Discussion Intel current and future Lakes & Rapids thread

[ondma]

Because it's a reflection of how bad the 10 nm yield continues to be. That affects Rocket Lake if laptop demand continues to be strong, and OEMs decide they want more 14 nm if they can't get 10 nm (or AMD). Right now it does look like there should be room for Rocket Lake.

Would not Tiger Lake selling better than expected indicate 10 nm yields are improving?

 

[jpiniero]

Would not Tiger Lake selling better than expected indicate 10 nm yields are improving?

Possibly. We'll find out during earnings I guess. Of course everything is relative.

 

[dmens]

Would not Tiger Lake selling better than expected indicate 10 nm yields are improving?

No. It is an absolutely meaningless statement. Pure bloviation.

 

[Zucker2k]

So TSMC 7nm (Renoir, Zen 2, Zen 3), and Intel 10nm (IceLake, IceLake Server, and TigerLake) are all having little to no effect on Intel 14nm availability? Please.

 

[DrMrLordX]

So TSMC 7nm (Renoir, Zen 2, Zen 3), and Intel 10nm (IceLake, IceLake Server, and TigerLake) are all having little to no effect on Intel 14nm availability? Please.

TSMC 7nm? Certainly, at least in instances where AMD taking market share causes Intel to produce fewer Xeon dice.

 

[dacostafilipe]

So TSMC 7nm (Renoir, Zen 2, Zen 3), and Intel 10nm (IceLake, IceLake Server, and TigerLake) are all having little to no effect on Intel 14nm availability? Please.

I don't know, but I feel that both nodes had little impact compared to the end of the surge of those companies that "needed" new 14nm CPUs because of the mitigation.

I had clients wanting those new 14nm CPUs, even in their private servers with no shared resources, where one could disable SMT and/or where performance wasn't a major factor. The datacenter sold them the upgrade, because it's more "secure".

 

[Zucker2k]

TSMC 7nm? Certainly, at least in instances where AMD taking market share causes Intel to produce fewer Xeon dice.

Of course, I'm talking about market share. You didn't think about that? Renoir, Zen 2, and Zen 3 derivatives (including server) have all combined to make a dent in Intel's market share at mobile, client, and enterprise levels.

 

[Adonisds]

Intel Lynnfield had 296 mm². Why is comet lake and rocket lake now considered too big?

 

[Hulk]

Intel Lynnfield had 296 mm². Why is comet lake and rocket lake now considered too big?

I'm curious, how does Zen 2/3 compare to Comet Lake in a per core die area basis?
How much silicon for 8 core Zen 2/3 vs 8 core Comet Lake? Perhaps "too big" or "too small" depends on the size of the competition?
Also I would think yields have something to do with it. If defects are high then large dies are going to be a killer on yields.

I don't know. Just thinking out loud here.

I'm hopeful for Rocket Lake but I don't know, Zen 3 seems so damn strong. I've never purchased/built an AMD system and I've been building systems since the late 1980's. If Zen 3 prices weren't crazy high right now I'd probably go for it. But I'm in no rush and am going to wait for things to cool down and see what Rocket Lake has to offer. Either way I'm going to get a massive increase in compute from my 4770k.

 

[DrMrLordX]

You didn't think about that?

I didn't mention TSMC explicitly, but yes, that was implicit to what I was saying about Intel's main source of additional available 14nm capacity coming from reduced Xeon sales . . .

 

[SAAA]

Intel Lynnfield had 296 mm². Why is comet lake and rocket lake now considered too big?

They have been selling smaller and smaller dices for years, with the underlying technology getting more and more expensive despite of Moore's law that was key in keeping the same profit.
With 14 nm they wouldn't be able to sell 300 mm^2 dices to consumers for 100 to 300 $ with margins anywhere near Lynnfield.
On the other hand Lynnfield on 14 nm would probably be in the tens of mm^2 range if ported, so you are still getting more compute power in 2020 vs 2010 for the same money.

Another issue is power: a 300 mm^2, 16 core Comet Lake would burn a lot of power, and we already are at the limit with 10 cores...

 

[jpiniero]

They have been selling smaller and smaller dices for years, with the underlying technology getting more and more expensive despite of Moore's law that was key in keeping the same profit.
With 14 nm they wouldn't be able to sell 300 mm^2 dices to consumers for 100 to 300 $ with margins anywhere near Lynnfield.

14 nm should be pretty damn cheap by now. It's more of a question of whether there are better products to use the capacity on.

 

[shady28]

They have been selling smaller and smaller dices for years, with the underlying technology getting more and more expensive despite of Moore's law that was key in keeping the same profit.
With 14 nm they wouldn't be able to sell 300 mm^2 dices to consumers for 100 to 300 $ with margins anywhere near Lynnfield.
On the other hand Lynnfield on 14 nm would probably be in the tens of mm^2 range if ported, so you are still getting more compute power in 2020 vs 2010 for the same money.

Another issue is power: a 300 mm^2, 16 core Comet Lake would burn a lot of power, and we already are at the limit with 10 cores...

I don't think Intel is interested in the market you guys are talking about anyway. Very, very, very few home PCs have 16 cores.

If you look at Steam HW survey for example - and I would consider this weighted heavily towards enthusiasts and not typical home users, so more cores than the overall market - the total number of systems with more than 8 cores is right around 1%.

This means 99%+ of users is 8 cores or less.

If Intel really wanted to target the halo group of 12 / 16 core users, that tiny speck of a market, the solution would be easy.

Drop the iGPU which takes up 20-25% of the die space (and which AMD doesn't have) and fill it with 4 more cores + more cache. Bang, you have a 14 core / 28 thread 14nm CPU on a die about the same size as a 10900/10900K which is just under 200mm2.

I wouldn't be horribly surprised to see something like that with RKL, since despite its size that particular market is quite vocal, and halo products like those are all about the marketing.

 

[jpiniero]

Drop the iGPU which takes up 20-25% of the die space (and which AMD doesn't have) and fill it with 4 more cores + more cache.

Uh, Rocket Lake is 8 max.

 

[shady28]

Uh, Rocket Lake is 8 max.

Has a complete SKU listing been leaked? I've only seen leaks on 8 core parts, does this mean there are no 6 core parts?

I'm not saying it's definitive, I'm just saying there is nothing definitive and I wouldn't be surprised if the marketing guys didn't get a say-so on how many cores to sell. But if there is one, it is just like AMDs 12 / 16 core chops - for edge case halo products that 99% of people aren't going to buy.

 

[SAAA]

14 nm should be pretty damn cheap by now. It's more of a question of whether there are better products to use the capacity on.

That last part definitely, like server parts with several times the profit.
As for how cheap is 14 nm I think it isn't still cheaper than a 32 nm node from a purely technical stand point, OK it's their best process ever, but the wafers must be somewhat more pricey than a simpler node, when all else is equal.
Of course you can't keep using an older node too long as 14 nm shows because you run into other issues, like power, insufficient transistor density and so on to sell newer parts.

 

[jpiniero]

Has a complete SKU listing been leaked? I've only seen leaks on 8 core parts, does this mean there are no 6 core parts?

There will be 6 core parts for sure. However there is no smaller die so the lower end 11th Gen will still be Comet Lake.

 

[Dayman1225]

So Sapphire Rapids is pretty huge.

LGA 4677?

I think many of you have seen the LGA4189 Ice Lake-SP picture posted by my friend, but here's an LGA4677 Sapphire Rapids(SP?) CPU. Can't delid now cause it's not mine.

forums.servethehome.com

Edit: apparently it could be Cooper Lake

https://twitter.com/x/status/1337242960291893248

 

[Zucker2k]

View attachment 35467
So Sapphire Rapids is pretty huge.

LGA 4677?

I think many of you have seen the LGA4189 Ice Lake-SP picture posted by my friend, but here's an LGA4677 Sapphire Rapids(SP?) CPU. Can't delid now cause it's not mine.

forums.servethehome.com

Edit: apparently it could be Cooper Lake

https://twitter.com/x/status/1337242960291893248

What in God's holy name is that?!

 

[Edrick]

What in God's holy name is that?!

Looks like it could be Cascade Lake AP (2 dies). Copper Lake only had 1 die.

*Edit*

No, it clearly looks like socket 4677. https://wccftech.com/intel-xeon-sapphire-rapids-granite-rapids-cpus-lga-4677-compatible-ddr5-pcie-5/

 

[Spartak]

There was a huge run on lappies earlier this year. In August, it was getting hard to find laptops anywhere, at all.

I know it's just anecdotal evidence, but as a small engineering office we also moved all our personnel to laptops in the summer. It was just two additional laptops since most already moved to laptops but that decision really simplified our IT and support structure.

 

[DrMrLordX]

I know it's just anecdotal evidence, but as a small engineering office we also moved all our personnel to laptops in the summer. It was just two additional laptops since most already moved to laptops but that decision really simplified our IT and support structure.

You weren't alone in that.

 

[jpiniero]

[Intel-gfx] [PATCH 22/22] drm/i915/adl_s: Update memory bandwidth parameters

Intel added Alder Lake-S IGP code to the Linux kernel back in early December. What caught my eye is this, which suggests that the stock DDR5 speed is probably 6000.

(For comparison, the deprogbwlimit value on Rocket Lake is 20, on Alder Lake-S it's 38, Tiger Lake is 34)

 

[Mopetar]

I don't think Intel is interested in the market you guys are talking about anyway. Very, very, very few home PCs have 16 cores.

It doesn't matter that there aren't a lot of consumers in the market for this because the few that are will spend a lot of money which gives Intel good margins. Apple sells under 10% of the worlds PCs and only around 15% of the world's smartphones, yet they make more profit than everyone else because they sell to the part of the market where the margins are the highest.

 

[dmens]

I don't think Intel is interested in the market you guys are talking about anyway. Very, very, very few home PCs have 16 cores.

If you look at Steam HW survey for example - and I would consider this weighted heavily towards enthusiasts and not typical home users, so more cores than the overall market - the total number of systems with more than 8 cores is right around 1%.

This means 99%+ of users is 8 cores or less.

If Intel really wanted to target the halo group of 12 / 16 core users, that tiny speck of a market, the solution would be easy.

Drop the iGPU which takes up 20-25% of the die space (and which AMD doesn't have) and fill it with 4 more cores + more cache. Bang, you have a 14 core / 28 thread 14nm CPU on a die about the same size as a 10900/10900K which is just under 200mm2.

I wouldn't be horribly surprised to see something like that with RKL, since despite its size that particular market is quite vocal, and halo products like those are all about the marketing.

Hahahaha, 14 cores, burning 350 watts? 8 cores of Rocketlake is already spec'ed to max out at 250W.

Nobody will buy that trash when AMD's offering from two years ago was already superior in every aspect. There is literally no configuration of RKL that is technically competitive. Price cutting is the only viable play.