Intel Skylake / Kaby Lake

[IntelUser2000]

Is 4790K or 6700K 65 watts or do they have GT3e? That is why the previous poster is saying it does not have low clocks, relative to the TDP and big igpu.

At these TDP levels, the iGPU is a relatively small part of power use. It's mostly the CPU. Also the advanced power management of these chips means you can put GT5e with 256 EUs and have 15W TDP(not that it will perform good). Anyway that's kind of off point.

We had 3.4GHz back in 2011. Would you say 3.3GHz in 2015 is high, regardless of TDP? Nope. "Low" is always relative. Of course its not low, but compared to the regular variants they are.

 

[CHADBOGA]

Why do you have 3 user accounts? Or are Dave2150, Dave5120, and Dave3000 all different people? Is this in case one gets banned for being too obnoxious on the CPU forums you already have a backup plan? :biggrin:

Nothing wrong with being well prepared. :biggrin:

 

[frozentundra123456]

At these TDP levels, the iGPU is a relatively small part of power use. It's mostly the CPU. Also the advanced power management of these chips means you can put GT5e with 256 EUs and have 15W TDP(not that it will perform good). Anyway that's kind of off point.

We had 3.4GHz back in 2011. Would you say 3.3GHz in 2015 is high, regardless of TDP? Nope. "Low" is always relative. Of course its not low, but compared to the regular variants they are.

It just seems like you are never satisfied. What do you want Titan X graphics and 32 cores at 5 ghz in a 10 watt TDP?

 

[Sweepr]

Sure it is. Compared to 4790K/6700K.

88-95W parts with slower GT2 graphics. Who would have guessed they would clock higher than quad-cores sharing 65W with a massive GT3e iGPU. So according to your logic TDP doesn't matter and anything below Devil's Canyon clocks is low clocked, got it.

I think Cannonlake will have a staggered launch just like Broadwell and Skylake. Mainstream Desktops get Cannonlake at 65W, 95W for 6700K replacement with GT2, and Skylake GT4e @ 65W with low clocks. Of course, Cannonlake may be a firmly 2017 part and they won't have to do that.

If Cannonlake will be launched in 2017 the most likely outcome is a 2016 Skylake Refresh and a new tock for desktops @ H2-2017.

Sure we do. It's quite obvious Broadwell-K/C(w/e its called) is THE graphics-oriented chip for this year. That means GT2 in Skylake is quite a bit lower performing than that. GT3e is about 60% faster than GT2: http://www.anandtech.com/show/6993/intel-iris-pro-5200-graphics-review-core-i74950hq-tested/6

No we don't. There are very slim details about Gen 9 and you are extrapolating Broadwell-K GT3e performance using Haswell data. Not to mention the fact that even if Gen 9 is a big improvement parts without eDRAM will most likely be BW limited and perform worse than Broadwell-K GT3e anyway.

 

[frozentundra123456]

Yea, that was what amazed me about the poster complaining of cpu clocks on BDW-K. CPU performance is the best part of that chip still, at least extrapolating from the minimal gains on graphics we saw in the low power chips. I really thought we would see more gains from Gen 8. Maybe Gen 9, but I am not holding my breath. If one wants to rip intel, it is pretty much justified on the igpu, considering how much die area they are devoting to it and the still mediocre (relative) performance. They better step up their game before AMD's apus get HBM or they could be in for some serious embarrassment. Right now it doesnt matter, because AMD is so bandwidth limited their APUs are still not compelling solutions, but HBM could drastically improve that.

 

[Idontcare]

Personally I think the clocks on 14nm are great compared to the FUD the 14 nm naysayers were spreading earlier.

Are you referring to the clock-targets set by the upper-ends of the SkyLake SKUs, or the yields at which Intel is getting when binning to those targets?

My impression of these forums was that most people expected Intel to target "me too" clockspeeds for Skylake regardless of the health of the 14nm line (binning yields), but that the reason we haven't seen desktop clockspeeds from Intel on 14nm to date are squarely to do with those substellar binning yields.

Not attacking your post or position, just trying to figure out if the Skylake clock targets really change anything in the conversation that has been held since last fall when Broadwell was delayed due to yields.

 

[frozentundra123456]

What you said could well be. I dont really concern myself with that part of it, I only am interested in the final product. It is just that a few posters continually were predicting 14nm would never clock as high as 22nm, etc., etc., from the usual suspects who love to bash intel. I would definitely not deny 14nm was a troubled node, but it looks like we will finally get good products from it. OTOH, desktop chips were never officially announced for Broadwell anyway, so I dont really know what it means that they basically skipped to Skylake.

Where I am more disappointed is that 14nm atom was (still is pretty much) delayed. I finally got fed up waiting for cherry trail and just bought one of the cheap Bay Trail winbook tablets from Microcenter. I also would have liked to see more improvements is U and Y products as far as battery life and thermals (and being on time), but products using those chips are too expensive for my taste anyway. But this is where 14nm really hurt intel. They are so dominant in the laptop/desktop/server market they could afford delays. But in low power mobile, phones and tablets, they really needed a slam dunk to better compete with ARM, and to get the mobile products with a lower BOM to market. What they got unfortunately was a product that was late to the market, and only an incremental improvement when they needed a giant leap.

 

[Idontcare]

What you said could well be. I dont really concern myself with that part of it, I only am interested in the final product. It is just that a few posters continually were predicting 14nm would never clock as high as 22nm, etc., etc., from the usual suspects who love to bash intel. I would definitely not deny 14nm was a troubled node, but it looks like we will finally get good products from it. OTOH, desktop chips were never officially announced for Broadwell anyway, so I dont really know what it means that they basically skipped to Skylake.

Oh yeah, never mind me then, you make sense.

Where I am more disappointed is that 14nm atom was (still is pretty much) delayed. I finally got fed up waiting for cherry trail and just bought one of the cheap Bay Trail winbook tablets from Microcenter. I also would have liked to see more improvements is U and Y products as far as battery life and thermals (and being on time), but products using those chips are too expensive for my taste anyway. But this is where 14nm really hurt intel. They are so dominant in the laptop/desktop/server market they could afford delays. But in low power mobile, phones and tablets, they really needed a slam dunk to better compete with ARM, and to get the mobile products with a lower BOM to market. What they got unfortunately was a product that was late to the market, and only an incremental improvement when they needed a giant leap.

Intel went aggressive with the targets set for 14nm, and with that came a commensurately higher risk-profile. I agree with you that they needed 14nm to really enable themselves establishing at a few key customers and that clearly is not happening.

Will be interesting to see if 10nm is a repeat in risk-profile, or if they push it out (delay) to lower the risk profile in HVM.

 

[tenks]

But what about Broadwell X99 server???

That is all tied to HEDT.

err...what about it? Its not out. HEDT skus are derivatives of Xeons and they typically come out around the same time. Both were delayed and neither are out. I dont understand your point?

 

[Ajay]

Intel went aggressive with the targets set for 14nm, and with that came a commensurately higher risk-profile. I agree with you that they needed 14nm to really enable themselves establishing at a few key customers and that clearly is not happening.

Will be interesting to see if 10nm is a repeat in risk-profile, or if they push it out (delay) to lower the risk profile in HVM.

Seems like they already made they choice for higher risk @ 10nm - given the 9 month delay. There probably weren't any great options, given the delays with EUV. Or, are you talking about 10nm product rollout?

 

[Arachnotronic]

Oh yeah, never mind me then, you make sense.



Intel went aggressive with the targets set for 14nm, and with that came a commensurately higher risk-profile. I agree with you that they needed 14nm to really enable themselves establishing at a few key customers and that clearly is not happening.

Will be interesting to see if 10nm is a repeat in risk-profile, or if they push it out (delay) to lower the risk profile in HVM.

If Intel had actually hit its target with Broxton, it would have been able to take serious advantage of the Qualcomm 20nm chip situation. 14nm leadership SoC + competent XMM 7260 modem? Could have been a winner.

 

[VirtualLarry]

If Intel had actually hit its target with Broxton, it would have been able to take serious advantage of the Qualcomm 20nm chip situation. 14nm leadership SoC + competent XMM 7260 modem? Could have been a winner.

Seems like Intel is missing market opportunities left and right. At least the new Broadwell Core Xeon-D SoCs look promising for micro-servers. (I'd love to see a mini-HEDT platform based on them as well, as cbn has suggested.)

 

[Sweepr]

Seems like Intel is missing market opportunities left and right. At least the new Broadwell Core Xeon-D SoCs look promising for micro-servers. (I'd love to see a mini-HEDT platform based on them as well, as cbn has suggested.)

Me too. I'm sure some people would be interested in an iGPU-less LGA1151 ~2.5GHz 8C/16T Skylake for the right price. Does anyone know Xeon-D's die size?

 

[Enigmoid]

Me too. I'm sure some people would be interested in an iGPU-less LGA1151 ~2.5GHz 8C/16T Skylake for the right price. Does anyone know Xeon-D's die size?

~160 mm^2 or around 3770k/4770k size.

 

[Sweepr]

~160 mm^2 or around 3770k/4770k size.

A slightly higher clocked version of this relatively small 45W chip would probably be enough to crash AMD's party next year if their 8-core Zen isn't another failure.
Per core performance still reigns supreme in desktops though, and I doubt >4GHz Skylake will be threatened in this respect.

 

[Dave2150]

I wonder if anyone knows the maximum number of cores LGA1151 can actually support?

I assume that the more cores on the CPU, the more pins required.

Or does LGA2011 need almost double the pins only to support quad channel memory and more PCI-E lanes?

 

[ShintaiDK]

I wonder if anyone knows the maximum number of cores LGA1151 can actually support?

I assume that the more cores on the CPU, the more pins required.

Or does LGA2011 need almost double the pins only to support quad channel memory and more PCI-E lanes?

There is no limit besides the electrical(total wattage) and physical size limit.

LGA1151 as such could support a 1000 cores.

LGA2011 is due to more memory channels, more PCIe lanes, higher powerdraw and the QPI links.(Not used on desktop version.)

 

[BigDaveX]

From a purely physical standpoint, if Skylake's die has similar dimensions to Haswell then they could quite easily make an MCM 8-core chip by just slapping a second die in there, assuming there was an internal QPI link or somesuch to hook it in.

 

[Justinbaileyman]

So is Intel's Skylake competing against AMD's Zen??Or will they be competing against another Intel CPU??

 

[ShintaiDK]

So is Intel's Skylake competing against AMD's Zen??Or will they be competing against another Intel CPU??

Haswell, Broadwell, Ivy Bridge, Sandy Bridge. I think the Nehalem crowd is giving up with Skylake. :biggrin:

 

[Sweepr]

So is Intel's Skylake competing against AMD's Zen??Or will they be competing against another Intel CPU??

Assuming H2-2016 launch it will most likely compete against a higher-clocked Skylake Refresh (LGA1151), Broadwell-E 6C/12T and 8C/16T (LGA2011, Skylake-E could arrive in late 2016 - perhaps a 10C/20T version?), Skylake-Y/Skylake-U mobile parts (possibly Cannonlake-Y/Cannonlake-U will be close to launch depending on 10nm rollout @ Q4-2016).

Also later in mid 2017 it will probably face Intel's 10nm Tock.

Looks like they recognized that the Intel way with fewer faster cores and SMT is the way to go for desktops. Something tells me performance per core will become important again to some of the ''moar cores'' apologists.

 

[StinkyPinky]

Let's all hope Zen is a good alternative to Intel, because it's in no ones interest for AMD to continue tanking. Could be their last chance in the desktop space.

 

[myocardia]

Looks like they recognized that the Intel way with fewer faster cores and SMT is the way to go for desktops. Something tells me performance per core will become important again to some of the ''moar cores'' apologists.

I hope very much that you are correct. I've been waiting 9 years now for AMD to have another competitive CPU. When I say competitive, I mean more or less commensurate in IPC with Intel's latest/greatest, even if from a generation or two ago.

 

[frozentundra123456]

Assuming H2-2016 launch it will most likely compete against a higher-clocked Skylake Refresh (LGA1151), Broadwell-E 6C/12T and 8C/16T (LGA2011, Skylake-E could arrive in late 2016 - perhaps a 10C/20T version?), Skylake-Y/Skylake-U mobile parts (possibly Cannonlake-Y/Cannonlake-U will be close to launch depending on 10nm rollout @ Q4-2016).

Also later in mid 2017 it will probably face Intel's 10nm Tock.

Looks like they recognized that the Intel way with fewer faster cores and SMT is the way to go for desktops. Something tells me performance per core will become important again to some of the ''moar cores'' apologists.

Yea, lets see how smoothly the launch goes, and when the products really hit the market in good supply. So far no mention of delays, but it seems like a lot of changes at once, especially for a company with the resources of AMD. Even intel doesn't do a totally new architecture plus a new process node at the same time, not to mention designing an ARM core as well, and possibly implementing HBM, although these slides have not mentioned that. IMO, they really need that (HBM) to have a compelling product though.

Edit: sorry getting off topic. My mind is still spinning from the ever increasing expectations for Zen in the other thread.

 

[Justinbaileyman]

Assuming H2-2016 launch it will most likely compete against a higher-clocked Skylake Refresh (LGA1151), Broadwell-E 6C/12T and 8C/16T (LGA2011, Skylake-E could arrive in late 2016 - perhaps a 10C/20T version?), Skylake-Y/Skylake-U mobile parts (possibly Cannonlake-Y/Cannonlake-U will be close to launch depending on 10nm rollout @ Q4-2016).

Also later in mid 2017 it will probably face Intel's 10nm Tock.

Looks like they recognized that the Intel way with fewer faster cores and SMT is the way to go for desktops. Something tells me performance per core will become important again to some of the ''moar cores'' apologists.

Errr,So many socket upgrades between now and AMD's ZEN. With only a year or less away for ZEN I'd really hate to have to upgrade again after spending all this money on Intel's Skylake-K just to compete with AMD's ZEN.I would like to purchase something that will be futureproof enough to last for at leased 2-3 years and something that will be somewhat comparable to ZEN.
I just got that feeling ZEN is going to end up kicking some ass and taking names.
I dont mind Upgrading from a Skaylake-k 6700K to another CPU on the same socket but I really dont want to do all this socket hoping if you can understand everything I just said and make sense of my ramblings.