Some slides have leaked out with information about Intel's forthcoming 6th Generation Core processor, code named Skylake. We have known that Skylake was coming, and coming this year, but there have been a lot of questions about enthusiast parts and what that means for DIY builders. The slides were first seen over at WCCFTech.com and show some interesting new information.
Dubbed Skylake-S, the LGA (socketed) processor will use a new derivative with 1151 pins as well as a new set of chipsets, the Intel 100-series. Skylake is built on the same 14nm process technology used with Broadwell but will feature a new microarchitecture for both the IA cores and the graphics systems. Obviously you can read the slide yourself above, but some of the highlights are worth touching on individually. Skylake will support both DDR3L and DDR4 memory systems with the enthusiast grade parts likely the only ones to attempt to push the newer, faster DDR4 speeds.
Enthusiasts will also be glad to know that there are planned 95 watt quad-core SKUs that will support unlocked features and overclocking capability. Intel lists an "enhanced" BCLK overclocking with the term "full range" which likely means there will no longer be a need for straps to 125 MHz, etc. A 95 watt TDP is higher than the 88 watt limit we saw on Haswell processors so there is a chance we might actually witness usable performance gains if Intel can get the clock speeds up and above where they sit today with current generation parts.
The use of DMI 3.0, the connection between the processor and the chipset, sees the first increase in bandwidth in many generations. Rated at 8 GT/s, twice that of the DMI 2.0 interface used on Haswell, should allow for fewer bottlenecks on storage and external PCIe connections coming from the chipset.
The new Intel 100-series chipsets will come in three variants at launch: the Z170, the H170 and the H110. The one we are most concerned with is the Z170 of course as it will be paired wit the higher end 65 watt and 95 watt enthusiast processors. Based on these specs, Skylake will continue to operate with only 16 lanes of PCI Express 3.0 capable of running at 1 x16, 2 x8 or 1 x8 and 2 x4 connections. With either DDR3L or DDR4 you will have a dual-channel memory system.
For storage, the Z170 still has six SATA 6.0 Gb/s ports, moves to 14 USB ports maximum with 10 of them capable of USB 3.0 speeds and it upgrades Intel RST to support PCIe storage drivers. Of note here is that the Intel chipset does not include USB 3.1 capability so motherboard vendors will continue to need an external controller to integrate it. Without a doubt the 100-series chipsets will be able to support booting and compatibility with the new Intel 750-series PCIe SSDs, the current king of the hill.
As for timing, the roadmap lists the Z170 chipset and the Skylake-S processor as a Q3 2015 release. I would normally expect that to line up with Computex in early June but that doesn't appear to be the case based on other information I am getting.
I was hoping for a few more
I was hoping for a few more pci-e lanes, but it’s great that they’re finally increasing the DMI bandwidth. Thanks for the info.
Since scaling in the DX12
Since scaling in the DX12 draw call test went all the way up to 6 core, it seems like the dual core parts should not be of interest unless you are in an extremely power constrained situation. Although, real games will probably still be graphics cards limited for most settings. At 14 nm, a dual core chip without graphics would only be like 30 to 40 square mm. I was trying to select components for a friend’s system and it still cost close to $200 for quad core Intel parts. I would like to see quad core for a lot cheaper. If I was looking for something closer to $100, it seems like the AMD FX-8320E is a good deal (8 thread, 4 module rather than 4 thread, 2 core Intel parts at this price).
Not even any hints at whether
Not even any hints at whether MorphCore is implemented or not.
MorphCore is Intel’s attempt
MorphCore is Intel’s attempt to get its x86 microarchitecture more competitive with ARM RISC designs in power usage, it’s similar to ARM big/little usage for power saving, not a big interest for desktop gamers were performance in more important. Getting more SMT threads will be more important, if AMD’s Zen does provide at least Sandybridge levels of single threaded performance. Once AMD achieves comparable single threaded performance to Intel, even if it’s just at the SandyBridge level, Intel will be at a big disadvantage on the GPU side of the equation.
AMDs further HSA style integrating of its integrated GPUs more closely to its CPU cores, will give AMD an advantage for floating point acceleration of gaming physics/etc. as well as graphics. There even may be a renewed core war, for more cores, or even more processor threads per core, among AMD and Intel, and with the DX12, and Vulkan APIs able to take advantage of more cores/Processor threads, if Zen’s single threaded performance is at the Sandybridge levels, the pressure may be on to compete with more cores, or more processor threads per core. Everybody is running out of available/affordable DIE shrinks to get the needed performance gains, so there will be more pressure to stack resources like memory(HBM, or other) and begin to offer wider internal system buses, to cores with more execution resources, and AMD may have its CPU/GPU integration to such a level that its APUs will be able to utilize the floating point resources of its integrated GPU with very little latency should the CPU need to call on the GPU to have the work done. All of the SOC industry will be pushed towards using UMA between CPU, and integrated GPU, and the further integration between CPU, and GPU will become necessary for continued relevance in the market.
AMD with its K12 custom ARM cores will have its own low power RISC ARMv8 ISA based custom microarchitecture to give Apple, and Nvidia the competition in the tablet market. And AMD’s K12 based APUs will be available for those OEMs that do not have Apple’s engineering resources. Intel is going to be forced into using MorphCore to get its x86 CISC based core design’s transistor count down towards even approaching the lower transistor counts of the RISC designs like ARM/MIPS/Power8 RISC base microarchitectures. MorphCore may be utilized more for mobile, and some server workloads where power usage trumps processing ability workloads, but for enthusiast SKUs were performance counts not so much. MorphCore will be Intel’s big/little solution to providing a few more powerful cores, alongside of less powerful cores for mobile devices.
DX12, 4K, and HEVC. Sounds
DX12, 4K, and HEVC. Sounds like a nice HTPC chip.
I have a question related to
I have a question related to the limit of PCI lanes. If I have a discrete GPU in my PC and I add M.2 PCI storage (my PC M.2 port is only PCI 2.0 4x) Will I have full speed in my GPU or will it split 8x to both?
You will not have full speed,
You will not have full speed, but you should have all the speed you need.
In teory as you wil be using
In teory as you wil be using PCIe 2.0 lanes for the m.2 slot, the Grphics card will continue at x16. Also, what motherboard are you using?
Well AMD’s server APUs based
Well AMD’s server APUs based on the ZEN microarchitecture, and whatever consumer derivative AMD makes from the HPC/Server part, if it has more PCIe lanes offered, may force Intel to offer more PICe lanes on any future variants. Intel for sure will be trying to limit any overlap between their Server SKUs and their high end enthusiasts SKUs so that sales are not cannibalized from their server business.
Intel without any competition will continue to limit as much as possible any overlapping between their server and high end enthusiasts SKUs, all while not having to offer any extra features like extra PCIe lanes. If AMD can get ZEN’s performance to just Sandybridge levels, and do so for less cost, many gamers will be interested. That AMD HPC/server APU will definitely lead to the introduction of a more powerful gaming APU, most likely a desktop gaming APU that can be paired with the top end AMD GPU, and offer gamers the extra value for their gaming dollars. AMD is continuing to merge the CPU with the GPU, in their APUs and with the Introduction of HBM on interposer/package, Giving its APUs more powerful graphics should be easy, epically if the CPU portion, HBM, and the GPU portion could be fabbed separately and joined together on an interposer package, with the CPU and GPU sharing HBM memory, and a very wide BUS.
AMD at the 14nm process node could probably have an APU on a single die with the graphics resources that Intel could not match, and it would not take much more than the Zen microarchitecture providing Sandybridge levels of performance to put Intel in a bind.
Ugh, maxing out at 4 cores
Ugh, maxing out at 4 cores again for consumer parts in 2015. We had consumer level quad cores from Intel in 2007. We haven’t seen a core count increase from them in 8 years!
Really I’m more interested in the Skylake Xeons but their pricing and core count will of course roughly reflect their consumer equivalents.
To pay the hyperthreading tax or go with AMD…. I’ll have to make up my mind this year.
AMD’s Zen microarchitecture
AMD’s Zen microarchitecture will support SMT, SMT(Simultaneous multithreading) is the generic computing science term for what Intel Calls HyperThreading. Even AMD’s K12 custom ARM cores will support SMT, and AMDs custom K12 based ARMv8 ISA running APUs are going to give Apple something to worry about in the tablet market. That AMD Zen microarchitecture based HPC/Server APU will be the basis for some high performance GPU accelerated consumer SKUs. AMD should be able to get some second generation HSA(AMD’s implementation of HSA) compliance and who knows how much more the Zen cores will be closely integrated with the GPU cores by the time Zen products beginning arriving for the server/HPC market.
AMD may have some extremely GPU accelerated desktop variants following shortly after the server/HPC variant of the ZEN APU hits the market. HBM on a high-end Desktop SKU will be coming, so expect some further APU style integrating of the CPU with the GPU, for more than just graphics workloads. With each new iteration of their APUs, AMD is gradually merging the CPU with the GPU.
This is nothing but pure
This is nothing but pure speculation at this point. Using the terms “will” for rumors is not usually a good idea. Many architectures from many companies were speculated as being good but fell short when reviewed. Hopefully, the new AMD architecture does pan out and exceed our expectations. Until I see “official” leaks I’ll chalk this one up to nothing more than a hopeful wish with no baring in truth whatsoever.
Maybe i am missing something,
Maybe i am missing something, but the article says this is gen 6 in 2015? Aren’t we still waiting for Gen 5 in 2015 for desktops? I mean we have gen 5 for x99 high end, supposedly 5th gen laptop is coming any day, but i’d assume 6th gen desktop is definitely 2016, right? Maybe a typo? Or are we considering the i7-9xx gen 1 and the i7-8xx gen 2? In my world i always considered i7-2xxx gen 2, i7-3xxx gen 3, and so on.
Intel sucks.
Intel sucks.