Rumor: Intel has 14nm+++ CPUs on the way

[asus killer]

it will Bulldoze the competition 

[Brooksie359]

35 minutes ago, williamcll said:

Too bad most applications still rely on less than 4 threads. Especially non-AAA games.

Yeah it's too bad everyone is limited to running one application at a time. Oh wait you can run more than one thing at a time meaning this type of argument is silly. Also games are starting to utilize more cores as well and with the increase in core counts across the board prgramers will tailor their software to utilize this new hardware. 

[ravenshrike]

56 minutes ago, WereCat said:

What does the "+" even mean? 

 

Higher density, better yields, higher quality, lower power consumption,...? 

It means whatever they want it to mean. In practice it appears to just mean better yields with a minor clock uptick. Although how much of the clock uptick is just from the more consistent yields themselves is hard to say.

 

[S w a t s o n]

20 minutes ago, InertiaSelling said:

Tick - tock - tock - tock - tock - tock - tock - tock - tock ...

 

- Intel

Broken moore's law is right twice a decade?

[Wolther]

+++

[Mihle]

1 hour ago, xentropa said:

Meanwhile TSMC has their 7nm out (intel 10nm equivalent).  If AMD ditches  Glofo and goes full TSMC, the I am going to have to say goodbye to intel and its been a good 13 years that I swore to only use their products.

They are most likely going TSMC for 7nm only untill GloFo 7nm is ready, and then use both. I think there was a rumor that said TSMC for Epyc and GloFlo for Ryzen/threadripper, if I remember right.

[Taf the Ghost]

1 minute ago, Mihle said:

They are most likely going TSMC for 7nm only untill GloFo 7nm is ready, and then use both. I think there was a rumor that said TSMC for Epyc and GloFlo for Ryzen/threadripper, if I remember right.

Epyc 2nd Gen and the Vega 7nm part will be on TSMC. For the high-volume parts, they'll be on GloFo. So that's Navi, Ryzen and the APUs. The interesting question is if the TSMC & GloFo Zen2 designs are the same. TSMC wafers cost AMD a good bit more due to agreements, so they'll only run out high-margin parts from there.

[S w a t s o n]

Just now, Taf the Ghost said:

The interesting question is if the TSMC & GloFo Zen2 designs are the same.

If the rumours are true AMD is gonna spin 8core ccx zen2 in addition to the originally planned 6 core ccx....

[Taf the Ghost]

3 minutes ago, Swatson said:

If the rumours are true AMD is gonna spin 8core ccx zen2 in addition to the originally planned 6 core ccx....

Everyone, at least in this type of discussion space, is still arguing about the CCX size. I'm firmly in the camp it's 4c CCXs for several generations. (Until there is some new 3D connection layers that work.) With that in mind, I think there will be 3 designs from Zen2.

 

1) Big Zen: at TSMC for servers. 16c per die.

2) Small Zen: at GloFo, 12c (3 CCX) and maybe a small iGPU.

3) APU: at GloFo, I forget the name of this model. Won't be until late 2019 and will have Navi tech.

[PlayStation 2]

+ is the new v[x].

[S w a t s o n]

1 minute ago, Taf the Ghost said:

Everyone, at least in this type of discussion space, is still arguing about the CCX size. I'm firmly in the camp it's 4c CCXs for several generations. (Until there is some new 3D connection layers that work.) With that in mind, I think there will be 3 designs from Zen2.

 

1) Big Zen: at TSMC for servers. 16c per die.

2) Small Zen: at GloFo, 12c (3 CCX) and maybe a small iGPU.

3) APU: at GloFo, I forget the name of this model. Won't be until late 2019 and will have Navi tech.

16C per die would be 8 core CCX's unless you think they are going to increase CCX count? Everything I have seen up till now suggested bigger CCX rather than more

[porina]

As well as 14nm is doing now, those with a longer memory may recall that Intel's first 14nm architecture was delayed quite badly too. Broadwell did make its way into mobile space but on desktop, they only just avoided a paper launch by shipping two parts shortly before Skylake. On that note, maybe I should try to pick up an i7-5775C for fun...

 

For those wishing for software using more threads, do consider that not all software really needs it. We're certainly at the point where enthusiasts can use more than 4 cores, but that doesn't scale indefinitely. How many of us would really make good use of a 32 core processor if they had one? Maybe 6 to 8 are the new 4, and will be for some time to come until such time there is a real need for it in the consumer space.

[Taf the Ghost]

Just now, Swatson said:

16C per die would be 8 core CCX's unless you think they are going to increase CCX count? Everything I have seen up till now suggested bigger CCX rather than more

You can put at least 4 CCX on a die before there's any issues. Part of Epyc's design, the way it routes, means cores on the other socket can address to the entire die before moving down to the CCX then Core. The tight interconnection of the CCX replicates itself going up each layer within the topology. 

 

Zen, as a design philosophy, is really these replication steps of the power of a 4-part module. However, I think on-die, below the memory controllers, they can actually put as many CCXes as they want. They connect out from the CCX via the Infinity Fabric, so it's as large as the space available on the die.

[Franck]

45 minutes ago, Brooksie359 said:

Yeah it's too bad everyone is limited to running one application at a time. Oh wait you can run more than one thing at a time meaning this type of argument is silly. Also games are starting to utilize more cores as well and with the increase in core counts across the board prgramers will tailor their software to utilize this new hardware. 

I understand both side as i am developing both light and heavy threaded application with and without DirectX/OpenGL. When multi threaded is limited it's because it wouldn't improve beyond that amount of cores or the situation doesn't allow to run more than a certain amount as the following tasks require previous one to be completed.

 

But when it's not either of these cases it is very easy to make thing automatically scale. I have multiple softwares that even if i don't change anything in the source code and send it in the future where there are 512 core CPU it will actually utilize what it can and perform much better.

 

Main problem i encounter with 3D programming is i assume the same as what game developer encounter. It is the fact that you have to double your code if you support both OpenGL and DirectX as OpenGL is only single threaded and DirectX can be multithreaded so there is 2. I understand the shift into Vulkan that is slowly happening.

[rcmaehl]

Can't wait for 14nm# if C++++ if anything to go by.

[Taf the Ghost]

7 minutes ago, rcmaehl said:

Can't wait for 14nm# if C++++ if anything to go by.

14nm "sharp" is actually a good name.

[chckovsky]

5 hours ago, WereCat said:

What does the "+" even mean? 

 

Higher density, better yields, higher quality, lower power consumption,...? 

Better yields because of fewer defects, leading to higher clock speeds and/or less power consumption. My understanding is that this is why more 8th gen chips are able to hit 5GHz at more reasonable voltages (and therefore temperatures) than with previous 14nm iterations. IPC is supposed to be the same for all Intel 14nm chips, so that should not be a factor.

[Hunter259]

3 hours ago, Franck said:

I understand both side as i am developing both light and heavy threaded application with and without DirectX/OpenGL. When multi threaded is limited it's because it wouldn't improve beyond that amount of cores or the situation doesn't allow to run more than a certain amount as the following tasks require previous one to be completed.

 

But when it's not either of these cases it is very easy to make thing automatically scale. I have multiple softwares that even if i don't change anything in the source code and send it in the future where there are 512 core CPU it will actually utilize what it can and perform much better.

 

Main problem i encounter with 3D programming is i assume the same as what game developer encounter. It is the fact that you have to double your code if you support both OpenGL and DirectX as OpenGL is only single threaded and DirectX can be multithreaded so there is 2. I understand the shift into Vulkan that is slowly happening.

Most people don't understand that for a great amount of software there is no benefit to multithreading and in others it can actually make it slower to multithread. Then there is the limit of just how much multithreading you can do without harming performance etc. It will be interesting to see how API's and languages evolve to allow for higher multithreading without becoming a low level shit show.

[Kamina]

What happened, Intel?

[Drak3]

9 hours ago, WereCat said:

What does the "+" even mean? 

 

Higher density, better yields, higher quality, lower power consumption,...? 

Lower power consumption. It also leads to Kaby being able to clock 200MHz higher and Coffee to reach Kaby speeds with 2 extra cores.

[cj09beira]

8 hours ago, porina said:

As well as 14nm is doing now, those with a longer memory may recall that Intel's first 14nm architecture was delayed quite badly too. Broadwell did make its way into mobile space but on desktop, they only just avoided a paper launch by shipping two parts shortly before Skylake. On that note, maybe I should try to pick up an i7-5775C for fun...

 

For those wishing for software using more threads, do consider that not all software really needs it. We're certainly at the point where enthusiasts can use more than 4 cores, but that doesn't scale indefinitely. How many of us would really make good use of a 32 core processor if they had one? Maybe 6 to 8 are the new 4, and will be for some time to come until such time there is a real need for it in the consumer space.

it also clocked like shit, its as bad as glo fo's 14nm

[Stefan Payne]

11 hours ago, williamcll said:

Too bad most applications still rely on less than 4 threads. Especially non-AAA games.

That is either because it can run on a calculator and nobody needs/wants 50000fps or because the programmer is lazy/incompetent.

With most indy stuff its because it runs well anyway, even with the shittiest CPU you can get right now....

10 hours ago, Swatson said:

Oh good so Intel is finally slowly accepting they gonna be stuck on nodes forever now. More pluses is more better

Just waiting to see this come true.

 

Intel is going AMD in the 90s. 

People might not remember but AMD was always late to the game and had always problems with their fabs to get the new process going wich gave Intel the edge.

 

Right now, as AMD has no more Fabs and is free to choose the Fab, they suddenly have the edge 

10 hours ago, InertiaSelling said:

Tick - tock - tock - tock - tock - tock - tock - tock - tock ...

- Intel

Reminds me of that

 

9 hours ago, Mihle said:

They are most likely going TSMC for 7nm only untill GloFo 7nm is ready, and then use both. I think there was a rumor that said TSMC for Epyc and GloFlo for Ryzen/threadripper, if I remember right.

No, they are not.

They are only using TSMC 7nm regardless of the manufacturer....

 

Rumor said that AMD paid GF to license the TSMC process and use that.

 

Although GF has its own 7nm Process, AFAIR its the IBM. But since the last IBM processes were rather garbage for what AMD wants to do, they seem to avoid it...

9 hours ago, Swatson said:

If the rumours are true AMD is gonna spin 8core ccx zen2 in addition to the originally planned 6 core ccx....

I highly doubt that as it would be too expensive to do that as it is quasi a redesign of the chips.

And why would you do that for something like 10mm²??

You wouldn't!

 

So they will either stick with the 4 Core CCX and probably increase the bandwith between the CCX or reduce the latency or they go for one bigger one. 

But to be honest: I find it more likely that they just increase the CCX than to do something about the CCX.

Especially since it allows them for easier design of processors from 4 to 16 Cores...

2 hours ago, Kamina said:

What happened, Intel?

They became so big they had to fail and get crushed by the cost of the fab and the development cost for new processes.

 

And you can guess that they probably didn't expect AMD to make a CPU this good.

There is also the statement somewhere from AMD that they planned for the next Intel CPU and them to be more agressive - but the next Intel CPU was nowhere to be seen...

[williamcll]

TSMC must be swimming in money even after the virus setback.

[Drak3]

1 hour ago, Stefan Payne said:

That is either because it can run on a calculator and nobody needs/wants 50000fps or because the programmer is lazy/incompetent.

With most indy stuff its because it runs well anyway, even with the shittiest CPU you can get right now....

Or because games are heavily serial in nature and parallelization can only be done so much.

[Taf the Ghost]

Quote

Cooper Lake will be built on a 14nm that’s further enhanced from what’s being used in Cascade Lake. So, I guess, 14++++

https://twitter.com/Ashraf__Eassa/status/1029440772213563393

 

Haha.