Analysis: Haswell "Refresh" Is Not Broadwell

[Arachnotronic]

I don't think "Haswell Refresh" is Broadwell. My view of the matter is this:

Intel is focused on mobile, and as a result will probably be rushing to get anything that's battery life constrained to the newest process node. By this time next year, that will be the 14nm node, right? Now, I assume that 14nm won't be particularly mature, at least compared to the 22nm parts, so it will be imperative to get 14nm into the mobile markets, while the less power-conscious parts go to desktop/LGA. 22nm will be mature and can be used to service these stagnant/slowing markets.

Now, if you notice, Intel brings "Skylake" to LGA by 1H 2015, which is probably in-line with the BGA stuff. Why? Well, people didn't really receive Ivy Bridge too warmly, particularly as it had minimal ST performance increase over Sandy Bridge and at the same time didn't OC much better (if anything worse). So, it makes sense for the enthusiast crowd to simply release "tocks" since the "ticks" focus on IGP and power consumption that nobody on the LGA/enthusiast desktop side care about anyway.

What do you guys think? Am I full of it or does what I wrote make sense? Thoughts?

 

[VirtualLarry]

Intel is definitely slowing down their desktop focus, and instead, focusing on low-power and IGP for mobile (ultrabook, and possibly tablet, down the road).

It wouldn't surprise me in the least if they did that.

 

[crashtech]

At this point it might be said that Intel is waiting for 99% of software to catch up and maximize the use of CPU resources that are available today, so skipping every other refresh for the (niche) desktop kinda makes sense.

 

[inf64]

If you have looked at the latest leaked intel desktop roadmaps you would see intel really states what it is (same node-> Haswell with new chipset and a probably a speed bin bump). No need to do any analysis.

 

[Arachnotronic]

If you have looked at the latest leaked intel desktop roadmaps you would see intel really states what it is (same node-> Haswell with new chipset and a probably a speed bin bump). No need to do any analysis.

It was more of an attempt to explain the rationale, but yeah, you're right - it's pretty clear LGA 1150 doesn't get any Broadwell love.

Do you think we will get a Broadwell-E?

 

[inf64]

There is no reason for intel to stop doing what they have been doing for years now, so yep E version of Broadwell should be arriving in the future. I just don't know when they will opt for it and how will it align with their next big update to the ISA (AVX512 or whatever they call it; it cannot be in Broadwell since it's Haswell's die shrink and this means it's next gen uarchitecture that comes after Broadwell).

 

[Arachnotronic]

There is no reason for intel to stop doing what they have been doing for years now, so yep E version of Broadwell should be arriving in the future. I just don't know when they will opt for it and how will it align with their next big update to the ISA (AVX512 or whatever they call it; it cannot be in Broadwell since it's Haswell's die shrink and this means it's next gen uarchitecture that comes after Broadwell).

I'm thinking though that another part of this "Haswell Refresh" is to realign LGA 115x with LGA 20xx. The gap gets narrowed to basically 1 generation, rather than two on desktop.

 

[inf64]

Yes that makes sense too. Intel sees the problem though, it's the slowing down of the desktop market and acceleration of mobile.

 

[NTMBK]

I'm thinking though that another part of this "Haswell Refresh" is to realign LGA 115x with LGA 20xx. The gap gets narrowed to basically 1 generation, rather than two on desktop.

Depends how long it takes to verify the server parts and platform, really. The i7-E series is just a byproduct of the E5 Xeons, and those take priority over the enthusiast niche.

 

[dealcorn]

Ashraf's spin on this topic throws a meaty bone to the enthusiast community: http://seekingalpha.com/article/1503982-intel-s-14-nanometer-setting-the-record-straight .

 

[Arachnotronic]

Depends how long it takes to verify the server parts and platform, really. The i7-E series is just a byproduct of the E5 Xeons, and those take priority over the enthusiast niche.

Absolutely. It's unfortunate that Intel can't keep the "EP" parts in lock-step with the mainstream cores, but I understand that validation of a large, multi-socket capable part is much more difficult than the mainstream stuff.

 

[Arachnotronic]

Ashraf's spin on this topic throws a bone to the enthusiast community: http://seekingalpha.com/article/1503982-intel-s-14-nanometer-setting-the-record-straight .

Hey, that guy seems to agree with me...

 

[crashtech]

Ashraf's spin on this topic throws a bone to the enthusiast community: http://seekingalpha.com/article/1503982-intel-s-14-nanometer-setting-the-record-straight .

Seems right on the money to me. In fact, I would not be surprised if all desktop parts start following the "E" model soon, such that all desktop customers will all be shopping for cut down Xeons in a few years.

 

[TuxDave]

Hey, that guy seems to agree with me...

lol

 

[Bryf50]

Seems right on the money to me. In fact, I would not be surprised if all desktop parts start following the "E" model soon, such that all desktop customers will all be shopping for cut down Xeons in a few years.

That's been my thought as well. Even if the consumer desktop market becomes completely unprofitable there's always going to be a need to create higher performing parts for servers and workstations. I think we tend to slightly underestimated the enthusiast and pc gaming market. It should be large enough to warrant the trickle down of those high end xeon parts.

 

[Arachnotronic]

That's been my thought as well. Even if the consumer desktop market becomes completely unprofitable there's always going to be a need to create higher performing parts for servers and workstations. I think we tend to slightly underestimated the enthusiast and pc gaming market. It should be large enough to warrant the trickle down of those high end xeon parts.

Indeed. The bummer, though, is that mainstream gets the "better" cores first. I expect Skylake to bring us 512bit wide AVX, and while I don't expect this will have a near-term impact, the chips will be more "future proof" than the higher end "-E" parts based on Haswell.

 

[blackened23]

I'm thinking though that another part of this "Haswell Refresh" is to realign LGA 115x with LGA 20xx. The gap gets narrowed to basically 1 generation, rather than two on desktop.

I can definitely see this happening as well.

 

[Piroko]

So, it makes sense for the enthusiast crowd to simply release "tocks" since the "ticks" focus on IGP and power consumption that nobody on the LGA/enthusiast desktop side care about anyway.

What do you guys think? Am I full of it or does what I wrote make sense? Thoughts?

It does make sense. The biggest impact of IVB (its beefed up graphics unit) was with Ultrabooks and Ultramobiles, Broadwell will probably be in the same position.

Also, this. Worded a bit extreme, but so far, seems to be spot on. I'm still amazed that he managed to dig this out in November '12.

 

[beginner99]

If there only is a tock and the tock has a new socket then the whole point of a sockted CPU gets a bit useless too...yeah of course for extreme overclockers that kill there CPUs and such it's very useful but that's even more niche than "normal" enthusiasts.

Or: if you can't upgrade you CPU to something better the socket because pretty useless.

 

[crashtech]

I assume there will be different CPUs with their own price points, even in this "cut-down Xeon" scenario, not to mention the desire for enthusiasts to modify their CPUs physically, so I would predict that the socket is here to stay for a while yet. Besides, in a server environment, the ability to swap out parts is cherished, and our parts will presumably be trickling down from that environment.

 

[IntelUser2000]

Indeed. The bummer, though, is that mainstream gets the "better" cores first. I expect Skylake to bring us 512bit wide AVX, and while I don't expect this will have a near-term impact, the chips will be more "future proof" than the higher end "-E" parts based on Haswell.

Actually I expect Intel to follow this path sooner or later.

http://images.anandtech.com/reviews/tradeshows/IDF/2005/Spring/Day3/FutureCPU/evolution.jpg
http://images.anandtech.com/reviews/tradeshows/IDF/2005/Spring/Day3/FutureCPU/vision.jpg

So "512-bit AVX" part may be either their iGPU with powerful DP capable FPU, or even few of MIC(Xeon Phi) cores integrated.

You would have traditional 256-bit FMA AVX2 capable cores and an option of using specialized units in the form of GPU or Xeon Phi-like cores.

Even if the consumer desktop market becomes completely unprofitable there's always going to be a need to create higher

Right now I don't think the de-focus on Desktop is much of being profitable or not but rather "should we sacrifice the perf/watt gains relevant for iGPU/Servers/Tablets/Ultrabooks for extra performance for the shrinking market of Desktop and enthusiast?". Obviously the answer would be no.

 

[Hulk]

I have to say I'm disappointed that 14nm Broadwell will not happen until 2015.
But why?

Is it because they are having a lot of problems with the 14nm process?

Or is it because they don't feel the need to stay on the yearly "tick-tock" schedule due to lack of competition?

I have a feeling it's the former. 14nm seems to be the point where they can really start competing with ARM so why would they put that off when all of the their focus as of late has been on efficiency?

With all of that being said if they offered a 6/12 core/thread Haswell socket 1150 "K" series part for the price of the 4770K they could buy me off

 

[IntelUser2000]

I have to say I'm disappointed that 14nm Broadwell will not happen until 2015.
But why?

Because its Desktop. Broadwell(not "Haswell Refresh") is coming to mobile earlier.

 

[USER8000]

Haswell has only just stated to appear in laptops and even the new generation 22NM Atom won't be out until the end of the year or early next year. So in reality,are we going to see Broadwell in retail computers in quantity before the middle of 2014?? After all they would need to make sure they have cleared enough of their Haswell inventory first,and replacing Haswell in under a year seems a tad far fetched.

If anything its new generation Atom,which is probably more important for Intel so they can better compete in phones and mainstream tablets.

 

[Arachnotronic]

I have to say I'm disappointed that 14nm Broadwell will not happen until 2015.
But why?

Is it because they are having a lot of problems with the 14nm process?

Or is it because they don't feel the need to stay on the yearly "tick-tock" schedule due to lack of competition?

I have a feeling it's the former. 14nm seems to be the point where they can really start competing with ARM so why would they put that off when all of the their focus as of late has been on efficiency?

With all of that being said if they offered a 6/12 core/thread Haswell socket 1150 "K" series part for the price of the 4770K they could buy me off

Broadwell is coming in 2014, just not for desktops. Desktops go straight to Skylake in 2015.