[Digitimes] Xeon Roadmap, Skylake launches in Q3'15

[RussianSensation]

So as I understand this, Intel decided to artificially delay Skylake after all?

BW-E delayed to 2016 and now Skylake delayed to late Q3/early Q4 2015. I guess it's better to have knowledge of these changes. Pretty much means I have at least a year before I have to think about a desktop CPU upgrade and a possible laptop upgrade, which means by that point my IVB i7 laptop will be almost 3 years old and my i5 2500K nearly 5 years old (!). Pretty crazy to think there will not be a real tangible upgrade from a 3.4Ghz IVB i7 on the mobile side for nearly 3 years. On the positive side this gives more time for SSD prices to drop (maybe 1TB Samsung 850 EVO / Crucial MX200 drive for $250-275 by then) and for mobile dGPU power to increase even more beyond 970M/980M.

I had a feeling Skylake-K would be delayed though given how late Intel is with Broadwell-K and after Intel announced the delay of BW-E.

 

[Fjodor2001]

It makes it sound like Intel is doing it for evil reasons instead of normal business practice of any company anywhere. That's why.

Yes, to me there's a difference between delaying a product because of technological reasons (development did not finish in time) compared to purely economical (intentionally delaying a product despite it being ready to make more money). I consider the latter more evil, and just shows there's a close to monopoly situation that Intel is taking advantage of. Intel is free to do so, but I as a customer am also free to consider it evil and lower my emotions of goodwill towards the company in question.

We really need some competition to get things moving again...

 

[kimmel]

Yes, to me there's a difference between delaying a product because of technological reasons (development did not finish in time) compared to purely economical (intentionally delaying a product despite it being ready to make more money). I consider the latter more evil, and just shows there's a close to monopoly situation that Intel is taking advantage of. Intel is free to do so, but I as a customer am also free to consider it evil and lower my emotions of goodwill towards the company in question.

We really need some competition to get things moving again...

Intel sells chips to companies like Dell, Lenovo, HP, Asus. Assuming these customers tell Intel that they won't buy Skylake or have systems ready for Skylake if it is released on a certain date. Intel of course will move the date. I'm not entirely sure why Intel would be at fault here. You are misunderstanding where the goodwill would be lost if they went ahead and released without their biggest customers being on board.

What point would releasing Skylake be without the support from the industry? Makes no sense at all from a buisness point of view.

 

[Lepton87]

BW-E delayed to 2016 and now Skylake delayed to late Q3/early Q4 2015. I guess it's better to have knowledge of these changes. Pretty much means I have at least a year before I have to think about a desktop CPU upgrade and a possible laptop upgrade, which means by that point my IVB i7 laptop will be almost 3 years old and my i5 2500K nearly 5 years old (!). Pretty crazy to think there will not be a real tangible upgrade from a 3.4Ghz IVB i7 on the mobile side for nearly 3 years. On the positive side this gives more time for SSD prices to drop (maybe 1TB Samsung 850 EVO / Crucial MX200 drive for $250-275 by then) and for mobile dGPU power to increase even more beyond 970M/980M.

I had a feeling Skylake-K would be delayed though given how late Intel is with Broadwell-K and after Intel announced the delay of BW-E.

I bought my mobo with the intention to swap the CPU for an 8-core BW-E. Is there any info when Skylake-E is going to be released? If it's within 6 months of BW-E then BW-E doesn't make sense and it's worth it to just wait for Skylake-E. Coincidentally BW-K may make little sense as well if skylake ends up clocked at 4GHz/4.4GHz turbo and BW-E overclocks under 5GHz on average. Assuming that the IPC increase is on the order of 10% over BW-E as expected. That would mean 4.85GHz(ST)/4.4GHz(MT) worth of BW performance. What do you guys think? Are IPC gains lesser or is skylake clocked lower then my prediction or both? If it is as I described BW-K would mostly make sense to extreme overclockers that certainly don’t use air-cooling. All of that entices enthusiast to move to the enthusiast platform.

 

[Fjodor2001]

Intel sells chips to companies like Dell, Lenovo, HP, Asus. Assuming these customers tell Intel that they won't buy Skylake or have systems ready for Skylake if it is released on a certain date. Intel of course will move the date. I'm not entirely sure why Intel would be at fault here. You are misunderstanding where the goodwill would be lost if they went ahead and released without their biggest customers being on board.

What point would releasing Skylake be without the support from the industry? Makes no sense at all from a buisness point of view.

Let me put it this way: If there was competition so e.g. AMD would have a superior product on the market, Intel would release Skylake as fast a possible no matter what. I guess AMD is as much to blame for this as well.

If there's anything good about these delays it's that it will allow AMD to stay in the market and maybe do a comeback with Zen on 14 nm in 2016 so we get some competition again. If Intel would be so far ahead of AMD that AMD would go bankrupt things would be even worse.

Let's hope Intel forgetting Andrew Grove's statement "Only the paranoid survive" will come back and bite them in the ass some day...

 

[Ajay]

Yes, to me there's a difference between delaying a product because of technological reasons (development did not finish in time) compared to purely economical (intentionally delaying a product despite it being ready to make more money). I consider the latter more evil, and just shows there's a close to monopoly situation that Intel is taking advantage of. Intel is free to do so, but I as a customer am also free to consider it evil and lower my emotions of goodwill towards the company in question.

It's not evil. Intel, as a for profit company, needs to, at a minimum, recoup it's investment in BW. They will be selling both SKL & BW at the same time, using an overlapping product segmentation strategy to make the most profit they can. It's annoying, but at least as an informed consumer - you can make the best choice for your needs.

We really need some competition to get things moving again...

Well, if Mubadala is willing to invest an extra $50B in GF & AMD over the next 5 years and poach a lot of talent from the best companies - maybe that could happen. But they won't, so I don't see it happening.

If Apple & Samsung really partnered up on ARM, they would have a good chance of creating a competitive line-up from phones to notebooks, but I don't see that happening either.

Intel is at the top of their game right now, and that means the ROI for an assault on a chunk of Intel's marketshare just doesn't exist.

 

[RussianSensation]

I bought my mobo with the intention to swap the CPU for an 8-core BW-E. Is there any info when Skylake-E is going to be released? If it's within 6 months of BW-E then BW-E doesn't make sense and it's worth it to just wait for Skylake-E. Coincidentally BW-K may make little sense as well if skylake ends up clocked at 4GHz/4.4GHz turbo and BW-E overclocks under 5GHz on average. Assuming that the IPC increase is on the order of 10% over BW-E as expected. That would mean 4.85GHz(ST)/4.4GHz(MT) worth of BW performance. What do you guys think? Are IPC gains lesser or is skylake clocked lower then my prediction or both? If it is as I described BW-K would mostly make sense to extreme overclockers that certainly don’t use air-cooling. All of that entices enthusiast to move to the enthusiast platform.

Intel Workstation platform is 1-1.5 years behind mainstream platform. For example, Haswell launched June 1, 2013, but HW-E August 29, 2014. Since BW-E is now Q1 2016, and Skylake is Q3/4 2015, Skylake-E earliest is Q3 2016, more likely Q4 2016-Q2 2017.

As far as Skylake goes, it should be a way bigger upgrade than BW. That's because the platform ditches VIR, brings PCI 4.0, AVX 3.2, and more mature 14 node than BW. Without VIR, I think it will overclock better than BW. Even without overclocking it seems Skylake will threaten the relevance of BW, which is the reason for the delay. If you are going to swap in an 8-core BW-E, I don't think you need to worry about how good Skylake is. Someone who spends $1000 on a CPU alone is at the cutting edge all the time so if Skylake is way better, you'll upgrade again anyway

 

[Lepton87]

As far as Skylake goes, it should be a way bigger upgrade than BW. That's because the platform ditches VIR, brings PCI 4.0, AVX 3.2, and more mature 14 node than BW. Without VIR, I think it will overclock better than BW. Even without overclocking it seems Skylake will threaten the relevance of BW, which is the reason for the delay. If you are going to swap in an 8-core BW-E, I don't think you need to worry about how good Skylake is. Someone who spends $1000 on a CPU alone is at the cutting edge all the time so if Skylake is way better, you'll upgrade again anyway

Well, I was hoping that 8 cores would trickle down to K series, even if only for the more expensive model. I don't want an X CPU, K model for the price of 5930K I could swallow but 1k$ not so much.

 

[Idontcare]

Yes, to me there's a difference between delaying a product because of technological reasons (development did not finish in time) compared to purely economical (intentionally delaying a product despite it being ready to make more money).

You do realize "delaying a product because of technological reasons" is also a "purely economical" decision, right?

IBM's decision to not invest $5B into developing an Intel-competing x86 processor (they have an x86 license) is "purely economical".

Apple's decision to not invest $10B into making Cyclone be a Haswell killer was "purely economical".

AMD's shareholder's decision to not invest another $2B into R&D to bolster the competitiveness of their existing product offerings was a "purely economical" decision.

For every argument you would make against Intel's decision makers, you cannot expect us to take you serious if you fail to recognize and acknowledge the same metric applies to every other business out there which voluntarily chose to not invest in developing a competing product because they too would rather maximize the returns on their investments to better enrich themselves and their shareholders.

In short, you have positioned yourself as having an ax to grind against any and every "for profit" business and employee on the planet because they are not wantonly wasting resources simply to pursue a product release timeline that is more to your personal liking.

But you can't really be espousing such an absurd viewpoint, are you?

 

[Fjodor2001]

In that case Intel should have though about that earlier, and invested less in R&D so Skylake would be ready at a later time. Now it will be ready several quarters ahead of actual launch, which is pointless. That is bad business.

 

[Fjodor2001]

On second thought maybe you are right.

Actually, what's the point of Intel even releasing desktop Broadwell at this point? I think we should be stuck with desktop Haswell from 2013Q2 until 2016/2017 or so. After all, there's no competition so what's the hurry. We should cheer them on and applaud them in such a decision.

 

[TechFan1]

There 10nm might be a factor too. They might be timing the release of Skylake to be about 1 year from when they expect to be able to release the 10nm shrink.

 

[III-V]

In that case Intel should have though about that earlier, and invested less in R&D so Skylake would be ready at a later time. Now it will be ready several quarters ahead of actual launch, which is pointless. That is bad business.

Not being able to tell the future is bad business?

On second thought maybe you are right.

Actually, what's the point of Intel even releasing desktop Broadwell at this point? I think we should be stuck with desktop Haswell from 2013Q2 until 2016/2017 or so. After all, there's no competition so what's the hurry. We should cheer them on and applaud them in such a decision.

They have themselves to compete with.

 

[Arachnotronic]

In that case Intel should have though about that earlier, and invested less in R&D so Skylake would be ready at a later time. Now it will be ready several quarters ahead of actual launch, which is pointless. That is bad business.

IMO, having a card to play when/if it is needed is better than not having it.

 

[Idontcare]

In that case Intel should have though about that earlier, and invested less in R&D so Skylake would be ready at a later time. Now it will be ready several quarters ahead of actual launch, which is pointless. That is bad business.

It is only bad business if they don't make the desired/targeted return on their investment over the life of the product.

As for Intel being "too prepared"...chance favors the prepared, and Intel's financials reflect a company which is well run. Not about to second-guess the experts there, they seem to be doing just fine.

 

[Fjodor2001]

There is also business logic in saving money by using child labor, or not taking care of toxic waste thereby polluting the environment. If done in certain countries it's still legal. It doesn't mean I approve of that or like it either.

 

[RussianSensation]

That's the thing -- since Intel has no competition for i5/7 CPU, the managers know that even with 6-12 months delays they will still get those 1st (Nehalem/Lynnfield) and 2nd/3rd gen (Sandy Bridge/Ivy Bridge) users to upgrade to BW-E/Skylake-K. Even if they delayed these products to Q4 2016, they would still get us to upgrade unless AMD pulls some miracle architecture. But with AMD not having access to 14nm node, even with a superior architecture, it's very unlikely for AMD to match or beat Skylake because they will continue to be behind on the manufacture process. Can you imagine if NV made 28nm GPUs but AMD was on 14nm? There would be no competition.

It's still possible Intel will release Slylake-S alongside BW-K as we haven't seen an official confirmation of the delay. Despite how exciting Skylake-K may seem, Intel will most likely cap it at i7 quad core for the 1151 Socket. I think Skylake-K 6-core as a real successor to 5820K is a lot more exciting but that is probably ways off at Q1-Q2 2017.

 

[jpiniero]

If anything, it was probably the OEMs who wanted more time on Broadwell laptop models and Intel gave in.

Yeah, Skylake-K isn't going to be 6 core, but it will have GT4e graphics.

 

[PPB]

What would be your performance expectations out of skylake, compared to haswell?

- We know already that Broadwell is in the 3-5% ballpark in regards IPC increase. OC headroom of 14NM is still yet to be seen.
- We pretty much dont know anything important regarding Skylake (except that it will be obviously wider than their predecesor). Some ̶S̶e̶r̶o̶n̶x̶ wild speculation regarding an improved SMT was floating around. Is there any feasibility behind this?


Im asking this now because Im thinking of going for a 4790K, the stock clock at 4.0 is really appealing if you are going for a H81/H97 platform, and the software that I use tends to be the most favoring scenario for IPC increase (to give you an idea, MT performance of 4770 is the same as a 3930K in Vray, and Haswell gains from Ivy had been more like 20% instead of the 10% the average is getting). I fear that the skylake bump would also be big enough to make the wait worthwhile.

 

[Arachnotronic]

What would be your performance expectations out of skylake, compared to haswell?

- We know already that Broadwell is in the 3-5% ballpark in regards IPC increase. OC headroom of 14NM is still yet to be seen.
- We pretty much dont know anything important regarding Skylake (except that it will be obviously wider than their predecesor). Some ̶S̶e̶r̶o̶n̶x̶ wild speculation regarding an improved SMT was floating around. Is there any feasibility behind this?


Im asking this now because Im thinking of going for a 4790K, the stock clock at 4.0 is really appealing if you are going for a H81/H97 platform, and the software that I use tends to be the most favoring scenario for IPC increase (to give you an idea, MT performance of 4770 is the same as a 3930K in Vray, and Haswell gains from Ivy had been more like 20% instead of the 10% the average is getting). I fear that the skylake bump would also be big enough to make the wait worthwhile.

There's *always* something better around the corner. If you need to buy now, then buy what you need today so you can get the most value out of it. If what you currently have is adequate for what you need to do and you won't be frustrated/lose productivity as you wait for Skylake, then wait for Skylake

 

[Qwertilot]

And ideally don't spend ages staring at road maps inbetween times. Then you'll feel happily surprised by the improvement rather than annoyed that something is taking slightly longer than once planned

 

[Dribble]

The delay isn't really in Intel's favour - the slower they move forward the slower we upgrade. They only make money when we upgrade. Personally I still have a i2500K, I would have built a new machine this year (well new MB, CPU, memory) but what's the point. In that way Intel are loosing out on money I would spend on them but instead I am spending on something else.

 

[III-V]

What would be your performance expectations out of skylake, compared to haswell?

- We know already that Broadwell is in the 3-5% ballpark in regards IPC increase. OC headroom of 14NM is still yet to be seen.
- We pretty much dont know anything important regarding Skylake (except that it will be obviously wider than their predecesor). Some ̶S̶e̶r̶o̶n̶x̶ wild speculation regarding an improved SMT was floating around. Is there any feasibility behind this?

I am expecting overclocks in the range of 5-5.2GHz. I am also expecting IPC improvement around 5-10% for legacy apps, over Broadwell.

Haswell already bumped the width of the back end up... I don't think it really makes sense to do it again. Unless you are talking about AVX.

Cache sizes haven't moved in a while, and I think it's about time for the front end to get some love. Especially when you consider that system memory accesses are very power hungry compared to grabbing code and data from caches, and the importance of power consumption today.

The delay isn't really in Intel's favour - the slower they move forward the slower we upgrade. They only make money when we upgrade. Personally I still have a i2500K, I would have built a new machine this year (well new MB, CPU, memory) but what's the point. In that way Intel are loosing out on money I would spend on them but instead I am spending on something else.

Yep. Especially when you realize that the key to making faster computers is faster computers themselves.

 

[witeken]

I am expecting overclocks in the range of 5-5.2GHz. I am also expecting IPC improvement around 5-10% for legacy apps, over Broadwell.

Haswell already bumped the width of the back end up... I don't think it really makes sense to do it again. Unless you are talking about AVX.

Cache sizes haven't moved in a while, and I think it's about time for the front end to get some love. Especially when you consider that system memory accesses are very power hungry compared to grabbing code and data from caches, and the importance of power consumption today.

Yep. Especially when you realize that the key to making faster computers is faster computers themselves.

You seem to know a lot about these things (AnandTech hasn't even done an article about the Broadwell architecture),.. you could even make a site :ninja:.

 

[Enigmoid]

Cache sizes haven't moved in a while, and I think it's about time for the front end to get some love. Especially when you consider that system memory accesses are very power hungry compared to grabbing code and data from caches, and the importance of power consumption today.

I agree. The skylake wiki page previously had doubled cache with lower latency.