Kaby Lake-G NUC "Hades Canyon" coming?

[jpiniero]

https://benchlife.info/wp-content/uploads/2017/09/Intel-NUC-Roadmap-2018-2019.png

Release according to the chart is 2Q 2018. It says dGFX which if the rumors are correct is technically true but it would be a quad core Kaby Lake-H die with an on-package Vega+HBM2 fused via EMIB. It does seem weird that they would release it after Coffee Lake but perhaps delays with AMD or EMIB happened.

 

[IntelUser2000]

Are they really going to make such a complicated chip for what would sell in mere thousands of units? The Skull Canyon NUC goes for $1000 if you spec the system out. This would cost at least that, if not more.

Also the 6 display output is kinda strange as well. HD 630 supports triple display, and none of the Vegas support more than 4 displays. That indicates the iGPU is active for display purposes.

The naming "VR" for the 100W part must be just a token gesture. With 35/45W for the KBL-H CPU there's only 55/65W for the GPU. You need an RX 480 for a minimum VR experience. To go lower than that you go with the uncertain Windows MR headset specs or the Oculus Rift's minimum spec using ATW/ASW, which is really a trick.

It does seem weird that they would release it after Coffee Lake but perhaps delays with AMD or EMIB happened.

Well, for one thing they had to wait until Vega chips were available. These would be smaller Vegas than the one used in RX Vega 56/64. Also, just putting two components into one package takes a bit of time. You'll always see two discrete components first before the integrated ones.

If they are going to use HBM, then so far, it would turn out to be the cheapest implementation of a HBM-using device. That would take time too.

 

[VirtualLarry]

Maybe they will, as an EMIB and HBM technology pipe-cleaner? I can't imagine that the limited sales numbers would make ROI for the packaging R&D expended.

 

[coercitiv]

We have briefly discussed this on the SKL/CFL thread. While the TDP numbers do fit with the old KBL-G rumors, this time the roadmap makes a clear distinction in that it openly mentinons the separation of chips - KBL-H + dGFX. In the light of the old SKU rumors and the way Intel would want to market their product, it would make little sense for them to start mentioning the EMIB based product as two different chips. Why go through the trouble to begin with?

On top of that, the top Hades NUC comes with a 7xxxK chip, which is an unlocked part. Why would Intel build this test package and allow overclocking? That would make power management even more difficult to handle.

I think there's a good chance this is a classic combo of CPU + dGPU, possibly based on MXM boards like Arachnotronic already mentioned.

The Skull Canyon NUC goes for $1000 if you spec the system out. This would cost at least that, if not more.

As an interesting side note, even though expensive when compared with DYI mini systems and even notebooks, the Skull Canyon NUC was offered for a good price when compared with similar systems from OEMs. They simply go crazy when it comes to these small factor units and ask for ridiculous prices.

The only exception to this rule that I've seen so far is the Asrock DeskMini 110 barebone which comes at a very appealing price in Europe too and is probably the cheapest and most effective way way to build a compact small factor PC. Points to VirtualLarry for spotting it early on.

 

[VirtualLarry]

The only exception to this rule that I've seen so far is the Asrock DeskMini 110 barebone which comes at a very appealing price in Europe too and is probably the cheapest and most effective way way to build a compact small factor PC. Points to VirtualLarry for spotting it early on.

Yeah, I've been a fan of those units for a while now. I just built my third one. (Had two weird freezes though, on that units. All parts are new. Not sure what's going on.)

 

[jpiniero]

We have briefly discussed this on the SKL/CFL thread. While the TDP numbers do fit with the old KBL-G rumors, this time the roadmap makes a clear distinction in that it openly mentinons the separation of chips - KBL-H + dGFX. In the light of the old SKU rumors and the way Intel would want to market their product, it would make little sense for them to start mentioning the EMIB based product as two different chips. Why go through the trouble to begin with?

Because of the 6 displays support and that they are using Kaby Lake when Coffee Lake-H should be readily available.

 

[coercitiv]

Because of the 6 displays support and that they are using Kaby Lake when Coffee Lake-H should be readily available.

Do we have an ETA for Coffee Lake-H?

Right now Intel is getting ready to launch Kaby Lake Refresh for mobile, and Coffee Lake mobile NUCs are set for the second part of 2018 as you already saw from the roadmap. Why do you think Intel would have 35W & 45W Coffee Lake mobile parts ready for NUC production in Q2 2018?

 

[jpiniero]

Do we have an ETA for Coffee Lake-H?

1Q, if not January.

 

[IntelUser2000]

Because of the 6 displays support and that they are using Kaby Lake when Coffee Lake-H should be readily available.

6 display support actually favors having a discrete GPU.

You don't actually need it on package. You can put the GPU on the motherboard like some rare vendors do. It would effectively be discrete graphics but save massive amounts of space.

The whole thing doesn't make sense anyway. Result of massive internal turmoil within Intel. I wish the problems Intel has blows up so the management is forced to admit defeat and resign, to be replaced with better ones.

 

[LTC8K6]

https://benchlife.info/wp-content/uploads/2017/09/Intel-NUC-Roadmap-2018-2019.png

Release according to the chart is 2Q 2018. It says dGFX which if the rumors are correct is technically true but it would be a quad core Kaby Lake-H die with an on-package Vega+HBM2 fused via EMIB. It does seem weird that they would release it after Coffee Lake but perhaps delays with AMD or EMIB happened.

Sounds like it just supports a dedicated graphics card.

 

[jpiniero]

You don't actually need it on package. You can put the GPU on the motherboard like some rare vendors do. It would effectively be discrete graphics but save massive amounts of space.

You could for sure, that's what laptops do more or less. It would save a little bit of space though, and it'd probably be easier to cool. Surely it's not worth the effort but is interesting at least.