Intel's 10nm only coming to servers in 2020 with Ice Lake

[cj09beira]

Intel's 10nm has been the subject of quite a lot of discussions recently, as it has been delayed time and time again, and now according to a intel official slide found in a chinese presentation called "Intel High performance Computing and Intelligent Computing" 10nm will only come to servers sometime in 2020, this revelation puts intel in a disadvantage against amd as they are said to release Epyc 2 on 7nm in the first half of 2019, meaning intel will have lost the process node race for the first time ever and by a full year.

Quote

This roadmap reveals some bad news for Intel, that the company will not be shipping any 10nm Xeon Processors until Q1 or Q2 2020, around a year after AMD's release of 7nm, which is roughly equivalent to Intel's 10nm, Zen 2-based EPYC 2 processors, giving AMD a significant advantage in 2019 thanks to their use of advanced process technologies.  

In 2019, Intel will be forced to compete with EPYC 2 on today's 14nm technology with Cascade Lake-SP and later Copper Lake-SP processors, giving AMD a potential advantage in terms of power efficiency and die size thanks to their use of a 7nm process node. This will, in turn, give AMD an opportunity to steal a lot of market share from Intel within the server market, springboarding off their success with today's EPYC processors with what is expected to be a significant performance leap. 

On the consumer side thanks to the smaller die sizes its expected that 10nm arrives earlier the question is how much earlier, considering intel right now is only able to produce a 2 core cpu with their 10nm node (2 cores + igpu with igpu disabled due to low yields) a full 6-8 core+igpu die its not expected anytime soon.

 

Personal opinion:

Hopefully the moment of the pendulum is just right so that Amd gains enough market share and R&D money to survive long term but not too much to where Intel is left out of the race (doubt it could even happen, they just have too much money).

Until now Amds investment on IF only meant that they could reduce prices but now we can see that it has an even better feature, which is it allows Amd to use nodes that are still in their initial phase as each die is much smaller, which will mean Amd will be able to use the newest node sooner than intel (until they start to do this also).

Epyc 2 is also expect to have 48 cores with a potential second sku coming later with up to 64 cores, this will mean that amd's one socket servers that were already really attractive will become even more so, having competition on the market is so fun.

 

Source: Overclock3d.net

[Deli]

The downfall of Intel. No one but itself to blame.

[D13H4RD]

Meanwhile, at AMD

 

7nm intensifies

[asus killer]

i'm not an expert but when people say "10nm Intel equivalent to 7nm AMD" aren't we ignoring things like thermals and power consumption that can make a lot of difference as we seen on the MBP 2018?

[D13H4RD]

1 minute ago, asus killer said:

i'm not an expert but when people say "10nm Intel equivalent to 7nm AMD" aren't we ignoring things like thermals and power consumption that can make a lot of difference as we seen on the MBP 2018?

I think the issue in that case was that Intel was pushing the limits of its core architecture alongside what that and its 14nm fab node could be capable of. 

[cj09beira]

1 minute ago, asus killer said:

i'm not an expert but when people say "10nm Intel equivalent to 7nm AMD" aren't we ignoring things like thermals and power consumption that can make a lot of difference as we seen on the MBP 2018?

we still don't have a product made using 7nm amd (Global foundries), so we can't say yet, though that 7nm process is based on the same one used by IBM which have frequency targets on the upper 5ghz range so it should be very good 

[GoldenLag]

2 minutes ago, asus killer said:

i'm not an expert but when people say "10nm Intel equivalent to 7nm AMD" aren't we ignoring things like thermals and power consumption that can make a lot of difference as we seen on the MBP 2018?

10nm and 7nm node size is kinda vague to describe how the dense the node actually is. Which is why 10nm will be a out the same as 7nm. Take the node numbers a indicative numbers and not pure fact

[jiyeon]

11 minutes ago, cj09beira said:

this revelation puts intel in a disadvantage against amd as they are said to release Epyc 2 on 7nm in the first half of 2019, meaning intel will have lost the process node race for the first time ever and by a full year.

DAMN.

[Taf the Ghost]

Most recent AMD roadmaps have had Zen2 then Zen2+ paths, so it's very possible AMD could have the Zen2+ servers out by the time Ice Lake-SP lands. Assuming AMD brings Zen2+ out on servers. AMD might just be straight up sampling Zen3 by that point.

[Master Disaster]

Anybody ever thought Intel are playing the smart game?

 

Everybody knows 7nm is just about the limit of silicon transistors and without some new breakthrough, be it either a compound or new material, once we reach 7nm we're kind of stuck there.

 

It's possible Intel are sitting back and waiting for AMD to show all their playing cards to the group before they push forward with more node shrinks. At the end of the day all they have to do is beat AMD at 7nm and they've effectively won the silicon race forever.

[LemonSpice]

2 minutes ago, Master Disaster said:

Anybody ever thought Intel are playing the smart game?

 

Everybody knows 7nm is just about the limit of silicon transistors and without some new breakthrough, be it either a compound or new material, once we reach 7nm we're kind of stuck there.

 

It's possible Intel are sitting back and waiting for AMD to show all their playing cards to the group before they push forward with more node shrinks. At the end of the day all they have to do is beat AMD at 7nm and they've effectively won the silicon race forever.

I feel like the everyone knows the "nm" limit has been used for years but this comapnys with huge rnd budgets have been able to make those breakthroughs and I almost cant see that not happening in the near future at least with intel.

[Emberstone]

5 minutes ago, JEman99 said:

I feel like the everyone knows the "nm" limit has been used for years but this comapnys with huge rnd budgets have been able to make those breakthroughs and I almost cant see that not happening in the near future at least with intel.

And I'm sure that, once the limits of silicon have actually been reached, components within CPU's will start containing other materials.

 

I highly doubt that microprocessor technologies are going to hit a hard wall any time soon.

[GoldenLag]

6 minutes ago, Master Disaster said:

Anybody ever thought Intel are playing the smart game?

 

Everybody knows 7nm is just about the limit of silicon transistors and without some new breakthrough, be it either a compound or new material, once we reach 7nm we're kind of stuck there.

 

It's possible Intel are sitting back and waiting for AMD to show all their playing cards to the group before they push forward with more node shrinks. At the end of the day all they have to do is beat AMD at 7nm and they've effectively won the silicon race forever.

"5nm" is allready on roadmaps though. And i believe its easier to imorove upon 7nm node than to scale down their size to then try to improve it.

 

Also AMD 7nm isnt AMD's. They have fab companies that manufactur the chips. 

 

Once 5/7nm is saturated its time to invrease the die size, which is a whole lot easier on a chiplet design.

 

There will allways be a node race. Currently 7nm has more momentum coming out of the corner than what intel 10nm has.

[cj09beira]

1 minute ago, Master Disaster said:

Anybody ever thought Intel are playing the smart game?

 

Everybody knows 7nm is just about the limit of silicon transistors and without some new breakthrough, be it either a compound or new material, once we reach 7nm we're kind of stuck there.

 

It's possible Intel are sitting back and waiting for AMD to show all their playing cards to the group before they push forward with more node shrinks. At the end of the day all they have to do is beat AMD at 7nm and they've effectively won the silicon race forever.

thats not what is happening, as right now intel is too busy trying to make their 10nm node work, and this is a field of iterative improvements so if you start being left behind its really hard to catch up, global foundries now also has all the expertise that ibm had which is a lot (only company selling 5.3ghz 10 core chips with 4 x smt, literally a monster of a cpu), now intel is probably pushing for photonics and electron spin transistors but that is well into the future 

 

1 minute ago, Emberstone said:

And I'm sure that, once the limits of silicon have actually been reached, components within CPU's will start containing other materials.

 

I highly doubt that microprocessor technologies are going to hit a hard wall any time soon.

you are right, we will just keep doing what we are doing now, divide the cpu into smaller parts and place them side by side, when that reaches its max we will build upwards 

[Master Disaster]

4 minutes ago, JEman99 said:

I feel like the everyone knows the "nm" limit has been used for years but this comapnys with huge rnd budgets have been able to make those breakthroughs and I almost cant see that not happening in the near future at least with intel.

There is a process called EUV which would supposedly allow shrinkage down to 5nm on bare silicon but my understanding is that it's incredibly complicated, incredibly hard to miniaturise for use in factories, incredibly expensive and still not guaranteed to work.

 

There was a huge buzz around silicon germanium a few years back but that seems to have died down recently.

 

At this point it really does seem like 7nm will be the limit, at least in X86 CPUs. I know Samsung said their looking at 4nm for mobile but I believe that's a different process so maybe the 2 things are equal?

[LemonSpice]

1 minute ago, Master Disaster said:

There is a process called EUV which would supposedly allow shrinkage down to 5nm on bare silicon but my understanding is that it's incredibly complicated, incredibly hard to miniaturise for use in factories, incredibly expensive and still not guaranteed to work.

 

There was a huge buzz around silicon germanium a few years back but that seems to have died down recently.

 

At this point it really does seem like 7nm will be the limit, at least in X86 CPUs. I know Samsung said their looking at 4nm for mobile but I believe that's a different process so maybe the 2 things are equal?

I mean forgive me if Im wrong but i would feel like creating a complete 1 billion plus transistor Micro processor whether it be arm based or an x86 cpu would be able to be done in the same manner.

[Taf the Ghost]

33 minutes ago, GoldenLag said:

10nm and 7nm node size is kinda vague to describe how the dense the node actually is. Which is why 10nm will be a out the same as 7nm. Take the node numbers a indicative numbers and not pure fact

From the publicly presented information, the total density between the Intel 10nm and GloFo 7nm are roughly similar, though it seems like Intel's cache density will be worse, which is the biggest part of any modern CPU. Thus, the same designs should be slightly smaller on GloFo/TSMC than on the Intel node. The main issue, right now, is if Intel ever releases a 10nm part.

 

Intel is currently roadmapping 4 years behind schedule.

[cj09beira]

1 minute ago, Master Disaster said:

There is a process called EUV which would supposedly allow shrinkage down to 5nm on bare silicon but my understanding is that it's incredibly complicated, incredibly hard to miniaturise for use in factories, incredibly expensive and still not guaranteed to work.

 

There was a huge buzz around silicon germanium a few years back but that seems to have died down recently.

 

At this point it really does seem like 7nm will be the limit, at least in X86 CPUs. I know Samsung said their looking at 4nm for mobile but I believe that's a different process so maybe the 2 things are equal?

the problem of euv is that most materials absorb it, which is a pain in the back side so they are having a hard time making masks for it (to protect the wafer from particles of dust, they need to be transparent to euv), it seems that creating euv is also not trivial and not very efficient at all which isn't helping, right now when euv is used they dont use masks which increases the chances of defects as the wafers aren't protected

[Taf the Ghost]

5 minutes ago, Master Disaster said:

There is a process called EUV which would supposedly allow shrinkage down to 5nm on bare silicon but my understanding is that it's incredibly complicated, incredibly hard to miniaturise for use in factories, incredibly expensive and still not guaranteed to work.

 

There was a huge buzz around silicon germanium a few years back but that seems to have died down recently.

 

At this point it really does seem like 7nm will be the limit, at least in X86 CPUs. I know Samsung said their looking at 4nm for mobile but I believe that's a different process so maybe the 2 things are equal?

Node names are just that these days. Samsung is going to have a "nm" node from 12nm through 4nm. That's just currently on their roadmap. The main thing is that transistors are getting taller. It's the space between the transistors, not the transistors themselves, that has been shrinking. We'll be seeing "1nm" nodes in the late 2020s without too much issue.

 

This is also why AMD is headed in the full Chiplet direction. 

[Stefan Payne]

49 minutes ago, cj09beira said:

intel will have lost the process node race for the first time ever and by a full year.

I think that is not true and there was a time where the others were better. But that would be around the 2000s...

 

But lets see how the Intel processors are to a competition with a superior process...

[cj09beira]

2 minutes ago, Taf the Ghost said:

From the publicly presented information, the total density between the Intel 10nm and GloFo 7nm are roughly similar, though it seems like Intel's cache density will be worse, which is the biggest part of any modern CPU. Thus, the same designs should be slightly smaller on GloFo/TSMC than on the Intel node. The main issue, right now, is if Intel ever releases a 10nm part.

 

Intel is currently roadmapping 4 years behind schedule.

though if i remember right intel's cache had more transistors which would lead to better clocks, so glofo's 7nm is denser if using the density optimized cell but slightly less dense if using the performance cell.

btw i would love amd to release a cpu using the 7nm HPC (ibm's 7nm) even just making a few and showing us what it could do would be cool

[Delicieuxz]

Is Cannon Lake as a desktop product still going to happen?

[Taf the Ghost]

1 minute ago, Stefan Payne said:

I think that is not true and there was a time where the others were better. But that would be around the 2000s...

 

But lets see how the Intel processors are to a competition with a superior process...

Intel has had a node advantage since the 1980s. The early 2000s the Intel node was just slightly better, but AMD had better designs at the time.

[Taf the Ghost]

Just now, Delicieuxz said:

Is Cannon Lake as a desktop product no longer going to happen?

Canceled in 2016. Some 5w parts came out, and they're bad. There was supposed to be 5w and 15w CNY parts, but all but a small run of the 5w was canceled in 2017.

[Nicnac]

Wow imagine how behind they will be in 2020! Maybe they will have switched places with AMD then?! Interesting times