Broadwell-EX/EP disappeared from Intel's enterprise roadmaps

[zMeul]

source: http://www.bitsandchips.it/52-english-news/5807-broadwell-ex-ep-disappeared-from-latest-enterprise-intel-roadmaps
 

Some of our internal sources told us that the Enterprise declination of Broadwell architecture is recently disappeared from the latest Intel roadmaps. The Enterprise CPUs will go directly from Haswell to Skylake

 

 
---
 
this is quite the surprise because it confirms Broadwell architecture is a failure; desktop models are limited to just two: i5-5675C and i7-5775C

 

there are speculations that the giant chipmaker may introduce its code-named “Purley” server platform based on “Skylake-EX” and “Skylake-EP” processors earlier than originally expected

 

also, there are rumors customers are canceling their Xeons orders in favor of waiting for Skylake: https://goparallel.sourceforge.net/16-petaflop-supercomputer-on-hold-to-wait-for-skylake-xeon/

[Curufinwe_wins]

source: http://www.bitsandchips.it/52-english-news/5807-broadwell-ex-ep-disappeared-from-latest-enterprise-intel-roadmaps

---

this is quite the surprise because it confirms Broadwell architecture is a failure; desktop models are limited to just two: i5-5675C and i7-5775C

there are speculations that the giant chipmaker may introduce its code-named “Purley” server platform based on “Skylake-EX” and “Skylake-EP” processors earlier than originally expected

also, there are rumors customers are canceling their Xeons orders in favor of waiting for Skylake: https://goparallel.sourceforge.net/16-petaflop-supercomputer-on-hold-to-wait-for-skylake-xeon/

I mean this isn't news considering leaked roadmaps had dropped broadwell-e since well before broadwell launched. A failure isn't the right term I would say, but since we are going to be on 14nm for a while (likely over two years) I wouldn't mind getting the new architecture right away instead of being stuck on haswell meets 14nm (which imho was always a bit weird).

Basically since 14nm production yields took so long to develop, their new architecture was basically ready right away. Aka skylake didn't face the same delays broadwell did and so the gimped release of broadwell is the reason skylake is going to go balls deep.

[qwertywarrior]

this might be good we are skipping a gen and we can get back on track with the same generation as the mainstream variants hopefully

[linuxfan66]

this might be good we are skipping a gen and we can get back on track with the same generation as the mainstream variants hopefully

I honestly don't even know why broad well was launched when it was going to be replaced so quickly...

[EpicAdom]

I honestly don't even know why broad well was launched when it was going to be replaced so quickly...

Because...y not? I guess?

[qwertywarrior]

I honestly don't even know why broad well was launched when it was going to be replaced so quickly...

very low demand from distributors/OEM because they still had large stocks of haswell chips

performance isnt increase with each new generation because intel is refusing to use larger dies, in fact die sizes have been decreasing in time.

just go look at the die sizes since sandybridge, Intel is cost saving, if they keep doing this its gonna be pointless buying new CPUs

Zen might just give us one more great CPU and keep intel on its toes

[bcredeur97]

just gonna sit here with my ivy and wait for something big to happen that is worth an upgrade....

 

I do have broadwell laptop though. Power savings is nice for mobile chips

[Norman_the_Owl]

That's a good thing

 

That means the X series and Z series will be on the same first number.

 

6700k will be skylake and so will the 6969x ( ͡° ͜ʖ ͡°)

[Deli]

I'm afraid with so little LGA Broadwell are produced. They will become collectable in the future. 

[CTR640]

That's a good thing

 

That means the X series and Z series will be on the same first number.

 

6700k will be skylake and so will the 6969x ( ͡° ͜ʖ ͡°)

[Norman_the_Owl]

but i wasn't laughing ( ͡° ͜ʖ ͡°)

[CTR640]

but i wasn't laughing ( ͡° ͜ʖ ͡°)

[patrickjp93]

very low demand from distributors/OEM because they still had large stocks of haswell chips

performance isnt increase with each new generation because intel is refusing to use larger dies, in fact die sizes have been decreasing in time.

just go look at the die sizes since sandybridge, Intel is cost saving, if they keep doing this its gonna be pointless buying new CPUs

Zen might just give us one more great CPU and keep intel on its toes

Before you go running your mouth, you might want to actually study the intricacies which govern CPU performance. It doesn't matter if Intel builds a bigger core. 4-way or 8-way SMT could be done, but no software on the consumer side uses that many threads simultaneously anyway. Widening the Out of Order engine wouldn't do much when statistically every 7-9th line of code is a branch. If you deepen the pipeline you also make the results of a branch miss more penalizing. So regardless of what Intel does, making the core bigger on its own will do nothing for you. Beyond that, most of the instructions used in consumer software today are optimized very close to their theoretical limits of possible instruction latencies. http://www.agner.org/optimize/instruction_tables.pdf

 

Software must evolve. Intel already did all the ground work providing new instruction set extensions and standards to use them easily, not to mention all the work which went into building multithreading standards and APIs to make it easy. Intel deserves no flak for your meager performance gains in legacy software modes. Almost all of the blame should be put on software companies themselves. The rest belongs squarely on Microsoft's shoulders for having such ancient legacy support for their Operating Systems. Even Windows 10 works with a Pentium IV. At this point not jumping up to at least Nehalem is astounding to me. Who the bloody F*CK is still using a P4?

[zMeul]

Who the bloody F*CK is still using a P4?

my sister's PC is a P4 derived Celly  :rolleyes:

will she jump from Xp to W10? most likely, not

[linuxfan66]

Before you go running your mouth, you might want to actually study the intricacies which govern CPU performance. It doesn't matter if Intel builds a bigger core. 4-way or 8-way SMT could be done, but no software on the consumer side uses that many threads simultaneously anyway. Widening the Out of Order engine wouldn't do much when statistically every 7-9th line of code is a branch. If you deepen the pipeline you also make the results of a branch miss more penalizing. So regardless of what Intel does, making the core bigger on its own will do nothing for you. Beyond that, most of the instructions used in consumer software today are optimized very close to their theoretical limits of possible instruction latencies. http://www.agner.org/optimize/instruction_tables.pdf

Software must evolve. Intel already did all the ground work providing new instruction set extensions and standards to use them easily, not to mention all the work which went into building multithreading standards and APIs to make it easy. Intel deserves no flak for your meager performance gains in legacy software modes. Almost all of the blame should be put on software companies themselves. The rest belongs squarely on Microsoft's shoulders for having such ancient legacy support for their Operating Systems. Even Windows 10 works with a Pentium IV. At this point not jumping up to at least Nehalem is astounding to me. Who the bloody F*CK is still using a P4?

From what I've heard multithreaded coded works against human thinking practices. That said I won't defend game programmers who spend years on cathedrals while releasing a broken game...

[patrickjp93]

From what I've heard multithreaded coded works against human thinking practices. That said I won't defend game programmers who spend years on cathedrals while releasing a broken game...

I would disagree, but I have Asperger's and have always excelled at mathematical reasoning. I have a blog with a basic template of how to set up a multithreaded game and an HPC blog where I discuss optimization both in multithreaded and single-threaded scenarios that I'm trying to extend without going off the deep end on a rant about Microsoft (though I may just have to publish that page and let everything think what they wish). http://linustechtips.com/main/blog/684/entry-854-start-multithreaded-and-separated/

 

Now, working in assembly is very much contrary to human thinking systems, and it takes a special sort of nut case to be able to build a complex system in assembly these days, especially in x86. With 6502 on the Atari you only had I think 16 instructions, and you knew how many clocks each instruction took for the Atari 2600/5200/7800, NES, and SNES. You were given very simple communication protocols to follow to get the graphics chip to do something (change the color of pixels basically). I built 2 Atari 2600 video games in 3 weeks spending 8+ hours a day coding and trying things for a winter term class. That was a nightmare, but it was fun. Multithreading is easy on its face. And it's not like games actually have so many moving parts at the system level for there to be an excuse. On the GPU rendering side, yes, very complicated. The stuff the CPU should be doing is very simple unless you want to go balls to the wall on game AI or are running an MMO and need to do lots of socket communications.

[zMeul]

Multithreading is easy on its face. And it's not like games actually have so many moving parts at the system level for there to be an excuse. On the GPU rendering side, yes, very complicated. The stuff the CPU should be doing is very simple unless you want to go balls to the wall on game AI or are running an MMO and need to do lots of socket communications.

but rendering falls on the API and the driver, so developer doesn't have to get into that

[patrickjp93]

but rendering falls on the API and the driver, so developer doesn't have to get into that

The developer has to know how to use the API, and according to Nvidia a lot of games ship with fundamentally broken code that they just find workarounds for on the driver side. I doubt Nvidia's lying given its drivers are bigger pieces of software than the Windows 7 OS itself. So in the end if games aren't fully utilizing AMD's hardware, it comes down to the developers being careless and/or AMD not writing its drivers in such a way as to let common code paths do more with their hardware. AMD already translates all the DX kernels to compute workloads anyway, so it's not as if this is beyond possibility.

[Curufinwe_wins]

Btw, it seems people really underestimate how incredible the advancements in compilers has been over the last decade.

[patrickjp93]

Btw, it seems people really underestimate how incredible the advancements in compilers has been over the last decade.

Well, they're incredible if you're not Microsoft's compiler which still doesn't have full C++11 support.

[LucidMew]

Btw, it seems people really underestimate how incredible the advancements in compilers has been over the last decade.

Yeah, but are they cleverer than Mel?

[Curufinwe_wins]

Yeah, but are they cleverer than Mel?

Backwards compatibility is one of the single most limiting factors in the performance of moderns cpus.

[LAwLz]

just gonna sit here with my ivy and wait for something big to happen that is worth an upgrade....

 

I do have broadwell laptop though. Power savings is nice for mobile chips

It's pretty sad that it's been like 4 years with next to no CPU performance improvements from Intel.

Lower power usage and better integrated graphics are nice and all, but Intel haven't really given desktop users with a Sandy Bridge processor or newer any reason to upgrade.

 

I wish they would give us unlocked 6 core processors on LGA 1152 or whatever whey will call the next socket. They could drop all their dual cores while they were at it as well.

[Curufinwe_wins]

It's pretty sad that it's been like 4 years with next to no CPU performance improvements from Intel.

Lower power usage and better integrated graphics are nice and all, but Intel haven't really given desktop users with a Sandy Bridge processor or newer any reason to upgrade.

I wish they would give us unlocked 6 core processors on LGA 1152 or whatever whey will call the next socket. They could drop all their dual cores while they were at it as well.

Really Intel hasn't given you any reason to upgrade? Because Intel makes the software that uses legacy shitty instruction sets and doesnt even tax the cpu? And Intel definitely must be responsible for dx11 basically not scaling past two cores....

They have done their shit, few software developers have given two shits about it and continue to press the same poorly optimized crap down consumers throats.

[LAwLz]

Really Intel hasn't given you any reason to upgrade? Because Intel makes the software that uses legacy shitty instruction sets and doesnt even tax the cpu? And Intel definitely must be responsible for dx11 basically not scaling past two cores....

They have done their shit, few software developers have given two shits about it and continue to press the same poorly optimized crap down consumers throats.

Nope, Intel hasn't given me any reasons to upgrade. If they made a 6 core processor for let's say ~300 dollars on their consumer platform then I would get it in a heart beat.