Discussion Intel current and future Lakes & Rapids thread

[PeterScott]

Yeah but AMD's APUs are on a node that works. That's sort of the problem, especially when yields are only going to get worse as (if?) the shrinks continue. One of the big benefits to an EMIB design is that you could have some tiles on older nodes.

I guess you are implying this is only for Intel 10nm, that currently isn't good enough for large scale production.

Something you assume will persist forever?

The reality is that eventually Intel will fix the issues with 10nm, just like they did with 14nm, and until they do, 10nm won't see widescale production.

Breaking the CPU into multiple pieces to use 10nm is just a bit of absurd nonsense.

 

[Dayman1225]

Intel to Invest $5bn into Israel 10nm Manufacturing Plant to Expand it's Production and Capacity

They are starting work on it this year and plan to have it finished by 2020, back in 2014 they invested $6bn to upgrade this fab to 10nm and now they are spending an insane (at least to me) $5bn just to expand it, a little comparison, the 7nm Arizona Fab cost's $7bn alone just to tool (and whatever is left to finish)

 

[jpiniero]

The reality is that eventually Intel will fix the issues with 10nm, just like they did with 14nm, and until they do, 10nm won't see widescale production.

I don't think the issue is going to be resolved fully until 10++.

 

[LTC8K6]

I don't think the issue is going to be resolved fully until 10++.

I thought they basically abandoned the initial 10nm process to go with 10nm++?

 

[Dayman1225]

I thought they basically abandoned the initial 10nm process to go with 10nm++?

Original 10nm is abandoned for High Perf Desktop Procs and is to be used for Cannonlake (if it ever releases), possibly Modems(XMM 7660/8060 according to Charlie at SA), FPGA's(?) and Atom. 10nm+ will be used for Icelake and 10nm++ for Tigerlake and Sapphire Rapids supposedly.

 

[LTC8K6]

https://www.anandtech.com/show/12436/intel-10nm-dualcore-cannon-lake
Intel apparently shipped 10nm chips in 2017 somewhere.
To a black hole, apparently.

The microcode update was one brief official mention of Cannon Lake CPUs, and it was removed after it became public.
And we don't know what process they might be using, if they exist.

 

[Dayman1225]

https://www.anandtech.com/show/12436/intel-10nm-dualcore-cannon-lake
Intel apparently shipped 10nm chips in 2017 somewhere.
To a black hole, apparently.

The microcode update was one brief official mention of Cannon Lake CPUs, and it was removed after it became public.
And we don't know what process they might be using, if they exist.

Yeah the PDF mentioned Cannonlake 2+0(2 Core No GPU) and 2+2(2c +GT2 GPU) Cannonlake is 10nm because I doubt they backported it to 14nm at this point.

It's likely that they shipped a variant of either of the mentioned chips to Intel's customers (OEMs like Dell and HP)

 

[LTC8K6]

Yeah the PDF mentioned Cannonlake 2+0(2 Core No GPU) and 2+2(2c +GT2 GPU) Cannonlake is 10nm because I doubt they backported it to 14nm at this point.

It's likely that they shipped a variant of either of the mentioned chips to Intel's customers (OEMs like Dell and HP)

I'm thinking they shipped 10nm chips, but not for PCs.

I think by now someone would have run CPU-Z and noticed that they had a 10nm chip?

 

[Dayman1225]

I'm thinking they shipped 10nm chips, but not for PCs.

I think by now someone would have run CPU-Z and noticed that they had a 10nm chip?

I can imagine a 2+0 chip being used for embedded but not much else. It also depends if OEMs plan to even release systems with these chips. Though I do find it odd we have had no CNL leaks but we just seen ICL U Platform appear on SiSoft with 48 LP EUs

 

[jpiniero]

I can imagine a 2+0 chip being used for embedded but not much else. It also depends if OEMs plan to even release systems with these chips. Though I do find it odd we have had no CNL leaks but we just seen ICL U Platform appear on SiSoft with 48 LP EUs

There have been leaks - the 8121U is definitely Cannonlake. What's not clear is if that is the 2+0 or 2+2 model. Figure there won't be much availability of either - the 8130U is likely the replacement for the 2+2 since yields at this point are still terrible.

 

[Dayman1225]

There have been leaks - the 8121U is definitely Cannonlake. What's not clear is if that is the 2+0 or 2+2 model. Figure there won't be much availability of either - the 8130U is likely the replacement for the 2+2 since yields at this point are still terrible.

Forgot about that leak, guess we gotta wait and see at this point

 

[IntelUser2000]

There have been leaks - the 8121U is definitely Cannonlake. What's not clear is if that is the 2+0 or 2+2 model. Figure there won't be much availability of either - the 8130U is likely the replacement for the 2+2 since yields at this point are still terrible.

So there are references to Cannonlake with 16EUs.

The 2+2 CNL may be a GT2 chip but with 16EUs enabled. Rumors were pointing to Intel having problem with the larger GPUs on Cannonlake. Since Gen 10 allows asynchronous EU/subslice configuration, 16EUs become a possibility even if it only has odd number of subslices.

It also takes care of the problem of a CNL-based 15W Core i3 being faster than any other 15W Core chip not based on Cannonlake for graphics, and maybe even threatening the larger GT3e part. If it had 40EUs, and Gen 10 has further bandwidth saving features, it may be possible that it'll overlap GT3e performance without needing eDRAM. That complicates positioning especially when higher clocked chips can't be pumped out easily with Cannonlake.

I think 8121U is going to feature lower top Turbo frequency than 8130U, with a cut down GT2 EU part but based on Cannonlake and Gen 10.

I can imagine a 2+0 chip being used for embedded but not much else.

"Embedded" usage for consumer chips usually require iGPUs. Digital signage, PoS kiosks, cash registers, all need display out.

There are only two very niche scenarios I can think of. One, being used in a low end laptop where the manufacturer will put a discrete GPU anyway. Two, a company selling storage servers where the high end is taken up by Xeon D, and low end by this chip.

 

[LTC8K6]

We have had Ice Lake leaks as well, at least one.

https://wccftech.com/intel-10nm-cannonlake-ice-lake-processors-spotted/

 

[IntelUser2000]

Two, a company selling storage servers where the high end is taken up by Xeon D, and low end by this chip.

It may just not be selling at all. If they could put out a 2+2 part out, even a partially disabled one, why launch GT0(no graphics) part at all? Heck, a basic display adapter with 4EUs would be better than nothing.

I thought they basically abandoned the initial 10nm process to go with 10nm++?

You mean, they cancelled 10nm to go with 10nm+.

 

[LTC8K6]

It may just not be selling at all. If they could put out a 2+2 part out, even a partially disabled one, why launch GT0(no graphics) part at all? Heck, a basic display adapter with 4EUs would be better than nothing.



You mean, they cancelled 10nm to go with 10nm+.

As if it couldn't get any worse, by Intel's own admission, its first- and second-generation 10nm technologies -- 10nm and 10nm+, respectively -- will offer worse performance than its upcoming 14nm++ technology . Intel says the company's 10nm technology won't open up a clear performance lead over its 14nm++ technology until its third iteration -- known as 10nm++ -- which should go into production sometime in 2020.

I was assuming 10nm++ was now coming earlier than 2020, since they said 10 and 10+ were disappointing next to 14++, and we've had some urgency added...

https://www.fool.com/investing/2017/07/26/the-price-of-intel-corporations-10-nanometer-failu.aspx

 

[jpiniero]

It may just not be selling at all. If they could put out a 2+2 part out, even a partially disabled one, why launch GT0(no graphics) part at all? Heck, a basic display adapter with 4EUs would be better than nothing.

Because a large percentage of the chips that manage to have a functional CPU have a defect in some critical part of the GPU. That's the only real explanation.

 

[IntelUser2000]

I was assuming 10nm++ was now coming earlier than 2020, since they said 10 and 10+ were disappointing next to 14++, and we've had some urgency added...

The top transistor performance may lag behind few %, but even the base 10nm has benefits over 14nm++ that can easily overshadow the differences. Designers can use additional density and power reductions(which are greater than conventional node jumps), to improve performance. Since chips are largely thermal/power bound nowadays, improvements in power efficiency with 10nm can be used instead for better performance.

Even if the 10700K Icelake-S clocks 100-200MHz behind 9700K, long-waited architectural improvements will make up for it, then some. I think due to the possibility of 8 cores clocking at 4.5GHz or more that pushes the thermal/power boundaries of the 14nm++ process, 10nm's power efficiency will allow it to clock just as high.

 

[Dayman1225]

According to Ashraf Eassa over at TheMotelyFool, Intel's 7nm plans are aggressive and the first Client 7nm Architecture is named Meteor Lake and the successor to Sapphire Rapids is called Granite Rapids

Intel's 7nm plans are REALLY aggressive. Sadly, their plans for 10nm were aggressive too and look where that is - Ashraf

First Intel 7nm architecture is Meteor Lake (MTL). - Ashraf

First Intel 7nm server architecture is Granite Rapids, successor to Sapphire Rapids. - Ashraf

 

[jpiniero]

I was assuming 10nm++ was now coming earlier than 2020, since they said 10 and 10+ were disappointing next to 14++, and we've had some urgency added...

As it stands, Cascade Lake is like a end Q3- early Q4 product... so you figure Icelake Server would be at least a year after that... even with Matisse out there looming. So you're already near the end of 2019 already. 10++ sounds like it is essentially a reboot so I don't know if it'd be possible to pull it into 2019 if they say were to cancel Icelake Server and release Sapphire Rapids early.

 

[Bouowmx]

aggressive

What is aggressive about it? Some details not for public?

 

[Eug]

OK, with the introduction of Kaby Lake Refresh and the mixed 8th generation, I'm getting quite confused.

From this thread it seems people are saying there is likely no such thing as Kaby Lake Refresh Y series, because it'd still be dual-core, which already exists as Kaby Lake. I guess that would explain why there are no leaks about KL-R Y. And there won't be Coffee Lake Y either.

So what is likely the next Generation Y Series then and why? 10 nm Cannon Lake? Will that be called 8th generation or 9th generation? When?

 

[jpiniero]

So what is likely the next Generation Y Series then and why? 10 nm Cannon Lake? Will that be called 8th generation or 9th generation? When?

Intel had planned for Cannonlake to be the 8th gen Y models, and since Cannonlake is pretty much toast they don't really have a replacement. I suppose they could rebrand the Kaby Y parts to cover for it like they did with the dual core U models.

After that is Icelake in early 2019 sometime.

 

[Eug]

Intel had planned for Cannonlake to be the 8th gen Y models, and since Cannonlake is pretty much toast they don't really have a replacement. I suppose they could rebrand the Kaby Y parts to cover for it like they did with the dual core U models.

After that is Icelake in early 2019 sometime.

That wouldn't be good if Cannon Lake is toast. It sounds like from your previous posts that you believe the Cannon Lake U parts are almost just PR products.

But then again, if Y's update really is going to take that long, it makes my 2017 Kaby Lake Y purchase sound like smart timing on my part.

No there were changes to Kabylake that made video decoding more energy efficient. Also Kabylake-R has updated video output (HDMI 2.2 iirc). Since the i3 KL-R has this as well it's probably a salvaged die.

I think you probably mean HDCP 2.2. However, Kaby Lake 7Y32 (2017) already has HDCP 2.2 (and HDMI 2.0 with the right adapter). That would give it full 4K HEVC DRM support for an external monitor in theory, at least if the OS supports it.

http://www.cpu-world.com/sspec/SR/SR346.html

Furthermore, I already know it's relatively efficient for video decoding, as it has full hardware support for HEVC 10-bit 4K. I can multitask on this Y series chip while playing back (smoothly) 75+ Mbps 4K 60 Mbps HEVC 10-bit HDR.

http://4kmedia.org/sony-camping-in-nature-4k-demo/

It was the first SR2ZY stepping of 7Y30 that did not support HDCP 2.2, but the later SR347 stepping of 7Y30 does support HDCP 2.2. However, AFAIK, all of the 7Y32 chips support HDCP 2.2.

https://www.notebookcheck.net/Intel-HD-Graphics-615.174029.0.html

The revised video engine now supports H.265/HEVC Main10 profile in hardware with 10 bit colors. Furthermore, Googles VP9 codec can also be hardware decoded. The first models do not support HDCP 2.2 and therefore Netflix 4K. This was added with the new models in 2017 and beginning of May 2017 also the older CPU models like the m3-7Y30 are now shipped with a new S-Spec Code that supports HDCP 2.2. HDMI 2.0 however is still only supported with an external converter chip (LSPCon).

 

[jpiniero]

Yeah I don't know what Apple is gonna do if they intend to refresh the Macbook this year.

 

[Dayman1225]

Yeah I don't know what Apple is gonna do if they intend to refresh the Macbook this year.

Kaby Lake R, Coffee Lake U and Coffee Lake H.