Naturally, financial analysts are the people to ask for a reliable opinion on these technical matters.

Anyone who gambles on Intel, of all companies, not catching up within one or two years should definitely be kept away from (influencing) any financial decisions. Betting on them not jumping ahead in that time frame is only marginally less bad.

This does not even take into account the very likely event that the competition will have problems of their own in developing new process production technologies.
 
Last edited:
I think the future is one where intel and amd compete on the merits of their architectures not on their process technologies, which will be broadly/roughly equivalent. That in itself is quite the thing.
Unless I'm missing anything on the Intel side aside from new instructions the actual architecture is unchanged since Haswell? Standing completely still since 2013 and well into 2019 and perhaps beyond (if the 10nm 'lake stuff only gets avx512 and meltdown and spectre fixes) is one giant pigeon coming home to make one fat-assed roost. Intels only recourse is to get off the lovely lovely gravy train and offer far more chip for the money in every single segement. It's here in the consumer desktop space and it's going to happen in the server realm to.
 
Intel is behind on 10nm. AMD is fabless and rely on tsmc and global foundries, which have 7mn prototypes out already. It will take Intel years to redesign their flawed meltdown ridden design. Look at the encryption features of Ryzen pro. Intel has no match for AMD's current lineup and are at least a couple years away from even matching. You can't undo bad engineering in a year.
 
Mega6 said:
Intel is behind on 10nm. AMD is fabless and rely on tsmc and global foundries, which have 7mn prototypes out already. It will take Intel years to redesign their flawed meltdown ridden design. Look at the encryption features of Ryzen pro. Intel has no match for AMD's current lineup and are at least a couple years away from even matching. You can't undo bad engineering in a year.
Click to expand...
Well good thing is for them, i think they can just copy AMD?.. so step one.. copy AMD, and step two.. make the chips with TSMC or Global Foundries.... Oh, wait a minute...
 
I'm still waiting for a bios fix on my sb-e that will never materialize.
 
Mega6 said:
I'm still waiting for a bios fix on my sb-e that will never materialize.
Click to expand...
You mean your MB manufacturer (ASUS if I'm correct) hasn't released a new BIOS update? I have an X79 DARK, and they released a BIOS for it in April.
 
Even if Intel tried to slow the pace: https://www.technologyreview.com/s/601102/intel-puts-the-brakes-on-moores-law/ , doesn't mean that AMD did. Looks like some of those decisions are starting to bite Intel in the butt. And that is ignoring the recent security issues. Assuming AMD didn't slow down the pace, if Intel winds up 12~18 months behind by the time they get regrouped, that will be a very significant technology gap. Moore's Law is going to be a harsh thing for Intel. 18 months behind could mean their products have half the transistors as competing products.

Of course, even if AMD does everything right, they could still be screwed if one of their contract fab factories gets taken offline by some kind of disaster or someone making the factory a deal it can't refuse.
 
SolidBladez said:
If you're up to doing it yourself there's this: https://www.win-raid.com/t154f16-Tool-Guide-News-quot-UEFI-BIOS-Updater-quot-UBU.html
Click to expand...

Thanks, still debating moded bios https://rog.asus.com/forum/showthre...OS-4901-MODDED-With-0713-Microcode-patch-here!

Matt Christie said:
You mean your MB manufacturer (ASUS if I'm correct) hasn't released a new BIOS update? I have an X79 DARK, and they released a BIOS for it in April.
Click to expand...

That's correct, 4901 is pre-meltdown. Rive
 
That's correct, 4901 is pre-meltdown. Rive[/QUOTE]



Have you contacted their customer service about BIOS updates? Maybe even check newer boards (eg X99) to see if there have been updates, as it may be taking them time to implement them on X79. Or maybe they just don't give a damn!
 
Matt Christie said:
That's correct, 4901 is pre-meltdown. Rive
Click to expand...



Have you contacted their customer service about BIOS updates? Maybe even check newer boards (eg X99) to see if there have been updates, as it may be taking them time to implement them on X79. Or maybe they just don't give a damn![/QUOTE]


Asus isn't covering x79. https://www.asus.com/News/V5urzYAT6myCC1o2

Hence iffy home made hack on rog, thanks again guys. First checked in Jan 2018 and nothing has changed.

Let's get this thread back on target.
 
Last edited:
tordogs said:
To heck with their 10nm chips--when are they gonna fix all the microcode problems that cause their latest and greatest to run like Q6600s?--or worse.
Click to expand...
Their old chips are hampered by the fixes even more, so they can actually claim that their new chips have become faster compared to the old ones thanks to the fixes :D
 
craigdt said:
So is there something intrinsically different about a 10nm chip than, say, a 9nm chip?
Click to expand...


10nm and 7nm are based on finFET. It doesn't scale the same as other nodes, so it takes significantly more time to make a workable design, let alone get your foundary to be able to produce those at acceptable yields. This article does a decent job breaking it down: https://semiengineering.com/10nm-versus-7nm/

I'm surprised Intel hasn't just outright poached engineers like crazy from TSMC/GF.
 
tordogs said:
To heck with their 10nm chips--when are they gonna fix all the microcode problems that cause their latest and greatest to run like Q6600s?--or worse.
Click to expand...

You can't "fix" a physical design flaw. The only thing they can do is mitigate it via microcode which cripples the functionality of the chip.

In other words, Intel a pretty well fucked. Expect them to lose a large portion of the server market over the next couple years.
 
It's truly ironic.

AMD: Won't make a premium GPU for who know's how long but has come back swinging and then some with CPU.

Intel: Allowed their CPU's to stagnate with minimal gains while users from nearly 10 years ago are still on the fence for upgrades but now delving into GPU w/ Raj's team.

10mm, 7mm, heck even 1mm won't matter. They need a new architecture to get away from x86 if they really want to grab the future. Moore's laws have proven themselves and then some. Sure there's still some water to be squeezed from that stone but those few percentages may not even be a stop gap for the competition at this point.
 
15352934406bstrk4mgg_1_1.jpg



Farting in that suit would be brutal!!!!!!! All that heat rolling up the back of your neck, fogging up your goggles, everyone would know right off, Kyle STOP it your goggles are fogged up again!
 
Elios said:
and that is based on the P3 lol so its pretty old at this point
Click to expand...
Which itself is based on the Pentium Pro from the mid-1990s, haha.
I think this is more than just "moving away from the x86/x86-64 ISA", mainly since AMD are rolling with it with great success, but they do need a new microarchitecture for this generation.

Hopefully that won't be like Netburst all over again, but really, after all of the hardware exploits being discovered and security-corners being cut by Intel to make their latest "performance gains" (which are all quickly going away with patches), this has officially become far worse than Netburst ever could be.
The only reason they haven't attempted the vast anti-consumer push against AMD this time around is due to all of said exploits, their CEO's bullshit, and the numerous lawsuits against them hitting the press, front page - kind of hard to make all of that look like AMD's fault when it is so clearly Intel's.
 
10nm issues, exploits such as Spectre, Meltdown, Foreshadow etc (with who knows how many more undiscovered ones)...maybe it's time I seriously look into AMD...why can't they focus on making a good gaming chip to go along with their impressive multi-tasking CPU's?...if that happened AMD could totally play catch-up...in fact AMD should be marketing the hell out of this by focusing on Intel's current weakness- the exploits which lower performance...they should be putting ads online, talking about it in every interview etc
 
Riccochet said:
I'd be interested in a [H] review of Intel's current CPU's that have been patched up. See the real impact of the Foreshadow patches.
Click to expand...


Could be part of a CPU generational change article. 870k, 2600k, 3770k, 4790k, 6700k, 7700k, 8700k
 
Rauelius said:
Could be part of a CPU generational change article. 870k, 2600k, 3770k, 4790k, 6700k, 7700k, 8700k
Click to expand...

summary of that article: no major changes from 2600K to 8700K...additional cores is the main addition along with AVX instruction set in Sandy Bridge...the main SSE2 instruction set is the same today as on my i7 980X...Intel has barely evolved their chips in almost a decade because of lack of competition
 
Riccochet said:
You can't "fix" a physical design flaw. The only thing they can do is mitigate it via microcode which cripples the functionality of the chip.

In other words, Intel a pretty well fucked. Expect them to lose a large portion of the server market over the next couple years.
Click to expand...

Of course you can fix it...it's called a new chip. As they've said, they've caught it in the new Canyonlake chips coming in 2019 and going forward. I'm not trying to suggest it's great, but there's a solution not far off and they say the chips will regain the lost power. Personally I have a 4000 series so I'm looking to upgrade soon enough for better NVME and when PCI 4.0 comes out, but until then I'm happy, as many others are.

I'm open to AMD but so far they haven't been exactly showing great performance for gaming, and I'm not interested in content creation personally. If they change that, I'd be happy. I used to love them back in their early Athlon days and before.

Also this analyst forgets that Intel's 10nm process is not "behind" AMD's 7nm process at all. They are different styles which do not compare to say 7nm is smaller so it must be better.
 
Sonickrunch said:
Of course you can fix it...it's called a new chip. As they've said, they've caught it in the new Canyonlake chips coming in 2019 and going forward. I'm not trying to suggest it's great, but there's a solution not far off and they say the chips will regain the lost power. Personally I have a 4000 series so I'm looking to upgrade soon enough for better NVME and when PCI 4.0 comes out, but until then I'm happy, as many others are.

I'm open to AMD but so far they haven't been exactly showing great performance for gaming, and I'm not interested in content creation personally. If they change that, I'd be happy. I used to love them back in their early Athlon days and before.

Also this analyst forgets that Intel's 10nm process is not "behind" AMD's 7nm process at all. They are different styles which do not compare to say 7nm is smaller so it must be better.
Click to expand...

Except that I've heard nothing about Intel's next CPU being anything other than another iteration of what they already have. That's not going to fix the architectural problems since they are still going to be there.

Add into that the fact that Intel's 10nm process is years behind schedule and even what little they are showing from it isn't promising in the least.

The only shot Intel has at the moment is to quickly fix whatever problems their 10nm process has which is extremely unlikely considering how many years it's already behind. That has additional complications since whatever chips they had planned were likely supposed to use that 10nm process and since it's basically screwed at this point it puts them even farther behind on new chips. It will likely require whole new designs or major redesigns which is going to put them even farther back. Finished or mostly finished designs likely can't be done on the 10nm process due to issues with the process itself. You can't just take those designs and punch them out on the old process because they may not work at all and would require a redesign to even think of attempting them and it's likely performance and features would suffer on an older process.
 
SmokeRngs said:
Except that I've heard nothing about Intel's next CPU being anything other than another iteration of what they already have. That's not going to fix the architectural problems since they are still going to be there.
Click to expand...
Intel just recently hired Jim Keller, but it will still take 3-5 years for a new architecture.
 
Sonickrunch said:
Of course you can fix it...it's called a new chip. As they've said, they've caught it in the new Canyonlake chips coming in 2019 and going forward. I'm not trying to suggest it's great, but there's a solution not far off and they say the chips will regain the lost power. Personally I have a 4000 series so I'm looking to upgrade soon enough for better NVME and when PCI 4.0 comes out, but until then I'm happy, as many others are.

I'm open to AMD but so far they haven't been exactly showing great performance for gaming, and I'm not interested in content creation personally. If they change that, I'd be happy. I used to love them back in their early Athlon days and before.

Also this analyst forgets that Intel's 10nm process is not "behind" AMD's 7nm process at all. They are different styles which do not compare to say 7nm is smaller so it must be better.
Click to expand...

You seem to forget that server parts are always a couple years behind desktop parts. And, honestly, these vulnerabilities have more to do with server parts than desktop. OEM's may try to rush these new parts out the door, but in the long run 2 years is an eternity to expect companies to live with a huge reduction in performance. Especially where virtualization is concerned and hyperthreading is heavily utilized. I agree that about 25%, or more, of the server market will go AMD. That's massive.
 
Intel has stated in their own chip roadmap the fixes are in place in canyon lake. and they’ve also stated the expectation of the in-hardware fix to the issue will completely restore the performance loss of the microcode and software fixes. It shouldn’t take 3-5 years. The lithograph issue as well. I think the analyst forgets that not all processes are made the same. Intels 10nm is in some ways more complex than amds 7nm. I’m not going to argue which is better here but just because 7 is a smaller number means nothing. Also yea intel had issues with their 10nm process. But it’s not like they stopped. They are technically on 10nm++
 
Still not convinced that all these flaws are that real. Sounds like the you need a new car pitch because you have a bad air bag. Must tell you though I am a complete cynic.....
 
Sonickrunch said:
Intel has stated in their own chip roadmap the fixes are in place in canyon lake. and they’ve also stated the expectation of the in-hardware fix to the issue will completely restore the performance loss of the microcode and software fixes. It shouldn’t take 3-5 years. The lithograph issue as well. I think the analyst forgets that not all processes are made the same. Intels 10nm is in some ways more complex than amds 7nm. I’m not going to argue which is better here but just because 7 is a smaller number means nothing. Also yea intel had issues with their 10nm process. But it’s not like they stopped. They are technically on 10nm++
Click to expand...

3-5 years is the life cycle of server hardware. Companies are faced with a bad situation. Apply the patches and suffer huge performance losses, don't apply the patches and be vulnerable until Canyon Lake Xeon parts are released by OEM's, or replace their equipment with AMD.

Given the nature of the data most of these systems house and the performance needs for today's OTLP workloads that rules out options 1 and 2.

I say years because OEM's won't be ready to release Canyon Lake server parts for 8-10 months after the CPU's are released to OEM's. They don't just have the stuff laying around waiting for CPU's to slap in them. They're still producing current gen parts. That 8-10 months is pushing it too.

People seem to forget that nobody gives a shit about how this is going to effect desktop parts. At least not at the corporate level. It's high dollar VM hosts.
 
zalazin said:
Still not convinced that all these flaws are that real. Sounds like the you need a new car pitch because you have a bad air bag. Must tell you though I am a complete cynic.....
Click to expand...

I sympathize with intel. They dominated the industry based on two things, more performance on hand always, and being way ahead in the race to a new process in the fab.

They got a chunk of that performance from taking a calculated risk. The stuff in specter etc. isn't really a big threat to the desktop user. Nor is it in most server environments. It's not a very useful exploit unless you are letting bad neighbors execute arbitrary code on a shared system. Until cloud computing took off and lowered prices a lot, this couldn't really be a thing. There was no use case where arbitrarily fishing for random blocks of memory might yield anything good anytime soon. You don't bother and go for easier methods with more sure results. But with the cloud, well you can sit there and do this shit every so often. You pay a small fee for a few CPU cycles, and you might net a big fish for lite investment. Getting access to a machine with neighbors to snoop on and having those neighbors be constantly shuffled out is zero cost and happens regularly.

They'll fix this, and do so fairly quickly.

The bigger risk is that they lose their lead in the process race. Even if they CAN recover it, that won't be cheap and the market will punish them for that expense.
 
You must log in or register to reply here.
Back
Top