Intel kills 10nm ?? oO

[Rroff]

This was reported by Anandtech last week when they announced their quarterly results:

"As for 10 nm ramp in general, Intel is only talking about producing its relatively small Ice Lake-U processors in volumes this year, which is the company’s typical way of ramping up a new node. When it comes to their larger Ice Lake-SP server parts, Intel expects to launch those 10 nm Xeon products in 2020. The company says that its Ice Lake-SP CPUs will be available in less than 12 months after its Ice Lake-U products hit the market. In fact, Intel has even advised investors to expect 10 nm Xeons to arrive “rather sooner than later” in 2020, which would imply something earlier than Q4'2020."

Speculation suggests that desktop parts are still a problem as they require high clock speeds and process performance which might well both still be lacking.
For server parts ultimate clock speed is not usually so important and as others have stated it's an area where AMD can do more damage to Intel so they will prioritise that.
The fact that there was no official mention of 10nm desktop parts speaks volumes.

The big question is how soon will 7nm come and how will Intel prioritise that?
If they prioritise server first and laptop second then maybe that is why desktop will still only get 10nm even as late as 2020/21.
Could we even see Server @7nm before desktop @10nm!
That might partly depend on the strength of their designs and how they will compete with AMD @7nm & 5nm.
Some of the staff at Intel are certainly having to sweat a lot more than they did in the decade prior to Zen launching.

Intel seem to be focusing on creating/reworking designs that can be relatively quickly moved from 14->10->7nm which suggests they don't intend to sit on the 10nm node long and also has held things up a little but potentially pays off in the longer term. They also seem to have switched focus on a number of fabs to prepare them for 7nm instead of 10nm though not confirmed on all of them some of the equipment/work known to be going in doesn't make sense for 10nm - neither does the amount of money being thrown at some of the sites unless they were basically starting again from scratch on 10nm.

What actually happens might be another story but from what I can find out seems the most likely path is that 10nm will mostly be used for lower end products and transition to chipset production, etc. some of which is currently still on 22nm and/or has been moved back to 22nm due to 14nm shortages.

As an aside Intel's 14nm is actually some way from being tapped out performance wise - I'm not sure at what stage it is now but the last I heard the then latest incarnation was still missing optimisation (kind of like TSMC's 16FF to 12FF) that could give around 15-16% higher clock speeds as it hadn't been deemed worth the effort with a supposed imminent move to 10nm.

EDIT: This is very much an external perspective and might be wrong but from talking to people who deal with Intel professionally I very much get the impression that there are a lot of people at Intel that just tell people what they think they want to hear to get them to go away and they are increasingly put off working with them due to a seeming never ending loop of different managers each disconnected from the others when dealing with anything.

 

[Journey]

I am pretty sure that the title of this thread is becoming more and more accurate as the days go by, with only two fabs committed to 10nm and yields probably as healthy looking as my bank balance when I was a student, 10nm looks like a tick box exercise to show that they didn't scrap it, and the move to 7nm for 2022 will be the big push. The fact they back-ported fabs to 14nm speaks volumes, and protecting income/revenue should be and is their number one priority to shareholders.

Loss of the 10nm process in all but a few select products is going to cost them a great deal in market share, and it could not have happened at a worse time, what with AMD now firmly on the rebound with EPYC and offering significant benefits on multiple fronts to Xeon, and it's only going to get better for AMD in the next two years.

Have any heads rolled at Intel yet, maybe? But there are a lot more to follow once the start missing quarterly profit targets and shipments start falling quarter after quarter.

 

[Rroff]

I am pretty sure that the title of this thread is becoming more and more accurate as the days go by, with only two fabs committed to 10nm and yields probably as healthy looking as my bank balance when I was a student, 10nm looks like a tick box exercise to show that they didn't scrap it, and the move to 7nm for 2022 will be the big push. The fact they back-ported fabs to 14nm speaks volumes, and protecting income/revenue should be and is their number one priority to shareholders.

Loss of the 10nm process in all but a few select products is going to cost them a great deal in market share, and it could not have happened at a worse time, what with AMD now firmly on the rebound with EPYC and offering significant benefits on multiple fronts to Xeon, and it's only going to get better for AMD in the next two years.

Have any heads rolled at Intel yet, maybe? But there are a lot more to follow once the start missing quarterly profit targets and shipments start falling quarter after quarter.

I guess their mindset works against it but I'm surprised they haven't outsourced some production as an interim measure - both Samsung and TSMC have proven nodes that would work as a stop gap even if not as optimised for a CPU specifically as would get the best results for Intel.

There was some talk of it but seems purely down to the production of ICs for other Intel products rather than CPUs.

 

[Zeed]

I am pretty sure that the title of this thread is becoming more and more accurate as the days go by, with only two fabs committed to 10nm and yields probably as healthy looking as my bank balance when I was a student, 10nm looks like a tick box exercise to show that they didn't scrap it, and the move to 7nm for 2022 will be the big push. The fact they back-ported fabs to 14nm speaks volumes, and protecting income/revenue should be and is their number one priority to shareholders.

Loss of the 10nm process in all but a few select products is going to cost them a great deal in market share, and it could not have happened at a worse time, what with AMD now firmly on the rebound with EPYC and offering significant benefits on multiple fronts to Xeon, and it's only going to get better for AMD in the next two years.

Have any heads rolled at Intel yet, maybe? But there are a lot more to follow once the start missing quarterly profit targets and shipments start falling quarter after quarter.

Look at it this way. Next month we are BUYING 7nm (that i know is like equal to 10nm of intel) From AMD. BUYING !!! AMD is selling 7nm GPU's ect....
Not followed situation close but not seen any benchmarks of 10nm intel's parts not even ES ones when Lisa showed running Zen3 in CB months ago.

 

[hominid]

Intel still need to qualify 10nm stages/processes so they need at least 1 10nm FAB pumping out qualification parts. They will, for sure, be using parts of the 10nm process when they can, to try and cheapen up the total process - Quad patterning is very expensive @ 7nm so doing what features they can @10nm will help with the final cost.

Intel are in a rough spot, when they do get 7nm out the door it'll have to compete with the ultra refined 14nm on performance else it'll be yet more time researching. The two main things that are a given is the increase in density, hence smaller chips, and lower power usage, both are for nought if the performance is nowhere. We are at the point where there is diminishing returns, and even performance regressions, with smaller chip lithography - this could be a horrible pit that Intel will struggle to get out of.

AMD did the correct thing, even if forced to for other reasons, to get out the fabrication business!

 

[Rroff]

Intel still need to qualify 10nm stages/processes so they need at least 1 10nm FAB pumping out qualification parts. They will, for sure, be using parts of the 10nm process when they can, to try and cheapen up the total process - Quad patterning is very expensive @ 7nm so doing what features they can @10nm will help with the final cost.

Intel are in a rough spot, when they do get 7nm out the door it'll have to compete with the ultra refined 14nm on performance else it'll be yet more time researching. The two main things that are a given is the increase in density, hence smaller chips, and lower power usage, both are for nought if the performance is nowhere. We are at the point where there is diminishing returns, and even performance regressions, with smaller chip lithography - this could be a horrible pit that Intel will struggle to get out of.

AMD did the correct thing, even if forced to for other reasons, to get out the fabrication business!

It is beyond my understanding though there are a few long threads on it at Semiwiki, Cadence, etc. but AFAIK Intel's approach is a bit different 10nm and 7nm in terms of patterning - with 10nm they tried to push out the boat and skate as close to needing EUV as they could without actually using EUV and so went with a somewhat unexpected (by commentators) approach to try and get away with it while at 7nm they won't need to employ that approach and will be going at it along similar lines to Samsung and TSMC.

Again for rather complex reasons that I only have a basic understanding of it is why a lot of Semi-accurate's claims in terms of how they were relaxing it, etc. didn't make much sense.

 

[4K8KW10]

AMD did the correct thing, even if forced to for other reasons, to get out the fabrication business!

Only TSMC and Samsung for now still have no problems with cutting edge lithography.
All the others, including AMD, GlobalFoundries, IBM, and the smaller manufacturers that rely on older process, can't do much if nothing at all with the tougher reality of physical limitations.

 

[smilingcrow]

Now that Intel are gluing together Server chips 10nm isn't such a bust for them.
Although in high density data centres where power efficiency is king they should be slaughtered by EPYC @7nm.
They really are in a pickle.

 

[Rroff]

Only TSMC and Samsung for now still have no problems with cutting edge lithography.
All the others, including AMD, GlobalFoundries, IBM, and the smaller manufacturers that rely on older process, can't do much if nothing at all with the tougher reality of physical limitations.

No one has had it easy at ~7nm or below.

As an aside IBM (in alliance with Samsung) has actually had good results at 5nm.

 

[Cyber-Mav]

Hmm if intels 14nm cpus are faster than amds 7nm cpus then amd are the ones who should be worried.

Look at the gpu side. Amd have the smaller process node but their gpus are considerably slower than nvidias.

 

[LePhuronn]

Hmm if intels 14nm cpus are faster than amds 7nm cpus then amd are the ones who should be worried.

Look at the gpu side. Amd have the smaller process node but their gpus are considerably slower than nvidias.

Clock speed means nothing in isolation, look at the Pentium 4 and FX 9590. It's overall performance that counts, and if a 7nm AMD chip can perform the same (or better) than a 14nm Intel chip at lower clocks then it's AMD who have the superior product and upper hand.

And I really don't know why people keep using Radeon VII as a basis for anything. It's a PR stunt made from repurposed silicon that was never intended to game with in the first place. And as AMD's first and only 7nm "gaming" product, what exactly are you comparing it to to say it's "considerably slower than Nvidia"? Radeon VII is pitched at the RTX 2080 and performs largely the same despite having lower clocks.

Same performance at lower clocks in any other discussion would mean superiority, but not for AMD products it seems.

 

[Rroff]

Clock speed means nothing in isolation, look at the Pentium 4 and FX 9590.

I remember the Pentium D and some of the Celerons of that era - you could get them upto almost 5GHz but still performance was woeful.

While quickly googling to remind myself what speeds they were running I found this https://arstechnica.com/civis/viewtopic.php?t=628898 a good laugh at some of the BS talked about on forums heh.

 

[4K8KW10]

Clock speed means nothing in isolation, look at the Pentium 4 and FX 9590. It's overall performance that counts, and if a 7nm AMD chip can perform the same (or better) than a 14nm Intel chip at lower clocks then it's AMD who have the superior product and upper hand.

And I really don't know why people keep using Radeon VII as a basis for anything. It's a PR stunt made from repurposed silicon that was never intended to game with in the first place. And as AMD's first and only 7nm "gaming" product, what exactly are you comparing it to to say it's "considerably slower than Nvidia"? Radeon VII is pitched at the RTX 2080 and performs largely the same despite having lower clocks.

Same performance at lower clocks in any other discussion would mean superiority, but not for AMD products it seems.

Radeon VII is an impossible card. If nvidia shrinks the TU102 silicon from its 754 sq.mm 12nm to 377 sq.mm 7nm, it probably will get something around 50-60% faster than the same sized Vega 20.
It's ridiculous.

 

[Rroff]

Radeon VII is an impossible card. If nvidia shrinks the TU102 silicon from its 754 sq.mm 12nm to 377 sq.mm 7nm, it probably will get something around 50-60% faster than the same sized Vega 20.
It's ridiculous.

AMD's current 7nm products are literally the least possible work to get a functioning 7nm chip - not very representative of something designed from the ground up to benefit from 7nm.

 

[smilingcrow]

Radeon VII is an impossible card. If nvidia shrinks the TU102 silicon from its 754 sq.mm 12nm to 377 sq.mm 7nm, it probably will get something around 50-60% faster than the same sized Vega 20. It's ridiculous.

It's a dog and they released it so it needs to go to Crufts and be judged as a worst in class, a stinky dog turd.
Hopefully, with AMD's new found CPU mojo back they will be able to also transform their GPU architecture.
They've had one miracle so why not two!

 

[pmc25]

Radeon VII is an impossible card. If nvidia shrinks the TU102 silicon from its 754 sq.mm 12nm to 377 sq.mm 7nm, it probably will get something around 50-60% faster than the same sized Vega 20.
It's ridiculous.

Not sure how you 'calculate' that. I.E. You didn't.

1) They probably won't shrink the design. 2) The shrink wouldn't be that large. 3) The shrink to TSMC 7nm probably wouldn't even be do-able without a major redesign. 4) Where are you getting huge performance uplift from? The 12nm node they're using is a bespoke SHP node designed specifically for NVIDIA and their large die GPUs. You might see frequency drops on the big chips. 7nm HP that AMD are using is certainly not SHP, and not a bespoke process designed purely for specific families of large die GPUs.

I would guess there'll be a new NVIDIA HPC card on TSMC 7nm+ HP in early 2020 (last month or so of this year at a push), then consumer 7nm+ stuff in Spring or Summer 2020. New architectures obviously, for both.

 

[LePhuronn]

AMD's current 7nm products are literally the least possible work to get a functioning 7nm chip - not very representative of something designed from the ground up to benefit from 7nm.

Eh? Do you have a full suite of benchmarks for Navi and Zen 2? Because the only 7nm product AMD have available right now is Vega 20, and that wasn't designed from the ground up to be 7nm.

 

[LePhuronn]

It's a dog and they released it so it needs to go to Crufts and be judged as a worst in class, a stinky dog turd.

So a repurposed compute GPU hurriedly cobbled together to capitalise on a PR opportunity that actually competes with Nvidia's 2nd best gaming product is somehow "worst in class" and a "stinky dog turd"? Lol yeah, OK.

 

[Murphy]

And I really don't know why people keep using Radeon VII as a basis for anything.

It's used as basis because it provides an idea of what to expect from 7nm in terms of clockspeeds and power draw, it's used because up until its release we didn't have much idea of the level of improvments 7nm offered over 14/12nm. Although i guess you meant how people compare VII to Nvidia offerings, on that basis yea it's pointless comparing the two.

 

[LePhuronn]

It's used as basis because it provides an idea of what to expect from 7nm in terms of clockspeeds and power draw, it's used because up until its release we didn't have much idea of the level of improvments 7nm offered over 14/12nm

I wouldn't even go that far because Radeon VII is bodged together and pushed beyond its original design, so any and all numbers are flawed. Zen 2 and Navi will be the first true indication of what 7nm design can do.