Intel Core "Skylake" CPUs Accompanied by 100-series Chipset

[btarunr]

Intel answered the burning question some of us had about what the desktop chipset that succeeds the 9-series will be named. For now, Intel is referring to it as "100-series," on early internal roadmap documents scored by VR-Zone. Much like the current 9-series, 100-series will consist of a single PCH silicon, from which several variants will be carved out by toggling features.

There will be four primary kinds of "Skylake" packages, SLK-S, which will be socketed LGA; SLK-U, which will likely be compact, ultra-low power BGA, for Ultrabooks; SLK-Y, which will probably be mainstream BGA for compact desktops and all-in-ones; and SLK-H, which will likely be mainstream BGA for conventional notebooks. This generation of CPUs and PCHs, will also be accompanied by four kinds of wireless network controllers, depending on the target form-factor, "Snowfield Peak" Wi-Fi + Bluetooth, "Douglas Peak" WiGig + Bluetooth, "Pine Peak" WiGig, and XMM726x 4G LTE controllers; and "Jacksonville" GbE wired Ethernet controller. Thunderbolt standard will undergo an evolution with the company's "Alpine Ridge" controller.



View at TechPowerUp Main Site

 

[Fiery]

I don't think the SKU differentiation is listed accurately. U would be the one succeeding the current Haswell-ULT (e.g. Core i7-4500U), BGA package, for ultrabooks and compact PCs (like Intel NUC). Y would be the successor to the current Haswell-ULX (e.g. Core i7-4610Y), a lower-TDP variant of the U SKU with similar features, targeting Windows 8.x tablets. H would be the successor to the current 4-core mainstream mobile processors (e.g Core i7-4900MQ).

What's interesting however is that the Skylake lineup does not seem to have any socketed mobile processors anymore.

 

[RejZoR]

Can't wait for the Skylake. I want these monsters for my next major upgrade.

 

[Aquinus]

Can't wait for the Skylake. I want these monsters for my next major upgrade.

Calling it a monster is probably an understatement. Considering latest improvements in performance from Intel, upcoming releases are probably not going to show staggering performance boosts and Intel has already addressed many of consumer's complaints about the PCH (I still think you should use PCI-E if it's not enough though,) but all in all, weather you get a new rig now or in a year, I doubt the difference in performance between the two will be all that significant. Even now, your 920 is probably pretty quick at 4Ghz.

 

[THU31]

They should finally make a SoC, what is the point of keeping the "south bridge" as a separate chip, with such a tiny die size?

 

[Aquinus]

They should finally make a SoC, what is the point of keeping the "south bridge" as a separate chip, with such a tiny die size?

Yeah, but then you're stuck with those components and every board needs to be able to support common functionality. It's harder to build a motherboard to support different south bridges, so an i3 would need to still have a similar SoC components as an i7 on the same platform. At least now, you have the option between different PCHs and you can drop in basically whatever CPU you like. If you go the SoC route, all motherboards will look like AM1, where most of the functionality will be the same. Same number of SATA, USB, and Ethernet ports as every other board unless they start using PCI-E lanes off the CPU to do what the PCH is already doing. Also the PCH adds more PCI-E lanes that aren't supplied by the CPU (X79 offers 8 lanes, some of which are used by components on the motherboard), so you're reducing the number of usable lanes because PCI-E does take up a good amount of die space and IC pins/contacts.

Also, there is little performance or power savings to be had by going to an SoC design. PCHs already use very little power which isn't adding to the CPU's TDP, even X79 only has a TDP of less then 8 watts, z87 uses half that. So I don't see much motivation beyond smaller platforms for having a SoC design. Just my 2 cents.

 

[THU31]

But are those PCHs not identical physically, but with certain features enabled or disabled? I honestly do not see the point of having so many different PCHs, purposely limiting features like the number of USB 3.0 or SATA III ports.
Why would you need so many different motherboards as well? Would it not be cheaper to manufacture just a few models?

 

[sparkyar]

another socket/package? so x97 chipsets are only for haswellrefresh/broadwell... so intel...

 

[fortiori]

Calling it a monster is probably an understatement. Considering latest improvements in performance from Intel, upcoming releases are probably not going to show staggering performance boosts and Intel has already addressed many of consumer's complaints about the PCH (I still think you should use PCI-E if it's not enough though,) but all in all, weather you get a new rig now or in a year, I doubt the difference in performance between the two will be all that significant. Even now, your 920 is probably pretty quick at 4Ghz.

Honest question, what complaints have they addressed? The PCH bandwidth is the biggest complaint i know of and it's been the same for as long as i can remember. That fruit doesn't just hang low, it's laying on the ground and for generations they have done nothing to increase it while at the same time adding more and more bandwidth intensive ports to suck up the limited bandwidth (SATA III, USB 3, M.2).

When you can saturate the entire CPU-PCH connection by using only 3 ports at one time, and you do nothing to address this glaring flaw for years on end...

 

[RejZoR]

Calling it a monster is probably an understatement. Considering latest improvements in performance from Intel, upcoming releases are probably not going to show staggering performance boosts and Intel has already addressed many of consumer's complaints about the PCH (I still think you should use PCI-E if it's not enough though,) but all in all, weather you get a new rig now or in a year, I doubt the difference in performance between the two will be all that significant. Even now, your 920 is probably pretty quick at 4Ghz.

I'm aiming at the eDRAM powered L4 cache Skylakes with HT and 6 cores (12 threads). Hopefully they won't skimp on features and keep on insisting with only quad core architecture.

 

[Steevo]

http://www.computerweekly.com/news/2240184800/Intel-buys-ST-Ericssons-GNSS-business

So your new computer comes with GPS, I understand the desire in portables, not sure if this is also coming in desktop formats too.....

 

[TRWOV]

Are we still at 16 PCIe lanes?

 

[Hilux SSRG]

Come to me my Skylake...

So intel is including native BT in the controllers now?

 

[dhdude]

I'm aiming at the eDRAM powered L4 cache Skylakes with HT and 6 cores (12 threads). Hopefully they won't skimp on features and keep on insisting with only quad core architecture.

This may actually be worth upgrading to. Haswell and Broadwell just aren't worth the upgrade from 2nd & 3rd gen stuff! Of course Haswell-E might change my mind, feeling the upgrade itch again...

 

[Hilux SSRG]

This may actually be worth upgrading to. Haswell and Broadwell just aren't worth the upgrade from 2nd & 3rd gen stuff! Of course Haswell-E might change my mind, feeling the upgrade itch again...

Hold off if you can, DDR4 prices will be lower once Skylake ships!

 

[dhdude]

Hold off if you can, DDR4 prices will be lower once Skylake ships!

Yeah that is a good point! Not looking forward to seeing inital DDR4 prices...

 

[Hilux SSRG]

I'm hoping that Skylake brings Hexacore to the mainsteam i7/i5 markets. But that would be a big wish tho!

 

[MikeMurphy]

Yeah, but then you're stuck with those components and every board needs to be able to support common functionality. It's harder to build a motherboard to support different south bridges, so an i3 would need to still have a similar SoC components as an i7 on the same platform. At least now, you have the option between different PCHs and you can drop in basically whatever CPU you like. If you go the SoC route, all motherboards will look like AM1, where most of the functionality will be the same. Same number of SATA, USB, and Ethernet ports as every other board unless they start using PCI-E lanes off the CPU to do what the PCH is already doing. Also the PCH adds more PCI-E lanes that aren't supplied by the CPU (X79 offers 8 lanes, some of which are used by components on the motherboard), so you're reducing the number of usable lanes because PCI-E does take up a good amount of die space and IC pins/contacts.

Also, there is little performance or power savings to be had by going to an SoC design. PCHs already use very little power which isn't adding to the CPU's TDP, even X79 only has a TDP of less then 8 watts, z87 uses half that. So I don't see much motivation beyond smaller platforms for having a SoC design. Just my 2 cents.

Bingo, and platform cost savings.

 

[Aquinus]

But are those PCHs not identical physically, but with certain features enabled or disabled? I honestly do not see the point of having so many different PCHs, purposely limiting features like the number of USB 3.0 or SATA III ports.
Why would you need so many different motherboards as well? Would it not be cheaper to manufacture just a few models?

No, because when you charge more for the nicer model and less for the not so good one, you make profit. Also you're not changing out the PCH on current boards, it's soldered in place, so you don't need to worry about things like components existing for the capabilities of the PCH. With an SoC, you need to expect all CPUs to have the same interfaces regardless of the motherboard. That isn't the case right now.

Honest question, what complaints have they addressed? The PCH bandwidth is the biggest complaint i know of and it's been the same for as long as i can remember. That fruit doesn't just hang low, it's laying on the ground and for generations they have done nothing to increase it while at the same time adding more and more bandwidth intensive ports to suck up the limited bandwidth (SATA III, USB 3, M.2).

When you can saturate the entire CPU-PCH connection by using only 3 ports at one time, and you do nothing to address this glaring flaw for years on end...

Then get a RAID card. Stop expecting the PCH to do everything under the sun. It doesn't work that way. You need more bandwidth, get a dedicated device for it. Also the number of users that will fully saturate 3 6Gbps SATA ports are few and far between which is why I think it's nuts when people complain about SATA on X79.

I'm aiming at the eDRAM powered L4 cache Skylakes with HT and 6 cores (12 threads). Hopefully they won't skimp on features and keep on insisting with only quad core architecture.

I have reservations about just adding another level of cache. I would like to see performance numbers to see that they got it right before jumping on board that ship.

Are we still at 16 PCIe lanes?

For Broadwell? Well, yeah. If you need more that's an argument for skt2011. It's 20 lanes actually, but 4 lanes aren't usually available. It might have something to do with DMI.

Bingo, and platform cost savings.

Exactly. It works well for the market. If you have a powerful system the likelihood you want wanting something in specific is a lot greater and for people who complain about bandwidth, I will say it again, get a fricken RAID card. You don't use an Intel IGP for gaming, so why would you use the PCH for some crazy I/O numbers? It's the same philosophy, giving you want you absolutely need, not what you absolutely want.

 

[fortiori]

Then get a RAID card. Stop expecting the PCH to do everything under the sun. It doesn't work that way. You need more bandwidth, get a dedicated device for it. Also the number of users that will fully saturate 3 6Gbps SATA ports are few and far between which is why I think it's nuts when people complain about SATA on X79.

I expect the PCH to be able to cope with a higher percentage of potential usage scenarios from its available ports than it can currently. It's simple math really:

Due to PCI overhead you get about 1800MBs max bandwidth divided between:

6 Sata III/1 M.2 ports at rough total of 3600MBs
6 USB 3 ports at roughly 3000MBs
8 USB 2 ports at a rough total of 240MBs
Integrated LAN at 100MBs

So here you have a paltry 1800MBs link trying to manage 6940MBs of potential bandwidth.

This isn't about I/O numbers, or more bandwidth for the sake of more. This is about smooth operation in high utilization scenarios using the ports integrated into the system. No matter how you slice it the bandwidth PCH bandwidth is woefully inadequate for the functionality provided and it should have been widened years ago to accommodate the faster and increasing number of interfaces (USB 3/SATA III) not to mention the introduction of SSDs.

I don't expect the PCH to do everything under the sun, I simply expect it to be able to cope perfectly fine with the bandwidth requirements of all its integrated (read: advertised and marketed) ports.

 

[Deleted member 24505]

WiGig sounds funny, wonder if Glenn Quagmire's PC has WiGiggedy

Also DDR4 prices wil be

 

[Aquinus]

I expect the PCH to be able to cope with a higher percentage of potential usage scenarios from its available ports than it can currently. It's simple math really:

Due to PCI overhead you get about 1800MBs max bandwidth divided between:

6 Sata III/1 M.2 ports at rough total of 3600MBs
6 USB 3 ports at roughly 3000MBs
8 USB 2 ports at a rough total of 240MBs
Integrated LAN at 100MBs

So here you have a paltry 1800MBs link trying to manage 6940MBs of potential bandwidth.

This isn't about I/O numbers, or more bandwidth for the sake of more. This is about smooth operation in high utilization scenarios using the ports integrated into the system. No matter how you slice it the bandwidth PCH bandwidth is woefully inadequate for the functionality provided and it should have been widened years ago to accommodate the faster and increasing number of interfaces (USB 3/SATA III) not to mention the introduction of SSDs.

I don't expect the PCH to do everything under the sun, I simply expect it to be able to cope perfectly fine with the bandwidth requirements of all its integrated (read: advertised and marketed) ports.

That problem with your logic is that according to your logic ALL I/O is used at the same time. This is never ever true with consumers. It's rarely even true with servers either. It's an unrealistic use case.

Tell me about how you have 6 SSDs running full-tilt constantly, and how you have 6 USB 3 ports (probably on two or three hubs, so it's actually half of what you think it is or less,) also running full tilt filled with... what? Hard drives? Flash Drives? USB 2.0 barely scratches the surface and neither does integrated LAN. So with that said, tell me more about how you have SSDs plugged into every SATA and USB 3.0 port running full tilt all the time. Let me tell you something, that's unrealistic.

Additionally, DMI does not use PCI-E. It's similar, yes, but two different signaling technologies nonetheless. DMI has 20Gbit to work with which is actually closer to 2500-2800MB/s in any one direction which is almost 1GB/s off from what you stated. DMI is also 20Gbit BI-DIRECTIONAL, where USB and SATA are not bi-directional and they're describing total bandwidth in any given direction so the max theoretical bandwidth of DMI 2.0 as a whole would be double that 5000-5600MB/s. You also left out the 8 lanes that the PCH tends to offer which takes a bite out of DMI's bandwidth when something is happening. If nothing is happening, you'll always have enough bandwidth regardless of how many devices are plugged in.

With that all said, you would need to do a lot to hit that limitation and if you are hitting I/O that hard and if it becomes a problem, it's still an argument for you to get a RAID controller and stop relying on the PCH. To think that your "integrated option" is the best answer for everything is ludicrous. I don't see gamers running around with Intel IGPs saying that it's awesome for gaming and you don't see sysadmins running around claiming that the PCH/MCH/South Bridge is the best option for I/O. Much like how anyone who actually has experience with a RAID controller knows that the benefit can vastly outweigh the costs when you actually need it.

So it's actually not just "simple math". That would be true if everything was running full steam ahead, but it doesn't work that way. Also I would gladly give up 10% of my performance across the board to run that much hardware at full tilt. I don't think you realize how powerful the PCH really is. This tends to be a common problem when people complain about something they've never encountered then try to claim they know something about it. Stick with what you know, please.

Until you can show me that you've reached this limit without doing anything too extreme (something that less than 1% of consumers would do,) I think you're arguing about something that doesn't actually matter. Can you saturate DMI? Sure... but what does it take to do it? The answer is: More than you'll ever should do with the PCH anyways or probably even with your tower in the first place.

Additionally, considering the nature of each bus. If everything was reading AND writing the same amount at full tilt (not just reading or just writing), DMI 2.0 actually has almost enough bandwidth to drive everything at full speed considering DMI 2.0 is 20Gbps bi-directional (40Gbps total).

Most people don't even have two SSDs, forget 6. Most people don't fill up all of their USB 3.0 ports with high-bandwidth devices either. Lets stick to reality.

 

[fortiori]

So it's actually not just "simple math". That would be true if everything was running full steam ahead, but it doesn't work that way. Also I would gladly give up 10% of my performance across the board to run that much hardware at full tilt. I don't think you realize how powerful the PCH really is. This tends to be a common problem when people complain about something they've never encountered then try to claim they know something about it. Stick with what you know, please.

I was politely disagreeing with you and you appear to be taking it personally for some reason...

Speaking of what i know: I have saturated DMI 2.0 so those mythical and unrealistic usage scenarios you mention aren't so mythical after all. What you get is stuttering and locks ups. Your opinions on whether or not i should be fine with this, or whether or not i should be using my computer that way are no more useful or impactful to me than a dog licking his own balls in the summer sun. Given that we're on an enthusiast technology site, I'm not sure why you would expect otherwise. You could have saved yourself a whole lot of time, trouble, and offense by implicitly understanding that and not replying at all.

 

[Aquinus]

I was politely disagreeing with you and you appear to be taking it personally for some reason...

Speaking of what i know: I have saturated DMI 2.0 so those mythical and unrealistic usage scenarios you mention aren't so mythical after all. What you get is stuttering and locks ups. Your opinions on whether or not i should be fine with this, or whether or not i should be using my computer that way are no more useful or impactful to me than a dog licking his own balls in the summer sun. Given that we're on an enthusiast technology site, I'm not sure why you would expect otherwise. You could have saved yourself a whole lot of time, trouble, and offense by implicitly understanding that and not replying at all.

...and I'm telling you that you're wrong and that your "polite" comment was riddled with incorrect facts.

You saturated it, huh? What did you do and how did you determine that because I bet you that you're computer is performing like crap because of something else. At least I can explain why your wrong, you just profess that you're correct. I intervened and took it seriously because you're talking about stuff without having accurate knowledge of how it works and spreading such false information only makes the problem worse. Also to top it all off, despite evidence to the contrary you remain ignorant of what's right in front of you. Numbers don't lie, something is happening and it's not DMI-related.

This is my way of nicely saying: Don't talk about what you don't know and you'll be fine. A little research can go a long way.

Edit: I would keep the colorful metephors on GN. They only denigrate your argument.

 

[fortiori]

This is my way of nicely saying: Don't talk about what you don't know and you'll be fine. A little research can go a long way.

http://www.intel.com/content/dam/ww...h-gen-core-family-desktop-vol-1-datasheet.pdf

Page 27:

Maximum theoretical bandwidth on interface of 2 GB/s in each direction
simultaneously, for an aggregate of 4 GB/s when DMI x4.

You were saying?