Intel Skylake / Kaby Lake

[LTC8K6]

This is that same rumor about Intel using AMD for integrated graphics, except it adds HBM2 to the rumor.

 

[mikk]

It isn't integrated according to the slide, Intel calls it dedicated. It even is possible to use Intels integrated GPU for display connection. There is a separate chip if true, but no idea how this is going to work.

 

[LTC8K6]

Kaby Lake G Series Processor Info Surfaces
http://www.guru3d.com/news-story/kaby-lake-g-series-processor-info-surfaces.html

 

[raghu78]

The root source for tpu, computerbase, guru3d and wccftech articles is benchlife

https://benchlife.info/intel-kaby-lake-g-confirm-in-roadmap-with-14nm-04032017/

If we see HBM2 on an Intel CPU in late 2017 that means AMD should probably be able to do the same with Raven Ridge. That would be some thing especially given how the usual Intel supporters said HBM2 is not possible on a mainstream CPU or APU in 2018 or even 2019. I am quite sure AMD is not licensing Vega to Intel as they would want to retain the competitive edge over Intel in graphics. I can see a Polaris based GPU in Kabylake G for sure.

I am very excited about the possibilities with HBM2 in APUs . Raven Ridge will achieve the best possible result due to a single die approach with Infinity Fabric connecting all the main components. Kabylake should also do very good with Polaris based GPU and HBM2.

 

[Sweepr]

Raven Ridge will achieve the best possible result due to a single die approach with Infinity Fabric connecting all the main components.

Raven Ridge packs a rather small 11 CUs iGPU and uses DDR4, no mention of HBM/HBM2 in any credible leak and no successor till 2019 according to the latest leaked roadmaps. Also, off-topic.

 

[Ajay]

Also, is it known whether Skylake-X will be on 14nm or 14nm+?

14nm+. Should be the best Intel HEDT yet. SKLX will be even better if completion from Ryzen leads to lower prices.

 

[ozzy702]

14nm+. Should be the best Intel HEDT yet. SKLX will be even better if completion from Ryzen leads to lower prices.

I'm assuming this will mean high (4.5ghz+) OC clocks for 6-core parts which is really what I'm interested in as a 7700k user. I don't want to sacrifice clock speed for moar cores.

 

[liahos1]

is 4.5ghz a big deal? My 5820k gets to 4.5 stable ...not sure if thats a good chip or not. I was hoping you could get to something like 4.8? I know the broadwell-e's were a step back in terms of OCing though

 

[ozzy702]

is 4.5ghz a big deal? My 5820k gets to 4.5 stable ...not sure if thats a good chip or not. I was hoping you could get to something like 4.8? I know the broadwell-e's were a step back in terms of OCing though

I don't think most six cores are able to easily hit 4.5ghz, at least not without high end cooling. When I say 4.5ghz+ I mean even chips that are on the low end of the silicon lottery will hit 4.5ghz or more. That would put golden chips in the 4.8ghz+ range.

 

[DrMrLordX]

is 4.5ghz a big deal?

Most Intel HEDT fans are looking for at least a return to the clockspeeds they could get under 22nm. Broadwell-E would generally not clock that high.

 

[Arachnotronic]

I'm assuming this will mean high (4.5ghz+) OC clocks for 6-core parts which is really what I'm interested in as a 7700k user. I don't want to sacrifice clock speed for moar cores.

I would think so. First gen 14nm and Broadwell architecture were both bad. As we saw, Skylake on 14nm clocked better than BDW did, and as we saw with KBL, 14nm+ gives a nice clock boost.

SKX-X should deliver both.

 

[LTC8K6]

The root source for tpu, computerbase, guru3d and wccftech articles is benchlife

https://benchlife.info/intel-kaby-lake-g-confirm-in-roadmap-with-14nm-04032017/

The Benchlife piece was already posted, but it's difficult to interpret...translating via Google is spotty at best for this sort of thing. Better to post another source that has figured out what the Benchlife article says.

 

[LTC8K6]

It isn't integrated according to the slide, Intel calls it dedicated. It even is possible to use Intels integrated GPU for display connection. There is a separate chip if true, but no idea how this is going to work.

No idea why AMD would shoot itself in the foot...

 

[DrMrLordX]

No idea why AMD would shoot itself in the foot...

Not sure that AMD would be shooting itself in the foot. If AMD can pair one of their dGPUs with a certain number of Intel CPUs then it would amount to a lot of sales that would/could otherwise go to Nvidia.

 

[LTC8K6]

Not sure that AMD would be shooting itself in the foot. If AMD can pair one of their dGPUs with a certain number of Intel CPUs then it would amount to a lot of sales that would/could otherwise go to Nvidia.

Also sales that would/could have gone directly to AMD in the form of APUs and low end cards.

Why would AMD want to help sell Intel chips with AMD graphics over their own new APUs?

Hey, maybe AMD could just buy finished APUs from Intel soon?

It sort of seems to indicate that their APUs aren't that good...even though that may not be true for the upcoming chips.

How good is the HBM2 supply?

I'll believe this pairing when I see it.

 

[Sweepr]

TechPowerUp's take on Kaby Lake G

...For now, rumors peg these Kaby Lake-G as special BGA processors based on Kaby Lake, with an additional discrete GPU on the package. The TDP of these processors (at 65 W and 100 W) is well above the Kaby Lake-H's known 45 Watts. Which begs the question: what exactly is under the hood? This, including Intel's modular approach to chip design for which it developed its EMIB technology, could probably account for the AMD graphic's chip TDP - a discrete-level GPU which would be integrated on-die, EMIB's routing layers handling the data exchange between GPU and processor. This is where HBM 2 memory integration would also come in, naturally - a way to keep a considerable amount of high-speed memory inside the package, accessible by the silicon slices that would need to. Nothing in the leaked information seems to point towards this HBM 2 integration, however.



Also helping these "AMD Radeon IP integration" story (besides TDP) is that the two chips that will be part of the Kaby Lake-G series will feature a package size of 58.5 x 31mm - bigger than a desktop Kaby Lake-S (37.5 x 37.5 mm) and the Kaby Lake-H series chips (42 x 28mm). The extra space would accommodate increased footprint of the GPU package - though for now, leaked information points only, again, to Intel's own GT2 graphics solution, though Benchlife seems to put much stock on the AMD side of the equation...

www.techpowerup.com/232067/rumored-intel-kaby-lake-g-series-modular-multi-die-hbm-2-amd-graphics-ip

 

[IntelUser2000]

If we see HBM2 on an Intel CPU in late 2017 that means AMD should probably be able to do the same with Raven Ridge. That would be some thing especially given how the usual Intel supporters said HBM2 is not possible on a mainstream CPU or APU in 2018 or even 2019. I am quite sure AMD is not licensing Vega to Intel as they would want to retain the competitive edge over Intel in graphics. I can see a Polaris based GPU in Kabylake G for sure.

There's is a difference. Cost to implement. If it uses EMIB like some articles are saying. It may mean Intel's claims of EMIB being much more cost effective are true.

So if it works as the rumors I am assuming this setup will occupy the space Iris Pro did, or at least tried to. The highest end with pricing to match. Presumably the lower end will be occupied by Raven Ridge.

 

[coercitiv]

I'm amazed by how many people (and and publications) chew on this "integrated" dGPU /w HBM story. There's every sign possibile that it's fake, yet some just want to believe.

 

[IntelUser2000]

I'm amazed by how many people (and and publications) chew on this "integrated" dGPU /w HBM story. There's every sign possibile that it's fake, yet some just want to believe.

I would take the claims with a good heap of salt, but the rumors persist. Also the KBL-G talks about being for graphics but only has GT2 according to the slide.

-It's not next enthusiast chip, that's for the -X version
-It's has a TDP rating of 100W. Unless they for some idiotic reason want a third enthusiast line aside from -X and the -K chips
-It can't be their own GPU, otherwise it would be on-die

Logically then just looking at the slide it connects very well to the GPU on package not from Intel rumor. Some company decisions do not immediately make sense(or won't until they explain). We can only think of the possibilities for them wanting to take such a road.

 

[coercitiv]

Logically then just looking at the slide it connects very well to the GPU on package not from Intel rumor. Some company decisions do not immediately make sense(or won't until they explain). We can only think of the possibilities for them wanting to take such a road.

I'm not talking about using external IP for a new GPU or the EMIB, it's the rest of the info that makes no sense.

• HBM is too expensive for midrange dGPUs from both Nvidia and AMD, but somehow Intel's gonna put it in a mainstream product. Sniff sniff.
• APUs shine in mobile / low power scenarios where dGPUs have a hard time competing, but Intel goes directly after desktop class with 100W SKUs. Guns blazing!
• It's Intel's first attempt at EMIB, presumably using external IP as well, yet they go BIG from the start.

Enthusiasts have been wishing for this kind of integration for quite a while now (the fast high performance APU), and this rumor tends to cater more to what we dream rather than what is viable from a financial/technological point of view.

Imagine the rumor was only about Intel's first attempt of building a commercial product based on EMIB, using external graphics IP, in the form of a very cost effective mobile solution. Hardly a click generator, needs some spices

 

[IntelUser2000]

Enthusiasts have been wishing for this kind of integration for quite a while now (the fast high performance APU), and this rumor tends to cater more to what we dream rather than what is viable from a financial/technological point of view.

I know, it does seem out of line with what's been done in the past few years.

Actually, you can't put much of a GPU in a 100W chip. Not what you'd call a high performance in discrete terms. RX 460 is at 75-100W depending on the version. I'd have to think based on how competitive Maxwell and Pascal was in terms of perf/watt against integrated solutions the cost of being discrete isn't as great as we thought before. It just puts a cap on how low you can go.

You are talking about 2-2.5x the performance difference between RX 460M and Iris Pro 580. Well, a little greater but I'd pick AMD to have better gaming drivers and the architecture being more mature, also the Iris Pro 580 has TDP limits. So if we talk about 30-50% gain over Iris Pro 580, we get 2/3rd the performance of the RX 460 part. Which seems like a good generational jump but not so out of line.

Its actually Intel's second attempt at EMIB. The first is a rather large chip in the Altera FPGA. Could we say conclusively that putting a single HBM/HBM2 chip is more expensive than Intel's custom eDRAM? I'd have thought based on looking at presentations the main cost of packaging is not so much the die itself but how you integrate the die. It's probably more the difficulty of integration possibly contributing to the risk and delay of the part.

 

[DrMrLordX]

Also sales that would/could have gone directly to AMD in the form of APUs and low end cards.

Why would AMD want to help sell Intel chips with AMD graphics over their own new APUs?

I have no idea what AMD will be able to command price-wise for Raven Ridge, but in general margins on their dGPUs have been much higher than their existing crop of APUs. Depending on what was "integrated" with KBL-G, AMD might wind up making more money just selling the dGPU parts at the expense of their own product stack.

 

[Sweepr]

Another DDR4 frequency world record with a Core i7-7700K, this time 5280 MHz using a T-FORCE XTREEM DDR4-3733Mhz CL18-20-20-39 16GB-Kit and an ASROCK Z170M OC FORMULA motherboard.

http://www.ocstation.com/reviews/Se...-force-xtreem-ddr4-5280mhz-world-record-again


Talking about fast memory...

The Joint Electron Device Engineering Council says the DDR5 standard should be finalized in 2018, although that doesn’t necessarily mean you’ll be able to buy the memory next year. It could take a little longer for DIMMs to actually hit the streets, since hardware makers will also need to update motherboards to work with DDR5 memory.

As Ars Technica points out, the DDR4 spec was finalized in 2012, but DDR4 memory wasn’t widely available until 3 years later.

https://liliputing.com/2017/03/ddr5-memory-will-twice-fast-ddr4.html
http://www.pcworld.com/article/3186737/components/staying-alive-ddr5-memory-is-on-its-way.html

 

[Sweepr]

PCLab: 35 Mobile CPUs Compared


- List of Processors Tested

Intel:
Core i7-7700HQ, Kaba Lake, 4R / 8W, 2.8 GHz + Turbo
core i7-6820HK, Skylake, 4R / 8W, 2.7 GHz + Turbo
Core i7-6700K, Skylake, 4R / 8W, 4.0 GHz + Turbo
Core i7-6700HQ, Skylake, 4R / 8W, 2.6 GHz + Turbo
Core i7-4980HQ, Haswell, 4R / 8W, 2.8 GHz + Turbo
Core i7-4790S, Haswell, 4R / 8W, 3.2 Turbo GHz +
Core i7-4720HQ, Haswell, 4R / 8W, 2.6 GHz + Turbo
Core i7-2630QM Sandy Bridge, 4R / 8W, 2 GHz + Turbo

Core i7-7500U, Kaba Lake, 2R / 4W + 2.7 GHz Turbo
Core i7-6500U, Skylake, 2R / 4W + 2.5 GHz Turbo
Core i7-5600U, Broadwell, 2R / 4W, 2.6 GHz + Turbo
Core i7-5500U, Broadwell, 2R / 4W + 2.4 GHz Turbo

Core i5-6300U, Skylake, 2R / 4W + 2.4 GHz Turbo
Core i5-6200U, Skylake, 2R / 4W + 2.3 GHz Turbo
Core i5-5257U, Broadwell, 2R / 4W, 2.7 GHz + Turbo
Core i5-5200U, Broadwell, 2R / 4W + 2.2 GHz Turbo
Core i5-4210U, Haswell, 2R / 4W + 1.7 GHz Turbo

Core i3-6100U, Skylake, 2R / 4W, 2.3 GHz
Core i3-5010U, Broadwell, 2R / 4W, 2.1 GHz
Core i3-4030U, Haswell, 2R / 4W, 1.9 GHz
Core i3-4000M. Haswell, 2R / 4W, 2.4 GHz

Core m7-6Y75, Skylake, 2R / 4W + 1.2 GHz Turbo
Core m5-6Y57, Skylake, 2R / 4W, 1.1 GHz + Turbo
core m3-6Y30, Skylake, 2R / 4W, 0.9 GHz + Turbo

Core M-5Y71, Broadwell, 2R / 4W + 1.2 GHz Turbo
Core M-5Y70, Broadwell, 2R / 4W, 1.1 GHz + Turbo
Core M-5Y51, Broadwell, 2R / 4W, 1.1 GHz + Turbo
Core M-5Y10c, Broadwell, 2R / 4W + 0.8 GHz Turbo
Core M-5Y10, Broadwell, 2R / 4W, 0.8 GHz + Turbo

Pentium N3700, Braswell, 4R / 4W + Turbo 1.6 GHz
Pentium 3805U, Broadwell, 2R / 2W, 1.9 GHz

Atom x5-Z8500, Cherry Trail, 4R / 4W + 1.44 GHz Turbo
atom X5-Z8300 Cherry Trail, 4R / 4W + 1.44 GHz Turbo

Celeron N3050, Braswell, 2R / 2W, 1.6 GHz + Turbo

AMD:
AMD FX 7500, Kaveri, 4R / 4W, 2.1 GHz Turbo +

- Average Performance (Core i7-6500U = 100)

Interesting bit here, look at how performance varies depending on the notebook model tested. These are all Core i7-6700HQ laptops:

http://pclab.pl/art72904.html

 

[LTC8K6]

I guess with mobile the main difference is the cooling of the cpu? Better cooling giving more turbo time?

I wonder what variation you'd get if you compared a bunch of desktop systems from different mfgs with the same CPU?