Discussion Intel current and future Lakes & Rapids thread

[IntelUser2000]

Late 2020 brings Tigerlake with Xe graphics. So by late 2021 we should see a 7nm version. I don't expect a datacenter oriented product to come earlier than that no matter how much they want to claim otherwise.

Late 2021 for a 7nm desktop is also possible. The enthusiast SKUs don't require a lot of volume.

If you look at Intel's history, they typically brought enthusiast parts first. QX9650, 2600K, 3770K comes into mind.

Source? If true, this would totally suck for Intel.

Not really. The node names mean little. Intel 14nm to 10nm is a full node jump. So is their 7nm. The foundry 7nm is almost a half node from 10nm. 5nm is another half from 7nm. Intel is just taking longer in between. In essence, the results are the same.

 

[Bouowmx]

2020: Tiger Lake/Willow Cove; 10 nm (++? according to marketing)

So, 2021 (the return of leading edge to desktop?) would be:
Golden Cove ("Alder Lake"); 10 nm (further optimization)
or
Golden Cove ("Meteor Lake"); 7 nm
?

Sorry for asking again.

 

[Ajay]

Xe in 2021 is their lead 7nm product and Navin Shenoy, head of DCG, said that 7nm DC CPU would come after the GP-GPU, so in 2022 - Granite Rapids - as for Desktop? Not sure, could see Meteor Lake in 2022 on 7nm.

1. So, a very slow ramp up to HVM for 7nm. More good news

 

[Ajay]

Not really. The node names mean little. Intel 14nm to 10nm is a full node jump. So is their 7nm. The foundry 7nm is almost a half node from 10nm. 5nm is another half from 7nm. Intel is just taking longer in between. In essence, the results are the same.

Right, but the timing, performance , density and yields mean everything for 7nm. With 10nm being such a disaster, Intel need to get high performance, high core count CPUs out into the market as fast as possible to recapture it’s brand image.

 

[jpiniero]

Late 2021 for a 7nm desktop is also possible. The enthusiast SKUs don't require a lot of volume.

Yeah not happening. The impression that I am getting is that even meeting Aurora's needs in 2021 with the GPUs is going to be tough and they are only doing it because they will lose the contract if they don't. And then Intel is going to be so anxious to get a Xeon out next by that point, that will surely be the second product out ASAP but not in 2021.

 

[TheGiant]

do we have any info on backporting sunnycove a next to 14nm more +?

 

[DrMrLordX]

Late 2020 brings Tigerlake with Xe graphics. So by late 2021 we should see a 7nm version.

That's very optimistic. If true, great, but . . .

 

[mikk]

do we have any info on backporting sunnycove a next to 14nm more +?

Not, not really. Intel told some time ago the first Sunny Cove products will be on 10nm process, so we "get the advantage of the shrink factor by moving to the new process technology".....which implied Sunny Cove is coming on 14nm some time after Icelake. The only product could be Rocket Lake as far as we know. And Rocket Lake might not come before 2012 given that Comet Lake-S is planned for H1 2020. If Rocket Lake is just anther Skylake refresh...horrible.

 

[IntelUser2000]

That's very optimistic. If true, great, but . . .

It has nothing to do with optimism. They said the datacenter dGPU as the lead product for 7nm in 2021. So I expect them to do that in 2021. Based on Tigerlake, it'll be late 2021.

2020: Tiger Lake/Willow Cove; 10 nm (++? according to marketing)

So, 2021 (the return of leading edge to desktop?) would be:
Golden Cove ("Alder Lake"); 10 nm (further optimization)
or
Golden Cove ("Meteor Lake"); 7 nm
?

Honestly? No one outside of Intel knows. And maybe, only the core group.

 

[DrMrLordX]

It has nothing to do with optimism.

Of course it does. You're assuming 7nm is going to work out on schedule and that Intel will have Xe enterprise ready before Dec 2021.

 

[Ajay]

Yeah not happening. The impression that I am getting is that even meeting Aurora's needs in 2021 with the GPUs is going to be tough and they are only doing it because they will lose the contract if they don't. And then Intel is going to be so anxious to get a Xeon out next by that point, that will surely be the second product out ASAP but not in 2021.

This is what I don't get, Aurora won't demand all of Chandler's new Fab, even with low yields. I also thought Intel brings up at least two fabs at once (which *was* the plan with 10nm). Surely there is enough capacity to ramp the first 7nm server core (Sapphire Rapids?). This won't likely be HVM, but at least final silicon in some reasonable volume, unless - 7nm is, again, in ridiculous state vis-a-vis functional yield.

 

[jpiniero]

This is what I don't get, Aurora won't demand all of Chandler's new Fab, even with low yields.

Doesn't sound like something that will be done at the Arizona fab. Probably just a small run in Oregon. It's essentially like risk production at the foundries, that's what it sounds like actually.

 

[Ajay]

Intel has already stated that the new Arizona fab (#42) will be coming online in 2020 with 7nm tooling. So it will be an ideal fab to ramp up 7nm EUV. I don’t know what is going on with D1D or D1X.

 

[Dayman1225]

Gen12 (or Xe) is starting to show up on GFXBench, no results though.

https://gfxbench.com/device.jsp?D=I...aseInternal&os=Windows&api=gl&benchmark=gfx40

 

[mikk]

https://browser.geekbench.com/v4/cpu/13954547

HP Spectre x360 gets i7-1065G7

 

[jpiniero]

Intel has already stated that the new Arizona fab (#42) will be coming online in 2020 with 7nm tooling. So it will be an ideal fab to ramp up 7nm EUV. I don’t know what is going on with D1D or D1X.

Presumably there's an Oregon fab involved, because of Copy Exactly, and the Aurora Xe chips will be used as part of the yield learning I guess. Maybe?

It is possible that Granite Rapids would start final production in late 2021.

 

[Ajay]

Presumably there's an Oregon fab involved, because of Copy Exactly, and the Aurora Xe chips will be used as part of the yield learning I guess. Maybe?

It is possible that Granite Rapids would start final production in late 2021.

Two fabs just making Xe chiplets - seems kind of crazy, unless 7nm EUV is looking to be hard to bring up. And late 2021 - the pressure on Intel's process development team must be relentless.; they *have* to get it right.

 

[Dayman1225]

https://twitter.com/x/status/1152501210630557697

TGL U - BGA 1449

TGL Y - BGA 1598

Why would the -Y socket be bigger?

Also what is ICL - I, YN and UN?

 

[jpiniero]

Why would the -Y socket be bigger?

My guess is big.LITTLE.

Also what is ICL - I, YN and UN?

Guess on I is Industrial, maybe with the Thunderbolt fused off. The N parts my guess is no IGP.

 

[Dayman1225]

Guess on I is Industrial, maybe with the Thunderbolt fused off. The N parts my guess is no IGP.

Why bother with a new BGA socket with that when you can just fuse it off and sell it in the standard one?

 

[jpiniero]

Why bother with a new BGA socket with that when you can just fuse it off and sell it in the standard one?

I was thinking the same thing but maybe it's not a big deal to design a new package and you save a few bucks.

 

[IntelUser2000]

Why would the -Y socket be bigger?

The reasons aren't always profound. Sometimes, they leave some pins unused for headroom.

Also, Amber Lake 4+2 has less pins than 2+2.

Also what is ICL - I, YN and UN?

The -I does look industrial. There were Apollo Lake-I. As for socket change, the changes might be better suited for the target market.

The reduced pin count could also mean the -I and -YN/-UN parts have less I/O to the outside world.

 

[Dayman1225]

The -I does look industrial. There were Apollo Lake-I. As for socket change, the changes might be better suited for the target market.

Seems to be the case. Codename Keystone Lake was just found and it is KSL - I. It is aimed at Infotainment systems in cars like Apollo Lake/Gordon Peak version. The "I" could be infotainment or Industrial. In this case, industrial makes more sense with the existence of ICL and Apollo - I

https://twitter.com/x/status/1152706281452392448

 

[mikk]

As for UN there are some Icelake SKUs floating around with this branding, one is called i7-1068NG7. Because of the G7 it should come with graphics. The number is higher than the I7-1065G7, maybe this is a 25W SKU. We don't have the full lineup, the 4.1 Ghz SKU is still missing.

 

[ikjadoon]

As for UN there are some Icelake SKUs floating around with this branding, one is called i7-1068NG7. Because of the G7 it should come with graphics. The number is higher than the I7-1065G7, maybe this is a 25W SKU. We don't have the full lineup, the 4.1 Ghz SKU is still missing.

i7-1068NG7. What a name...

Did Intel get model number advice from HP? That Ice Lake x360's model number: Convertible 13-aw0xxx, following in HP's MD5 hash consumer model name tradition, e.g., the HP Pavilion 13-an0010nr.

That i7-1065G7 might be the "4.1 GHz" SKU: since Whiskey Lake, i7 advertised "Max Turbo Frequency" clocks are not Turbo Boost, but some wholly and utterly unrealistic "Thermal Velocity Boost" clocks. So, 3.9 GHz Turbo Boost & 4.1 GHz TVB.

I fully expect TVB shenanigans with all Intel 10nm and <10nm SKUs, not just i7/i9 SKUs.

The marketing spiel from Coffee Lake R / Whiskey Lake:

Intel® Thermal Velocity Boost

Intel® Thermal Velocity Boost (Intel® TVB) is a feature that opportunistically and automatically increases clock frequency above single-core and multi-core Intel® Turbo Boost Technology frequencies based on how much the processor is operating below its maximum temperature and whether turbo power budget is available. The frequency gain and duration is dependent on the workload, capabilities of the processor and the processor cooling solution.

(emphasis mine)