Personal computing discussed

Moderators: renee, Flying Fox, morphine

 
Milo Burke
Gerbil XP
Topic Author
Posts: 490
Joined: Thu Nov 07, 2013 11:49 am

Skylake Core Count?

Thu May 22, 2014 2:42 pm

The infallible Wikipedia states that Skylake will have "quad-core as the default mainstream configuration".

Has anyone heard anything in confirmation or to the contrary?

If a Skylake i3 is quad-core, will the line-up be arranged like this?

i3 = quad core with Hyper Threading
i5 = eight core without Hyper Threading
i7 = eight core with Hyper Threading

Or perhaps something more like this:

i3 = quad core without Hyper Threading
i5 = quad core with Hyper Threading
i7 = six core and eight core with Hyper Threading

What do you guys think?

Some might suggest bringing the core counts too high would challenge the server markets. I know before we were limited to 6 and 8 core processors for servers. But don't they have 12 and 15 core processors for servers now? And isn't it likely that server core counts will continue to rise in the next 24-36 months?
 
The Egg
Minister of Gerbil Affairs
Posts: 2938
Joined: Sun Apr 06, 2008 4:46 pm

Re: Skylake Core Count?

Thu May 22, 2014 3:21 pm

Well citizen that's an excellent question and I thank you for it.

I think it's great that we live in a country where you can ask questions. Because without questions, we'd just have answers. And an answer without a question is a statement.
 
Bauxite
Gerbil Elite
Posts: 788
Joined: Sat Jan 28, 2006 12:10 pm
Location: electrolytic redox smelting plant

Re: Skylake Core Count?

Thu May 22, 2014 3:32 pm

Haswell EP (socket 2011-3) is recently rumored to go to even 18 cores (or have the same socket as haswell EX) core inflation is inevitable it is just a matter of when and how.
They could hit a "core wall" like they did with clockspeed, but even that is still increasing in the long run, also never has there been a performance wall with IPC and other improvements.
Also rumors of high end, many-core unlocked xeons, given amd and nvidia approaching this market with their "premium" gpus perhaps intel wants some of this pie. Some enthusiasts would definitely plop down 3k+ for one, hell so would some HFT and other niche markets.

I could see them bumping the price of the consumer i7s up closer to $500 if they get 6/8 cores, market dominance and all that. They might force iris pro cache on those models as well to "justify" the pricing.

I also wonder if they will continue to stick to 16 pci-e lanes for the consumer socket, with SSDs migrating to that bus it is overdue for more lanes IMO.

Also saw a rumor that the BGA broadwell desktop chips (which will be most of them baring iris pro) will actually regress to pci-e 2.0! "When our market abuse powers combine, we are captain monopoly!"

edit: read it here, take with a grain of salt but it is all plausible
TR RIP 7/7/2019
 
ClickClick5
Gerbil Elite
Posts: 581
Joined: Fri Aug 15, 2008 8:28 pm
Location: Alaska
Contact:

Re: Skylake Core Count?

Thu May 22, 2014 3:58 pm

The Egg wrote:
Well citizen that's an excellent question and I thank you for it.

I think it's great that we live in a country where you can ask questions. Because without questions, we'd just have answers. And an answer without a question is a statement.


Thank you Mayor West.

EDIT: But I vote for a 6 core i7 for mainstream and 8 core i7's for the E series. I don't see Intel going crazy with the core count, say a 16 or 24 core with HT.
i7-5930k (4.1GHz), 32GB DDR4-2400, 2080Ti, GA-X99-UD3, Samsung 860 Pro 256GB (os/programs), Toshiba 5TB (games), WD VelociRaptor 500GB (games/scratch disk), Win 10 Pro x64
Check out my site for trance sets and other goodies! clickclick5.com!
 
puppetworx
Gerbil Elite
Posts: 710
Joined: Tue Dec 02, 2008 5:16 am

Re: Skylake Core Count?

Thu May 22, 2014 4:31 pm

I definitely expect more cores now that per thread performance has roughly been the same for the last three generations. Adding more cores means more silicon (though 14nm process mitigates some of that) so I don't expect it to be a crazy core count. Dual-Core Pentiums, Quad-Core i3s, Hexa-core i5s and Octa-Core and up i7s doesn't sound outlandish. I presume by 'default mainstream' they mean i3.
 
Milo Burke
Gerbil XP
Topic Author
Posts: 490
Joined: Thu Nov 07, 2013 11:49 am

Re: Skylake Core Count?

Thu May 22, 2014 4:58 pm

I suspect Skylake will be the system to build. My hunch is that Broadwell won't bring us that much. Sandy was great. Ivy was just a process shrink, not a surprise there. While all the power savings with Haswell were impressive, the performance increase wasn't. And Broadwell is just a process shrink.

So if Sandy was a good new microarchitecture, and Haswell was a bit lame, hopefully Skylake will be good again.

And a decent IPC increase with a core count increase with PCI 4.0 and DDR4 sounds like a really nice PC. I have an Ivy i5 machine now. Maybe I'll replace it with a Skylake i7. :)
 
f0d
Gerbil XP
Posts: 422
Joined: Sun Jan 23, 2011 3:07 pm
Location: austrALIEN

Re: Skylake Core Count?

Thu May 22, 2014 6:18 pm

since its a guessing game i think it will go like this for skylake

celeron 2 core **maybe** (neutered pentium)
pentium/celeron 4 core (i think there still might be a chance for 4 core celerons)
i3 4 core - hyperthreading
i5 6 core
i7 6 core - hyperthreading
(imo it will be similar to what we have now but just with 2 added cores)

skylake - e
6 core 6820k?!?!?
8 core 6930k
8 core 6960extreme edition
all with hyperthreading
(i dont think we will see an increased core count on skylake-e compared to haswell-e)

yes thats right i think its time for death of the dual core and even the low end pentium/celeron will have quads (maybe the celeron **might** have a neutered 4 core and make it a 2 core)

thats the way i see things happening with skylake
Image
 
MadManOriginal
Gerbil Jedi
Posts: 1533
Joined: Wed Jan 30, 2002 7:00 pm
Location: In my head...

Re: Skylake Core Count?

Thu May 22, 2014 8:27 pm

Hmm, people buying into the MOAR COREZ hype? I'd rather see significant clockspeed increases to use the higher desktop versus mobile power budget than more cores. But of course scaling a single silicon process over a wide range of clockspeeds is hard to do, so moar corez is more likely. I say there might be a moderate increase, maybe with Turboboost going down to i3's, hyperthreading going down to i5's, and then more cores for i7's. Of course this is two generations away so it's hard to say.
 
dodozoid
Gerbil
Posts: 63
Joined: Thu May 09, 2013 4:21 am
Location: Czech Republic (central Europe)

Re: Skylake Core Count?

Fri May 23, 2014 5:10 am

f0d wrote:
i3 4 core - hyperthreading
i5 6 core
i7 6 core - hyperthreading

I highly doubt this assumption... this would make i3 8 thread whie i5 would be only 6 thread processor
 
Arvald
Gerbil Elite
Posts: 761
Joined: Tue Sep 27, 2011 12:14 pm
Location: Gerbil-land, Canada

Re: Skylake Core Count?

Fri May 23, 2014 1:47 pm

If you think right the mainstream I believe is as follows:
i3 - 2 Core
i5 - 4 Core
i7 - 4 Core Hyper threaded.

So this means if the mainstreams all move up then a good guess would be:
i3 - 4 Core
i5 - 4 Core Hyper threaded.
i7 - 6 Core Hyper threaded.
 
NovusBogus
Graphmaster Gerbil
Posts: 1408
Joined: Sun Jan 06, 2013 12:37 am

Re: Skylake Core Count?

Fri May 23, 2014 2:39 pm

IIRC Intel has long referred to i5 as mainstream or mainstream performance, so if this was an internal leak it may not reflect any change from the status quo. It's still unclear whether lots of cores will ever offer meaningful performance benefits and I doubt Intel is going to spend extra money fabbing chips that nobody wants, so I'll believe a quad-core i3 when I see it.
 
vargis14
Gerbil Jedi
Posts: 1900
Joined: Fri Aug 20, 2010 6:03 pm
Location: philly suburbs

Re: Skylake Core Count?

Fri May 23, 2014 2:56 pm

I believe...or better yet I hope Skylake gives us the per core improvement we gained when Sandy was released.

I cannot for the life of me find the article on the Chinese 1ghz Loongson quad or 8 core CPU that at such a slow speed without of order execution and 256kb vectors I think if I am remembering correctly are pushing out way more Gflops then Intels 3.5+ ghz hexa cores.
I remember something like the Logsoon putting out like 125+ Gflops @ under 1.25GHZ

I hope Intel and AMD is watching Chinas progress in CPU's to get that kinf of performance from 45nm fabrication is pretty amazing. I remember the next one will be made on 28nm.

I also hope you all do not think I am pulling for China to make the most powerful CPU's! But Intel needs some competition no matter where it comes from to make then give us the increases in performance we should have gotten since sandy bridge.

EDIT: Also Intel not Soldering on the IHS on IVY and HASWELL was a kick in the nutz to the public. If they had soldered on the IHS on IVY and HAswell the K chips would be hitting 5ghz+ with ease. Perhaps with devils canyon if they Solder on its IHS we will see what can be done when the thermal interface is the way it is supposed to be. I guess we will find out in a couple weeks.
2600k@4848mhz @1.4v CM Nepton40XL 16gb Ram 2x EVGA GTX770 4gb Classified cards in SLI@1280mhz Stock boost on a GAP67-UD4-B3, SBlaster Z powered by TX-850 PSU pushing a 34" LG 21/9 3440-1440 IPS panel. Pieced together 2.1 sound system
 
Wirko
Gerbil Team Leader
Posts: 296
Joined: Fri Jun 15, 2007 4:38 am
Location: Central Europe

Re: Skylake Core Count?

Fri May 23, 2014 5:23 pm

f0d wrote:
since its a guessing game i think it will go like this for skylake

celeron 2 core **maybe** (neutered pentium)
pentium/celeron 4 core (i think there still might be a chance for 4 core celerons)


Quite likely.

With Atom cores inside of course, not Core cores.
 
Kougar
Minister of Gerbil Affairs
Posts: 2306
Joined: Tue Dec 02, 2008 2:12 am
Location: Texas

Re: Skylake Core Count?

Sat May 24, 2014 10:34 am

Well, lets differentiate between Skylake and Skylake-E.

As a mainstream chip I'd bet we won't see anything above quadcore for vanilla Skylake. And that's even after Intel's recent re-awakening to the desktop enthusiast market segment. I am not even confident we will see anything >8 cores on Skylake-E chips for that matter.

Some insight should come when Haswell-E launches, if the eight-core chips have a model as cheap as $~600 then maybe Skylake-E will surprise us. But I don't think 8-core chips are going to debut that cheaply, the first Quad-core launched at $800 after all and Intel then waited a year to drop it down to ~$550.
 
Ninjitsu
Gerbil Team Leader
Posts: 219
Joined: Thu Feb 20, 2014 3:46 am

Re: Skylake Core Count?

Sat May 24, 2014 11:16 am

My speculation is:

Celeron = 2C
Pentium = 2C/4T
Core i3 = 4C/4T
Core i5 = 4C/8T
Core i7 = 8C/8T
Core i7 Extreme = 8C/16T to 10C/20T

Haswell-E will be 8C/16T, so they can't have Skylake i7s outperform them...and i don't think there will be any Broadwell-E.

Also the cut-off for regular consumer/semi-pro workloads according to Amdahl's Law would be 8 cores.
 
the
Gerbil Elite
Posts: 941
Joined: Tue Jun 29, 2010 2:26 am

Re: Skylake Core Count?

Sun May 25, 2014 2:40 am

Current Intel trends would put their main stream parts (i.e. socket 115x, soldiered mobile etc.) as topping out as quad core. What Intel has been doing is using the extra die space to move these to a full system-on-a-chip. Case in point is that there will be a Broadwell chip with the full chipset integrated on-die. It wouldn't surprise me if all mobile Sky Lake chips follow this lead.

What may change with SkyLake for the high end consumer parts is that it may support 4-way SMT. Thus the consumer break down would go something like this:
Celeron = 4 core Silvermont
Pentium = 2c/4t Sky Lake
i3 = 4c/4t Sky Lake
i5 = 4c/8t Sky Lake
i7 = 4c/16t Sky Lake

I also see Intel starting to differentiate the lines via what IO is possible. Currently the Z87 and Z97 chipsets have some pins that can be configured as SATA/PCIe or USB3/PCIe. I predict that the number of configurable ports will increase in the design but will be strictly locked to one configuration on the i3's. Essentially if you want more PCIe lanes, you'll have to opt for an i5 or an i7. DDR4 support is coming and I can see the i3's going single channel only in the mobile market with the i5's and i7's supporting dual channel. Due to the bandwidth increase of DDR4, this won't be as dire of a change.

Core counts for SkyLake-E is going to be interesting. Haswell-E for consumers will be 8 core tops with server parts going to 12 core according to rumor. (Haswell-EX is rumored to be 18 core.) Broadwell-E could jump to 10 core on the consumer side but I only see 8 cores enabled on the die. Boardwell-E die targeted at servers would go up to 16 core. I'd guess at 21 core count for Broadwell-EX. At this point I see Intel doing a major change to how their CPU cores are arranged internally. SkyLake-E is also going to bring a new socket though it is anyone's guess at how many pin it'll have and what features it'll support. I'd guess SkyLake-E for consumers at 10 core, possibly harvested from a 12 core die. I do see SkyLake-E for consumers supporting 4-way SMT at this point.

After this, Intel will move to 10 nm production which I suspect will face heavy delays. It would not surprise me to see a major Sky Lake refresh before 10 nm parts reach consumers. I do think that when 10 nm parts arrive, their will be 6 core chips in the mainstream socket. The SkyLake-E refresh would also be the one to use a fully enabled 10 core die (server parts won't see much of a refresh).
Dual Opteron 6376, 96 GB DDR3, Asus KGPE-D16, GTX 970
Mac Pro Dual Xeon E5645, 48 GB DDR3, GTX 770
Core i7 [email protected] Ghz, 32 GB DDR3, GA-X79-UP5-Wifi
Core i7 [email protected] Ghz, 16 GB DDR3, GTX 970, GA-X68XP-UD4
 
f0d
Gerbil XP
Posts: 422
Joined: Sun Jan 23, 2011 3:07 pm
Location: austrALIEN

Re: Skylake Core Count?

Sun May 25, 2014 3:34 am

There are some great speculative ideas here on what could be next so im going to steal some for my updated "guess"

i still believe that 2 cores will be gone when Skylake hits and as "wirko" and "the" speculates the low end will be Silvermont or Silvermont style atom cores

i also think that "the's" speculation is also right about 4 thread SMT but i think it will be on Skylake-E/Xeon but not normal consumer Skylake

Celeron 4 core Atom
Pentium 4 core Skylake
i3 4 core 8 thread Skylake
i5 6 core 6 thread Skylake
i7 6 core 12 thread Skylake

Skylake-E
6 core 6820k?!?!? 24 thread (a die harvested 8 core)
8 core 6930k 32 thread
8 core 6960 extreme edition 32 thread

i think the consumer cpu's will be 2SMT still but they will use 4SMT for the Xeon/extreme series of CPU's

i still think there will be a 6 core consumer Skylake it will be handy for the E3 Xeons as well as consumer i5's and i7's
Image
 
Wirko
Gerbil Team Leader
Posts: 296
Joined: Fri Jun 15, 2007 4:38 am
Location: Central Europe

Re: Skylake Core Count?

Sun May 25, 2014 8:02 am

3-way SMT is also possible, and Intel might decide that the fourth thread brings about too little benefit. So, another take:

i3 - 2 core 6 thread
i5 - 4 core 8 thread
i7 - 4 core 12 thread

f0d wrote:
i3 4 core 8 thread Skylake
i5 6 core 6 thread Skylake


This is unlikely as it would put i3 and i5 too close together, and i3 too high - actually, that's where Haswell i7 is.
 
f0d
Gerbil XP
Posts: 422
Joined: Sun Jan 23, 2011 3:07 pm
Location: austrALIEN

Re: Skylake Core Count?

Sun May 25, 2014 10:14 am

Wirko wrote:
3-way SMT is also possible, and Intel might decide that the fourth thread brings about too little benefit. So, another take:

i3 - 2 core 6 thread
i5 - 4 core 8 thread
i7 - 4 core 12 thread

f0d wrote:
i3 4 core 8 thread Skylake
i5 6 core 6 thread Skylake


This is unlikely as it would put i3 and i5 too close together, and i3 too high - actually, that's where Haswell i7 is.

you could say the same with current cpus that the i3 and i5 are close together i3 2 core 4 thread and i5 4 core 4 thread is pretty close (exactly the same thread wise) yet intel still sells them

since we are speculating we really dont know what is or isnt likely - for all we know they could keep the same core count as haswell or maybe even double it, everything in between is possible too so i wouldnt say anything is unlikely at this point

3 thread smt is possible but i dont think (im not saying its unlikely - just saying i personally dont think) they would use it on consumer CPU's yet and imo they would introduce features like this in xeon / skylake-e first which actually could use the extra threads
Image
 
the
Gerbil Elite
Posts: 941
Joined: Tue Jun 29, 2010 2:26 am

Re: Skylake Core Count?

Mon May 26, 2014 3:07 am

Well I’ve decided to put my predictions into a table. Take with huge grain of salt as in sheer probability some of it will most certainly be wrong. That’s what happens when trying to predict the future. Most stem from trends Intel has had of late.

Intel will continue to par higher clock speed parts with those of lower cores. For example, I’m predicting a 4 core, 4.0 Ghz Broadwell-E Xeon E5, not a 10 core 4.0 Ghz Broadwell-E. Overall, I’m not predicting much of a increase in clock speeds in the coming years, with most chips between 3.0 and 4.0 Ghz for their base clocks. Occasionally Intel will flirt with a turbo over 4.0 Ghz.

The high end Core i7’s are currently derived from dual/quad socket Xeon parts. I’ll go out on a limb and say that they’ll be rebranded as Core i9’s when socket 1150 Broadwell chips arrive with 128 MB of eDRAM. I see Intel charging a good price premium for that eDRAM part and selling it under the Extreme Edition title. This will enable Intel’s marketing machine to offer both the Core i7 Extreme Edition and the Core i9 Extreme Edition for similar prices ($1000 USD). When Sky Lake-E rolls around, I see the Core i9 branding being used to further increase prices.

Mobile Broadwell parts should arrive in late 2014 but I’m not including them in the chart below. Socketed parts I see as arriving in early 2015 (CES?). Broadwell on the consumer side has been delayed due to 14 nm production and while Sky Lake is also impacted, I still see that design arriving roughly 12 months after Broadwell. I do see a massive battle to get 10 nm production underway so I predict a major Sky Lake refresh before Cannon Lake arrives. Cannon Lake’s major improvement over Sky Lake refresh will be mainly a drop in power consumption. I see Intel implementing some of their near threshold voltage work finally making it out of their labs and into 10 nm parts. The server parts on 10 nm won’t see as much benefit as they’ll aim to keep clock speeds and cores both relatively high. IO is also what is going to increase power consumption in the server space.

I predict Sky Lake will bring some IPC improvements by going 8 wide and introducing 4 way SMT. I mainly see 4 way SMT as a server feature and something to distinguish the Core i7/i9 line up. Consumer versions of Sky Lake will be a full SoC incorporating the full chipset on-die. The only thing external are physical interfaces for IO (Ethernet, wireless), main memory and flash storage. eDRAM will continue to be an option but more widely available in all Sky Lake parts. On the server side I see Sky Lake focusing on scalability by dropping the ring bus topology and going to a multi-tiered crossbar approach. This will reduce latency when core counts continue to increase and be able to reduce power consumption. The other factor for scaling is that Intel will offer hybrid PCIe/QPI 2.0 configurable pins. This will give single socket servers a few more PCIe lanes to use. I see Sky Lake-E moving to a six channel DDR4 designs to further increase memory bandwidth as core/thread count continue to climb. This will also require the usage of registered or load reduced DDR4 memory, even for the consumer Core i9 versions. Sky Lake-EX will be Intel’s first processor to use silicon photonics. I see Intel going with a full six optical QPI links here to enable a 3D torus topology.

I’m confident Intel will have another part after Cannon Lake but what I’m not sure is if there will be 7 nm mass production. Intel is working on in it in their R&D labs but getting into mass production may just be too costly. There is also supposed to be a transition to 450 mm wafers during the 14 nm generation which will further compound the fab issue for 10 nm and 7 nm production. By this time, a replacement for silicon will have to be ready.

The server processors will continue to have a discrete chipset. While the number of sockets scales upward, not all IO scales up at the same rate. For example, if Intel's server chips were to move to an SoC style design, would there be a need for eight USB 3.0 per socket? One divergence will be SATA Express/NVMe support since that uses PCIe lanes for connectivity. I see support for that being integrated into the processors themselves so that it can scale with socket count. This is also the reason for why I've increased PCIe lane counts. I'm also predicting a RAID6 NVMe offload chip that'll accept 16 PCIe lanes into it and fan out to 24 lanes. This fan out chip would use the extra lanes for storing parity that is calculated on the chip. Much like Intel's recent chipset's, SkyLake-E and later will have flexible IO between QPI copper and PCIe. Thus only in a single socket config will there be a full compliment of 72 lanes of available (actually 68 due to 4 being used by DMI for the chipset). One QPI link will take up the allocation of eight PCIe lanes with a maximum of three possible copper QPI links. Sky Lake-E may also be limited to a max of two QPI links where as Sky Lake-EP may support three despite using the same socket.

I'm predicting that Sky Lake EX will be the first chip from Intel to feature silicon photonics. This technology will be used to provide each chip with six optical based QPI links in the package (not sure if they'll implement it fully on-die). This will enable each chip to sit in a 3D torus. In addition to the six otpical QPI links, each socket will have two copper QPI links shared with PCIe lanes. Copper will still be necessary to include as potential Knight's Landing successors may include copper QPI for HPC as well as a few FPGA products from 3rd parties.

Desktop/Server parts:
Code Name Product Cores Threads Clocks Power Release Date Notes
Haswell-E Core i7 4 to 8 8 to 16 3.3 to 3.7 Ghz 130W to 150W 2014Q3 First chip with DDR4. Unlocked multipliers.
Haswell-E Xeon E5 4 to 8 8 to 16 3.3 to 3.7 Ghz 130W to 150W 2014Q4
Haswell-EP Xeon E5 10 to 12 20 to 24 3.3 to 3.7 Ghz 130W to 150W 2014Q4
Haswell-EX Xeon E7 8 to 18 8 to 36 2.0 to 2.8 Ghz 95W to 150W 2015Q2 Uses same socket as Ivy Bridge-EX. Introduces DDR4 memory buffers.
Broadwell Core i3 2 4 2.4 to 3.7 Ghz 35W to 77W 2015Q1
Broadwell Core i5 4 4 2.6 to 3.3 Ghz 35W to 77W 2015Q1
Broadwell Core i7 4 8 2.6 to 3.5 Ghz 45W to 87W? 2015Q1
Crystalwell Core i7 Extreme Edition 4 8 3.6 Ghz? 95W? 2015Q2 128 MB of eDRAM
Broadwell-E Core i9 4 to 8 8 to 16 3.4 to 3.8 Ghz 130W to 150W 2015Q4 Uses same socket as Haswell-E. Unlocked multipliers. Harvested from 10 core dies.
Broadwell-E Xeon E5 4 to 10 8 to 20 3.4 to 4.0 Ghz 95W to 150W 2015Q4 Uses same socket as Haswell-E.
Broadwell-EP Xeon E5 10 to 15 20 to 30 3.4 to 3.7 Ghz 130W to 150W 2015Q4 Uses same socket as Haswell-E. eDRAM options?
Broadwell-EX Xeon E7 8 to 18 16 to 36 2.2 to 2.8 Ghz 95W to 150W 2016Q1 Same socket as Ivy Bridge-EX. eDRAM options?
Sky Lake Core i3 2 4 2.4 to 3.4 Ghz 25W to 65W 2016Q1 New socket. SoC. Single channel DDR4 and 16 PCIe 4.0 lanes
Sky Lake Core i5 4 8 2.4 to 3.2 Ghz 35W to 77W 2016Q1 New socket. SoC. Dual channel DDR4 and 20 PCIe 4.0 lanes max
Sky Lake Core i7 4 16 3.3 Ghz 35W to 87W 2016Q1 New socket. SoC. Dual channel DDR4 and 24 PCIe 4.0 lanes max
Crystal Lake? Core i7 Extreme Edition 4 16 3.4 Ghz 95W 2016Q2 New socket. SoC. Dual channel DDR4 and 24 PCIe 4.0 lanes, 256 MB of eDRAM, unlocked
Sky Lake-E Core i9 6 to 8 32 3.2 to 3.6 Ghz 130W to 200W 2017Q1 New socket. Six channel DDR4, registered memory required, up to 72 PCIe 4.0, NVMe, harvested from 10 core die
Sky Lake-E Xeon E5 6 to 10 24 to 40 3.5 to 4.0 Ghz 87W to 150W 2017Q1 New socket. Six channel DDR4, registered memory required, up to 72 PCIe 4.0, NVMe
Sky Lake-EP Xeon E5 12 to 16 48 to 64 3.5 to 4.0 Ghz 130W to 200W 2017Q1 New socket. Six channel DDR4, registered memory required, up to 72 PCIe 4.0, NVMe, eDRAM
Sky Lake-EX Xeon E7 12 to 20 48 to 80 2.2 to 3.0 Ghz 95W to 200W 2017Q3 New socket. Sixteen memory buffer channels, up to 56 PCIe 4.0, NVMe, eDRAM, six QPI 2.0 optical links
Sky Lake refresh Core i3 4 4 2.4 to 3.5 Ghz 25W to 65W 2016Q4 New socket. SoC. Single channel DDR4 and 16 PCIe lanes
Sky Lake refresh Core i5 4 8 2.4 to 3.5 Ghz 35W to 77W 2016Q4 New socket. SoC. Dual channel DDR4 and 20 PCIe 4.0 lanes max
Sky Lake refresh Core i7 4 16 3.4 Ghz 35W to 87W 2016Q4 New socket. SoC. Dual channel DDR4 and 24 PCIe 4.0 lanes max, 128 MB of eDRAM
Crystal Lake refresh Core i7 Extreme Edition 4 16 3.5 Ghz 95W 2016Q4 New socket. SoC. Dual channel DDR4 and 24 PCIe 4.0 lanes, 256 MB of eDRAM, unlocked
Sky Lake-E refresh Core i9 6 to 10 24 to 40 3.3 to 3.6 Ghz 130W to 200W 2017Q1 New socket. Six channel DDR4, registered memory required, up to 72 PCIe 4.0, NVMe
Cannon Lake Core i3 4 4 2.0 to 3.4 Ghz 20W to 65W 2017Q3 New socket. SoC. Dual channel DDR4 and 16 PCIe 4.0 lanes
Cannon Lake Core i5 4 8 2.4 to 3.4 Ghz 35W to 65W 2017Q3 New socket. SoC. Dual channel DDR4 and 20 PCIe 4.0 lanes max
Cannon Lake Core i7 6 12 3.4 Ghz 35W to 87W 2017Q3 New socket. SoC. Dual channel DDR4 and 24 PCIe 4.0 lanes max, 128 MB of eDRAM
Cannon Lake Core i7 Extreme Edition 6 24 3.6 Ghz 87W 2017Q3 New socket. SoC. Dual channel DDR4 and 24 PCIe 4.0 lanes, 256 MB of eDRAM, unlocked
Cannon Lake-E Core i9 6 to 10 24 to 40 3.3 to 3.6 Ghz 130W to 200W 2018Q3 Sky Lake-EP socket. Six channel DDR4, registered memory required, up to 72 PCIe 4.0, NVMe
Cannon Lake-E Xeon E5 8 to 12 24 to 40 3.5 to 4.0 Ghz 87W to 150W 2018Q4 Sky Lake-EP socket. Six channel DDR4, registered memory required, up to 72 PCIe 4.0, NVMe
Cannon Lake-EP Xeon E5 12 to 20 48 to 80 3.2 to 4.0 Ghz 95W to 200W 2018Q4 Sky Lake-EP socket. Six channel DDR4, registered memory required, up to 72 PCIe 4.0, NVMe, eDRAM
Cannon Lake-EX Xeon E7 12 to 24 48 to 96 2.2 to 3.0 Ghz 87W to 200W 2019Q1 Sky Lake-EX socket. Sixteen memory buffer channels, up to 56 PCIe 4.0, NVMe, eDRAM, six QPI 2.0 optical links


Edit: Added note on Broadwell-e for consumers. Explained NVMe in the server chipsets.
Dual Opteron 6376, 96 GB DDR3, Asus KGPE-D16, GTX 970
Mac Pro Dual Xeon E5645, 48 GB DDR3, GTX 770
Core i7 [email protected] Ghz, 32 GB DDR3, GA-X79-UP5-Wifi
Core i7 [email protected] Ghz, 16 GB DDR3, GTX 970, GA-X68XP-UD4
 
elemein
Gerbil In Training
Posts: 5
Joined: Tue Aug 05, 2014 3:41 pm

Re: Skylake Core Count?

Tue Aug 05, 2014 3:48 pm

Just gonna drop this entirely relevant link here despite the thread being old.

translate.google.com/translate?js=n&prev=_t&ie=UTF-8&tl=en&u=http%3A%2F%2Fchinese.vr-zone.com%2F89484%2Fgreenlow-platform-will-include-skylake-sunrise-point-pch-and-jasksonville-phy-and-ddr4-launch-at-2015-11072013%2F

As the number of core areas, and no change, still remain at 2C4T or 4C8T
 
UnfriendlyFire
Gerbil Team Leader
Posts: 285
Joined: Sat Aug 03, 2013 7:28 am

Re: Skylake Core Count?

Tue Aug 05, 2014 4:11 pm

I would assume that as long as AMD doesn't threaten Intel's desktop/laptop market, Intel has little incentive to up the core count, because that means using more silicon die when they could've created more CPUs with the same wafer.

I definitely expect Intel to continue using dual-cores for ULVs for a while.
 
elemein
Gerbil In Training
Posts: 5
Joined: Tue Aug 05, 2014 3:41 pm

Re: Skylake Core Count?

Tue Aug 05, 2014 4:58 pm

UnfriendlyFire wrote:
I would assume that as long as AMD doesn't threaten Intel's desktop/laptop market, Intel has little incentive to up the core count, because that means using more silicon die when they could've created more CPUs with the same wafer.

I definitely expect Intel to continue using dual-cores for ULVs for a while.


There's no real need to either.

For the consumer market, SMT has ruled as the main design philosophy in x64, ARM and MIPS for quite awhile. Dual and quad cores are great for consumer programs.
 
Forge
Lord High Gerbil
Posts: 8253
Joined: Wed Dec 26, 2001 7:00 pm
Location: Gone

Re: Skylake Core Count?

Tue Aug 05, 2014 5:44 pm

I think it's hilarious, how many people in this thread think Intel will give them more cores just because it's what the consumer thinks they want. Here at work, less than one person in ten can even tell you how many cores are in their machine.

Intel will keep 2C4T at low end desktop, 4C4T at midrange desktop, and 4C8T at high end desktop until AMD makes them change. Enthusiast might get as many as 8C16T in the Haswell-E, and laptops will continue to run 2C2T, 2C4T, and 4C4T for the Skylake generation. I'll bet you cash that this disappointing prediction comes true.
Please don't edit my signature for me. Thanks.
 
UnfriendlyFire
Gerbil Team Leader
Posts: 285
Joined: Sat Aug 03, 2013 7:28 am

Re: Skylake Core Count?

Tue Aug 05, 2014 5:52 pm

I think all Intel is going to do consistently is tack on more GPU cores.

Take a look at the trend from Sandybridge's IGP all the way to Haswell, and Intel plans on doubling the amount of GPU cores for Broadwell.

Assuming AMD also keeps pace with their GCN IGP, it's eventually going to result in Intel's IGP being good enough for low-end gaming, AMD being forced to create more efficient IGPs or IGPs that can handle mid-range gaming (quite likely if AMD adds stacked-DRAM to their APUs), and Nividia getting kicked out of the low-end GPU market and thus stop selling their Fermi-rebrands (GT 820M being very similar to GT 435M makes me cringe).
 
elemein
Gerbil In Training
Posts: 5
Joined: Tue Aug 05, 2014 3:41 pm

Re: Skylake Core Count?

Tue Aug 05, 2014 8:16 pm

UnfriendlyFire wrote:
I think all Intel is going to do consistently is tack on more GPU cores.

Take a look at the trend from Sandybridge's IGP all the way to Haswell, and Intel plans on doubling the amount of GPU cores for Broadwell.

Assuming AMD also keeps pace with their GCN IGP, it's eventually going to result in Intel's IGP being good enough for low-end gaming, AMD being forced to create more efficient IGPs or IGPs that can handle mid-range gaming (quite likely if AMD adds stacked-DRAM to their APUs), and Nividia getting kicked out of the low-end GPU market and thus stop selling their Fermi-rebrands (GT 820M being very similar to GT 435M makes me cringe).


I don't entirely agree. Intel has been doing some great work on improving the internal architecture. Haswell was quite an improvement (and an experiment... Sortof. I mean, a lot of work was put into making TSX and iVRMs work, and the results were interesting to say the least.), and Skylake is rumoured to have a few notable improvements as well (namely AVX512...)

Though I agree. Both companies are starting to go at it with the iGPs. While there isn't much any of a market for low end GPUs (at least, not in the desktop area. Now, the laptop area is a different story.) and the next few years will be interesting to see where iGPs start gaining on dGPUs.

In other related news:

tomshardware.com/news/amd-hsa-carrizo-stacked-dram,27246.html

Stacked RAM is already in the cards :) I'm solidly hoping for the IGP of the top Carrizo APU to contest atleast with the GTX 750. IF that happens... Well, that would be big.
 
Forge
Lord High Gerbil
Posts: 8253
Joined: Wed Dec 26, 2001 7:00 pm
Location: Gone

Re: Skylake Core Count?

Tue Aug 05, 2014 9:22 pm

elemein wrote:
Haswell was quite an improvement (and an experiment... Sortof. I mean, a lot of work was put into making TSX and iVRMs work, and the results were interesting to say the least.)


Off-topic and I apologize, but did you see where Intel has officially strangled integrated VRM? Apparently they were so "impressed" by the iVRM results on Haswell that they have cut the iVRMs out of all future designs. Rarely get to see Intel fully reverse on anything.
Please don't edit my signature for me. Thanks.
 
elemein
Gerbil In Training
Posts: 5
Joined: Tue Aug 05, 2014 3:41 pm

Re: Skylake Core Count?

Tue Aug 05, 2014 9:44 pm

Forge wrote:
elemein wrote:
Haswell was quite an improvement (and an experiment... Sortof. I mean, a lot of work was put into making TSX and iVRMs work, and the results were interesting to say the least.)


Off-topic and I apologize, but did you see where Intel has officially strangled integrated VRM? Apparently they were so "impressed" by the iVRM results on Haswell that they have cut the iVRMs out of all future designs. Rarely get to see Intel fully reverse on anything.


Exactly why the results were "interesting"... Intel doesnt usually say "This was actually a bad idea."

They were damn fast to fix it though.
 
UnfriendlyFire
Gerbil Team Leader
Posts: 285
Joined: Sat Aug 03, 2013 7:28 am

Re: Skylake Core Count?

Tue Aug 05, 2014 9:46 pm

Forge wrote:
elemein wrote:
Haswell was quite an improvement (and an experiment... Sortof. I mean, a lot of work was put into making TSX and iVRMs work, and the results were interesting to say the least.)


Off-topic and I apologize, but did you see where Intel has officially strangled integrated VRM? Apparently they were so "impressed" by the iVRM results on Haswell that they have cut the iVRMs out of all future designs. Rarely get to see Intel fully reverse on anything.


They last time they reversed something was Netburst.

In fact, the original Netburst successor was suppose to scale up to 7 GHz until it got canned and replaced with Core 2: http://en.wikipedia.org/wiki/Tejas_and_Jayhawk

And apparently Tejas was also suppose to be IA64 (Itanium) compatible: http://www.xbitlabs.com/news/cpu/displa ... 52741.html
 
Waco
Maximum Gerbil
Posts: 4850
Joined: Tue Jan 20, 2009 4:14 pm
Location: Los Alamos, NM

Re: Skylake Core Count?

Tue Aug 05, 2014 9:55 pm

Milo Burke wrote:
Some might suggest bringing the core counts too high would challenge the server markets. I know before we were limited to 6 and 8 core processors for servers.

Until Intel allows ECC use with desktop processors (very unlikely) there's zero competition for the server space by their consumer parts.
Victory requires no explanation. Defeat allows none.

Who is online

Users browsing this forum: No registered users and 1 guest
GZIP: On