Intel Broadwell

[ksec]

I thought i may as well start a new thread on Broadwell instead of posting it in Haswell threads.

The graph on Anand's Haswell Review about Process, and Architecture isn't very clear. But the Basic is the same. Intel normally do a ~2xx Die Size, with a mature process, and a new architecture. This year it didn't happen. Or at least the those die areas dont belong to the CPU side at all.

Are we likely to get 6 Core with Broadwell? Broadwell wont't provide many more performance benefits then Haswell. Assuming a perfect scaling, GT2 with Broadwell on 14nm, @ less then 100mm2 area it is going to the smallest die size CPU in decades.

Not everyone wants a mid level iGPU. As many would surely prefer a dedicated GPU solution. And Xeon could do some help with 2 extra Core count.

So to cut it short, @ 14nm Intel has alot of die space to work with, Are we going to get a 6 Core CPU?

 

[Headfoot]

I really hope they go for 8-core Haswell-E before Broadwell even. Until you can use the iGPU's resources as if they were in-core executions units, I'm not excited about using die space budget for iGPU over cores on desktop.

The die space is already there, I really think it's just that Intel is dragging its feet because of the relative lack of competition in the high-end space. Unfortunate, really

 

[Ajay]

I really hope they go for 8-core Haswell-E before Broadwell even. Until you can use the iGPU's resources as if they were in-core executions units, I'm not excited about using die space budget for iGPU over cores on desktop.

The die space is already there, I really think it's just that Intel is dragging its feet because of the relative lack of competition in the high-end space. Unfortunate, really

Haswell-E will come out well after Broadwell. Given the temperature problems with Haswell, Broadwell will probably still be 4 cores. Intel will likely opt for more graphics die space instead.

 

[Dave3000]

An 8-core Intel consumer-level CPU would be nice if clock speeds are not sacrificed to get 8 cores, at least no lower clock speeds than a 6-core CPU.

 

[ShintaiDK]

Broadwell will be quadcores too on LGA1150. The first chance of more will be with Skylake.

 

[NTMBK]

I wouldn't put too much hope in there being a desktop Broadwell.

 

[R0H1T]

Nope definitely not in fact if it were possible for Intel they'd cram the current quad core performance into a dual core(with HT) & slap an IGP onto the rest of the 90% die space left :biggrin:

 

[ksec]

So by the looks of things we wont even get 14nm in 2014?

Even at 8 Core for 14nm, without the iGPU the die space will still only be ~180, that is not large at all and Intel definitely make a huge margin on it.

 

[R0H1T]

So by the looks of things we wont even get 14nm in 2014?

Even at 8 Core for 14nm, without the iGPU the die space will still only be ~180, that is not large at all and Intel definitely make a huge margin on it.

Intel should be on track for the full node transition in 2014 however whether they'll release desktop parts first or the Silvermont successor is anyone's guess but the next 12 months will definitely dictate their pace of development going forward because even Intel cannot sustain that amount of R&D costs especially with shrinking revenues !

 

[ShintaiDK]

Intel should be on track for the full node transition in 2014 however whether they'll release desktop parts first or the Silvermont successor is anyone's guess but the next 12 months will definitely dictate their pace of development going forward because even Intel cannot sustain that amount of R&D costs especially with shrinking revenues !

Their CPU revenue didnt go down much. I think you got them mixed up with with AMD. They however took a massive beating:

 

[Lepton87]

If we want more then 4-cores we can only use their server platform aka socket 2011 and even that is a joke. They won't even sell us a not crippled die in an EE CPU. Extreme should be extreme, tdp shouldn't matter but the real reason is that they don't want to sell their 8 cores cpus clocked that fast for a 1000$ and they are unwilling to change the price of EE line so we are stuck with CPUs which are only Extreme in prices. EE used to be really extreme like gallatin now it's only a few mb of more cache, it's debatable if it really deserves being called an extreme CPU.

 

[R0H1T]

Their CPU revenue didnt go down much. I think you got them mixed up with with AMD. They however took a massive beating:

No I meant what I said, Intel's revenues were down 1% but I bet their R&D costs went upwards by atleast 10% then the fact that PC market is in a downward spiral & 2013/14 are gonna be much worse than 2012, Intel's development corpus is still gonna increase regardless, so unless they get a strong foothold in the mobile/tablet market be ready to witness double digit negative growth in their top &/or bottom line this year !

 

[mrmt]

Haswell-E will come out well after Broadwell. Given the temperature problems with Haswell, Broadwell will probably still be 4 cores. Intel will likely opt for more graphics die space instead.

Haswell is 22nm, Broadwell is 14nm, how can you relate the two designs in a parameter where process node plays a pivotal role?

 

[mrmt]

Intel should be on track for the full node transition in 2014 however whether they'll release desktop parts first or the Silvermont successor is anyone's guess but the next 12 months will definitely dictate their pace of development going forward because even Intel cannot sustain that amount of R&D costs especially with shrinking revenues !

Regarding funding and cost structure, they should now be working on Sky-Lake successor, and the chip beyond Sky-Lake sucessor should be on the concept phase, meaning that this is the next chip which you can change the scope.

Sky-lake successor should be a 10nm chip, meaning that if a problem occurs with the node, they might postpone or scrap the thing, but they will have to eat the R&D bill anyway.

So unless you are implying an AMD-like shakedown, Intel R&D spending can be taken for granted at least until 2017 if they can field 10nm, or 2015 if they wipe out CAPEX for 10nm.

I don't think they will go for an AMD-like shakedown. You don't cut the brains of your company except in the most terrible situations.

 

[CakeMonster]

I would guess Intel is going to ride out one more generation of low IPC increases, and devote most the additional transistors to IGP in Broadwell. Possible some of them in the form of cache/ondie memory.

Despite what people here say (and despite what I want) this is what most people have use for these days. People need graphics performance more than CPU performance in their current devices, and Intel is probably correct that this will be valid for the lifetime of Broadwell (mid/late 2015).

After that (Skylake 2015) there is a much greater chance of some major overhauls. Keep in mind that Skylake will still be 14nm like Broadwell, but the demand for 6 or 8 cores is probably much higher then and would be my best bet. If not then some form of cache/memory would be a good use for the transistors.

With Skymount (2016) (10nm) we will definitely see more cores. Whether they go for 6 or 8 is also an interesting debate. I would consider 6 as a possible intermediary, but I won't expect it to last for long until they go for 8, considering how widespread it will become on other platforms in the mean time.

 

[R0H1T]

Regarding funding and cost structure, they should now be working on Sky-Lake successor, and the chip beyond Sky-Lake sucessor should be on the concept phase, meaning that this is the next chip which you can change the scope.

Sky-lake successor should be a 10nm chip, meaning that if a problem occurs with the node, they might postpone or scrap the thing, but they will have to eat the R&D bill anyway.

So unless you are implying an AMD-like shakedown, Intel R&D spending can be taken for granted at least until 2017 if they can field 10nm, or 2015 if they wipe out CAPEX for 10nm.

I don't think they will go for an AMD-like shakedown. You don't cut the brains of your company except in the most terrible situations.

I'm not sure what you're referring to but as far as my previous comment is concerned I expect Intel to significantly slowdown their desktop roadmap in favor of the mobile market. Desktops as it is hardly make half of the earnings as compared to say the server market, so going forward more emphasis on mobile/tablets then server & lastly your avg PC's D:

This is what I see as being the case, given Intel's recent product mix, & they may even leave the high end enthusiast market altogether certainly as with the time they took to release IVB-E :\

Now the core microarch is their last major revamp, since Netburst, & I foresee their x86 market ending with it, albeit minor performance bump like SNB - IVB - Haswell ought to continue because their server market depends on the advances they can extract from this aging microarch. I'm interested in seeing the performance gains that can be had with Silvermont & if it can be slightly tweaked to accommodate all the three major markets i.e. mobile, desktop, server especially in the long run !

 

[JimmiG]

Broadwell may be SoC/Ultrabook CULV only

 

[mrmt]

I'm not sure what you're referring to but as far as my previous comment is concerned I expect Intel to significantly slowdown their desktop roadmap in favor of the mobile market. Desktops as it is hardly make half of the earnings as compared to say the server market, so going forward more emphasis on mobile/tablets then server & lastly your avg PC's D:

Desktop roadmap? Desktop is dead. What Intel - and AMD - is selling today is mobile chips on the mainstream, and server chips at the high end. If Intel is already selling mobile chips in the desktop market today, why can't it sell in the future? And why wouldn't it launch Broadwell on desktops? At worst they would save some money by selling smaller dies.

This is what I see as being the case, given Intel's recent product mix, & they may even leave the high end enthusiast market altogether certainly as with the time they took to release IVB-E :\

There is no IVB-E die, what you have is some IVB-EP dies being sold as IVB-E. And because of that, the critical path isn't the development of the core itself, but the validation of the die for server duty, which is far more complicated than validation of a server part.

 

[R0H1T]

Desktop roadmap? Desktop is dead. What Intel - and AMD - is selling today is mobile chips on the mainstream, and server chips at the high end. If Intel is already selling mobile chips in the desktop market today, why can't it sell in the future? And why wouldn't it launch Broadwell on desktops? At worst they would save some money by selling smaller dies.

If I were to paraphrase the "desktop or PC market of today includes desktop plus notebooks" because individually they're just a shrinking blip at this point in time certainly as compared to all the other computing devices out there.

There is no IVB-E die, what you have is some IVB-EP dies being sold as IVB-E. And because of that, the critical path isn't the development of the core itself, but the validation of the die for server duty, which is far more complicated than validation of a server part.

I know that & correct me if I'm wrong here but SB-E was the same isn't it ? This is similar to what Nvidia did with the Titan, as in selling gimped Tesla 20X for general/retail users, but my emphasis was on the assumption that the days of 1000$ CPU's are coming to an end & whatever Intel decides to release as an enthusiast grade desktop CPU in the next couple of years could well be the last of its kind !

 

[Imouto]

Their CPU revenue didnt go down much. I think you got them mixed up with with AMD. They however took a massive beating:

While other players are growing silly.

Interesting fact is that you brought AMD by yourself into this thread without a single reason.

 

[VirtualLarry]

So to cut it short, @ 14nm Intel has alot of die space to work with, Are we going to get a 6 Core CPU?

Doubtful. Instead, Intel will be laughing all the way to the bank.

Expect GMs to push 65%.

 

[Edrick]

So to cut it short, @ 14nm Intel has alot of die space to work with, Are we going to get a 6 Core CPU?

Broadwell is a die shrink of Haswell. Making 6 core version would be more than a die shrink. And Broadwell will bring us the 5770K with a TDP of below 75W, and a faster IGP.

As others have said, Skylake is your first real hope of 6-8 core CPUs.

 

[mrmt]

If I were to paraphrase the "desktop or PC market of today includes desktop plus notebooks" because individually they're just a shrinking blip at this point in time certainly as compared to all the other computing devices out there.

That's a hyperbole here, the PC market isn't small by any standards.

I said that the desktop market is dead because the trade off between mobile performance and desktop performance isn't as big as it was in the past. In the past notebooks were orders of magnitude slower than desktops, but today, a 47W Haswell is no slouch compared to the 4770k. The benefits of going from 47W to 84W aren't that pronounced anymore. Top over 100W and the benefits are even more reduced.

In the end, it's far more cost effective to design a line up for mobile and then factory overclock them to desktops.

I know that & correct me if I'm wrong here but SB-E was the same isn't it ? This is similar to what Nvidia did with the Titan, as in selling gimped Tesla 20X for general/retail users, but my emphasis was on the assumption that the days of 1000$ CPU's are coming to an end & whatever Intel decides to release as an enthusiast grade desktop CPU in the next couple of years could well be the last of its kind !

Yes, it's something like that. For most folks outside AMD marketing bubble 4 fast cores is enough, and for the folks that really need more than 4 cores, they will have to shell out money for bigger, expensive downgraded server parts.

 

[mikk]

Broadwell will be quadcores too on LGA1150. The first chance of more will be with Skylake.

Broawell won't come for LGA. Next LGA Mainstream CPU is Skylake.

 

[Edrick]

Broawell won't come for LGA. Next LGA Mainstream CPU is Skylake.

Source?