Things I'd like to see in Skylake E

Broadwell E is right around the corner and unless the prices are drastically decreased in comparison to what Haswell E chips were priced at I see no reason to upgrade so I'm just going to skip over that all together and jump to Skylake E and what I'd like to see from Intel in about a year or two:

  1. HSA: Such a weird request, I know. If this was implemented then in theory it would possible for tasks to be divided more intelligently between the processor and the graphics card. As it stands, only AMD products have support for this however I see potential for this if Intel jumped on board.

  2. An iGPU: Once again, weird. For those of you running Linux with a 5820k isn't it annoying looking at your build with two graphics cards that may not even match up? I would be annoyed. This would be very useful for KVM and troubleshooting in general... not to mention if HSA was involved the iGPU could calculate whatever the CPU had to if needed.

  3. Higher stock clocks: Let's face it; the 5820k could easily handle 4.2ghz at 1.1v tops. At stock mine went all the way up to 1.05v at 3.3ghz which isn't logical. Heat output isn't something that we would need to worry about due to how Skylake was redesigned.

  4. A major increase in IPC: Does the average user really notice the difference between a 2500k and something like the 4790k in daily tasks? No. Sandy Bridge owners don't really have a reason to upgrade as of now besides newer technologies.

  5. Lower prices: I know this is subjective, but spending $600 more for two cores and 12 lanes doesn't make sense to me. I understand H-E is geared towards enthusiasts, but would it hurt to charge maybe $250 for the 5820k, $375 for the 5930k, and $500 for the 5960x? I mean I'm already dishing out ~$250 on a board for an architecture that's a year old.

That's what I'd like to see from Skylake E next year/ early 2017.

I thought E series were the same as non E just moar cores and pcie lanes. I don't expect any IPC boost in them. We are nearly at the wall anyway.

E series parts usually have slightly more cache, soldered heatspreaders, 4 channel memory, more pcie lanes, support for more than 1 socket (xeon parts) and butt loads of server grade things.

What I think is going to be the major thing with skylake-e is more than 6 cores for the mainstream parts, I'm thinking 8 or even maybe 10 for the 7820k.
Onboard L4 cache would also be nice, since I don't really expect broadwell-e to have L4 cache like the mainstream chips.

1 Like

I really hope I can drop one into an X99 board...

Broadwell-e is definitely going to be X99
But I feel like skylake-e is going to be a really weird creature, for such massive CPUs you really need integrated voltage regulation.
If they're using FIVR on skylake-e I really don't see any technical reason to why they wouldn't be able to use the X99.
X99 already got all the features that the 170 boards have, but the only thing I can really think of would be pci-e 4.0, but that's set for 2017 I think and not late 2016 like skylake-e is going to be.

1 Like

Yeah. I don't see them moving that off the chip like they did for the mainstream even though it will generally keep temps that much higher. I mean it wouldn't make sense to ditch X99 two years in when X58 lasted 3 years and X79 lasted three years...

The thing about modern intel chipsets is that they aren't as locked down, since 95% of the stuff occurs on the CPU package itself.
The platform controller hub is mostly just responsible for stuff like sata, usb and other similar functions.

Yeah. I just wish their microcode was open source. How epic would it be if the 5820k was a locked down 5930k and we could unlock the other pcie modules? :)

Thing is, the 5820k is exactly that.
It doesn't make any sense other than marketing for the 5820k to be 28 lanes, no binning can do that because its not missing 12 pcie modules or anything, since all the lanes go through the same bus to the CPU.

I'm pretty sure you could unlock overclocking on the xeon parts and open up pcie lanes by fiddling with microcode and bios settings.

If skylake-E chips come out, there will be a new chipset for that.
And most likely not be backwards compatible with X99.
That would not be very logical, since X99 does not has dmi3.0.

I know it's possible to load custom microcode on top of the Linux kernel, but it's a question of flashing the code to the part itself... I need to try to find the 5930k microcode and loading it once my rig is back up and running... It would be an interesting expire want granted I still need a means of flashing it..

Points,1,2 and 5 would never happen.

but i think the core clocks of the alleged "6820k" would probably be up by 10% Intel is already pushing the clocks to the point where you don't even need to overclock them anymore.

My guess will be, that Intel will find a reason to sell new bunch of motherboards.
I would really like to see freaking performance increase, more that 2-3%. For the last may be 5 gens of Intel CPUs we barely have what, 20% performance increase? 2500K is still reliable cpu In 9/10 cases.
The only reason currently to spend money on X99 is the core count.

1 Like

well the biggest advantage that skylake has over haswell in general is dmi 3.0
Which means 4x pci-e 3.0 lanes availeble for the chipset.
2x more bandwidth for usb3.1, ultra m.2, sata express and all that kind of connectivity stuff.

Correct me if I'm wrong, but isn't DMI a feature dependent on the chipset? In theory, couldn't a DMI 3.0 enabled device work on a 2.0 link?

it will but not at its full potential speeds i think?
USB 3.1 after intergration on X99 boards is an example of this basicly.

I'm exaly currious "if" there is any significant diffrence between using usb3.1 on X99 and on Z170. (dmi 2.0 vs dmi3.0)
If you use any other ultra m.2, or sata express storage devices.
Maybe @wendell has some better idea´s about this.

Of course i can be totaly wrong, but it do think that dmi2.0 on haswell-E can be some kind of a limmited factor wenn you utilize more sata-e or ultra m.2 storage devices, and usb 3.1

1 Like

No idea about the linux thing, but I did fiddle around with microcode to make a lga 771 cpu work on a lga 775 board, the two sockets are physically different and have a different pin count, oddly enough only 2 pins needed to be swapped along with microcode.

1 Like

The issue is modern microcode uses blobs that are closed source so doing anything is going to be a pain in the arse.

Well, the DMI3.0 with skylake -- think of DMI as basically a "hub" bus connected to the CPU. With DMI 3.0, its almost sorta kinda like 4x pci E link, but a lot of peripherals share that on skylake.

On X99, DMI isn't used the same way. You've got 40 PCIe lanes wired directly into the CPU. So M.2/pcie storage, graphics, etc. are connected directly to the CPU. It doesn't need to go through the DMI hub. On Z170, you can use the "other" x16 slot (x8 eletrical) to plug directly into the cpu for pcie m.2 or whatever, or use the other pcie slots for pcie connectivity through the DMI interface. For "slow" pcie peripherals like usb adapters (including onboard usb 3.1 asmedia controllers) and that kind of thing, it makes sense. DMI connection for graphics adapters makes less sense. DMI for e.g. Intel 750 also is iffy, but for slower m.2 devices (< 1 gbyte/sec) DMI is probably fine.

So for X99 they'll probably do DMI 3.0, sure, and the USB 3.1 will probably be connected through that. But you can also buy a PCIe addin card with the same asmedia usb 3.1/type c if you want right now today and be totally fine.

Make sense?

2 Likes

That's a really great explanation as i was struggling understanding what DMI was. Thanks Wendell!

I've been considering an X99 upgrade myself (5820k is 299 @ microcenter right now!) but have been uneasy about making the move now. I'm really unsure if Broadwell-E is even and more generally Intel's plan for the enthusiast platform beyond that.