Kaby Lake information.

[IntelUser2000]

However it's unknown what it will actually support. if it is just ulra-fast SSD's or if it actually support RAM-sticks based on 3D Xpoint.

Initially its DIMM for servers only, specifically Purley. Consumer variants and the ones coming first are NVMe based. NVMe development was probably done to enable Optane devices at decent speeds.

 

[IntelUser2000]

Too bad edram is still a very nich product, except for Apple. First we got soldered in chips, then the aborted broadwell K.

Pretty easy to see nowadays why they do this. The eDRAM parts follow the same logic as the rumored $1500 Broadwell-E chip. It's for margins and revenue. The pricing is high enough that it seems almost more affordable to get a discrete card than an eDRAM equipped part.

No wonder we barely find any devices with eDRAM. It's not cheaper than discrete, it doesn't use less power and it still comes with all the disadvantages of Intel graphics - uninspired driver and support reminicient of HD Audio.

 

[zentan]

Intel Begins Shipment Of Seventh Generation Core: Kaby Lake

When Intel says they are shipping, they of course mean they are shipping to their device manufacturer partners, so we should start seeing Kaby Lake based computers this fall.
Another interesting point brought up during the call was on yields. Intel has found itself in a situation where it’s inventory levels are higher than they would like them to be, and the answer to this was yields. Intel’s yields improved in Q1, and to quote Stacy J. Smith, Intel’s CFO and EVP, “Frankly, they got a lot better in Q2 as well” which is likely another reason why Kaby Lake is being delivered on-time.
We should learn more about Kaby Lake at Intel’s IDF which is coming up in mid-August.

http://www.anandtech.com/show/10503/intel-begins-shipment-of-seventh-generation-core-kaby-lake

 

[Eug]

Asus Transformer 3 will be Kaby Lake.

http://www.pcworld.com/article/3098360/computers/intel-ships-kaby-lake-chips-pointing-to-pc-availability-in-a-few-months.html

Q3 will include Y and U right? Any other chip lines?

 

[jpiniero]

Q3 will include Y and U right? Any other chip lines?

H is like the end of the year.

 

[VirtualLarry]

I'm wondering if KBL desktop chips will run on H110 chipsets, and whether they will support DDR4-2400 and HDMI2.0.

 

[Luddite]

Anyone have any idea when the Kaby Lake Xeons would be coming out?

 

[sirmo]

Zen based APUs for client won't have HBM.

I wouldn't be so sure about that.

 

[jpiniero]

I wouldn't be so sure about that.

I don't know why you think otherwise. It's too expensive. Look at the dGPUs - they are all using GDDR5(X), and those are multiple times more expensive than the highest priced APU you will see.

Anyone have any idea when the Kaby Lake Xeons would be coming out?

Probably Q1/Q2 for the E3 ones and Q2/Q3 for the E5 1XXX. The latter of course assumes that they do release E5 Xeons that are similar to Skylake/Kabylake-X. Purley is still Q1 I think.

 

[IntelUser2000]

Probably Q1/Q2 for the E3 ones and Q2/Q3 for the E5 1XXX. The latter of course assumes that they do release E5 Xeons that are similar to Skylake/Kabylake-X. Purley is still Q1 I think.

There's no E5 Xeons coming based on Kabylake. E5/E7 goes from Skylake in 2017 to Cannonlake in 2018.

Server cores come later so they aren't bound by process problems and don't need interim steps like Kabylake.

 

[jpiniero]

There's no E5 Xeons coming based on Kabylake. E5/E7 goes from Skylake in 2017 to Cannonlake in 2018.

I'd have to think they would make some Xeons based upon the same die that Kabylake-X is. It could be branded as E3 but I'd think it would be confusing since it wouldn't be on the same socket as the regular Kabylake E3s.

 

[mikk]

E5/E7 goes from Skylake in 2017 to Cannonlake in 2018.

2019 is more likely for a highend server based 10nm product lineup because of the slow progress on the process side. But Kabylake won't come for server nevertheless.

 

[IntelUser2000]

2019 is more likely for a highend server based 10nm product lineup because of the slow progress on the process side. But Kabylake won't come for server nevertheless.

It was a nice surprise, but that's how it is.

By the way, the Xeon Cannonlake would likely be mid to late 2018, or nearly a year later from consumer Cannonlake, considering Xeon Skylake is mid 2017. That would give enough room to flesh out 10nm issues, but cut down the two year difference down to one.

 

[IntelUser2000]

2019 is more likely for a highend server based 10nm product lineup because of the slow progress on the process side. But Kabylake won't come for server nevertheless.

That won't matter so much. Cannonlake client is late 2017. Skylake Xeon is mid 2017, meaning Cannonlake Xeon is mid to late 2018. That would also reduce the difference between clien and xeon to 1 year from 2 years. It's possible it'll have just 1.5 year gap between the two.

I'd have to think they would make some Xeons based upon the same die that Kabylake-X is. It could be branded as E3 but I'd think it would be confusing since it wouldn't be on the same socket as the regular Kabylake E3s.

Kabylake-X is a different chip from Skylake-X. They just coexist on the same platform. The former is likely a high clocked gamer-focused 4 core chip while the latter being HEDT replacement.

 

[t0mt0m]

Does yield improvements for Skylake have any bearing on yield for Kaby Lake?

 

[Azuma Hazuki]

In theory yes since it's not so much "improvements for Skylake" as "improvements for Intel's 14nm process." I don't think Kabylake is going to be a very different chip from Skylake.

 

[Borealis7]

some KBL benches found:
http://www.nextpowerup.com/news/299...hows-decent-improvement-over-last-generation/
KBL was supposed to be an optimization of SKL, same architecture same node, but from what i hear Intel has a surprise in store for this product.

 

[Drazick]

What kind of a surprise you think?

 

[IntelUser2000]

What kind of a surprise you think?

According to Intel's CEO at the time, Brian Krzanich, Kaby Lake would be "built on the foundations of the Skylake micro-architecture but with key performance enhancements."

I'd feel a bit lied to if that "key" was something similar to circuitry improvements to enable higher clocks on Pascal, not architectural ones. If it was Skylake, would you not say its "Skylake with circuitry improvements for higher clock"? Plus there's almost no headroom for higher clocks, especially for overclockers.

 

[Borealis7]

Brian is still the CEO, what do you mean "at the time"?

What kind of a surprise you think?

i can't say. i only heard there was an internal mail at Intel which referred to bigger gains than expected, but it could be for mobile parts and less for desktop.

 

[Arachnotronic]

According to Intel's CEO at the time, Brian Krzanich, Kaby Lake would be "built on the foundations of the Skylake micro-architecture but with key performance enhancements."

I'd feel a bit lied to if that "key" was something similar to circuitry improvements to enable higher clocks on Pascal, not architectural ones. If it was Skylake, would you not say its "Skylake with circuitry improvements for higher clock"? Plus there's almost no headroom for higher clocks, especially for overclockers.

It's not just Skylake with circuitry improvements for higher clock. It is Skylake built on an enhanced 14nm process with circuitry improvements. There are also media encode/decode improvements.

Anyway, if KBL-S is hitting 4.5GHz max turbo out of the box, and if we know what KBL-X is coming at 112W TDP & iGPU disabled, I think it's pretty reasonable to expect a nice improvement in avg overclock achievable with these parts.

I don't want to get too hyped up, but 5GHz overclock sounds like a real possibility with Kaby Lake given the more mature 14nm yields, improved 14nm process, and improved circuit design.

 

[IntelUser2000]

Anyway, if KBL-S is hitting 4.5GHz max turbo out of the box, and if we know what KBL-X is coming at 112W TDP & iGPU disabled, I think it's pretty reasonable to expect a nice improvement in avg overclock achievable with these parts.

I don't want to get too hyped up, but 5GHz overclock sounds like a real possibility with Kaby Lake given the more mature 14nm yields, improved 14nm process, and improved circuit design.

The 4.2/4.5GHz part 7700K and KBL-X is apparently the same thing. The possibility of conflicting TDP information 95/112W may be just like with IVB when they said 95W-class but the actual TDP was 77W or something.

That 5GHz has been elusive for a decade. We have a tendency of thinking there's always going to be a breakthrough that brings massive improvements but trend in the last decade has been that progress has been progressively slowing. All at the same time when marketing hype is progressively increasing. The reality has been so-called breakthrough technologies were needed to get that extra squeeze we had in the past few years. Just like we won't see 8-second 100 meter sprinters, we may just be at human limits with technology.