Intel Cannonlake SoC will have 4-core, 6-core and 8-core versions

[VirtualLarry]

Even with 10nm having a non-FinFET xtor design (that I assume will be more efficient) clock speeds across all eight cores might not be that high.

Intel dropping FinFET at 10nm? I thought that "FinFET was the future", at least until Intel made a switch in the materials actually used for their transistors. If true, does this mean that Intel is going to switch materials for 10nm? Because I don't believe that I've read that anywhere, other than some speculation as to when Intel might switch materials.

 

[cbn]

Intel dropping FinFET at 10nm? I thought that "FinFET was the future", at least until Intel made a switch in the materials actually used for their transistors. If true, does this mean that Intel is going to switch materials for 10nm? Because I don't believe that I've read that anywhere, other than some speculation as to when Intel might switch materials.

Here is where I found the info:

http://forums.anandtech.com/showpost.php?p=34198524&postcount=1

 

[Eug]

Quad core sounds interesting in an ultraportable laptop if they can keep the single-core (and dual-core) clockspeed and performance up, and higher than previous years' dual core designs.

However, if it isn't a design that can be run fanless, I'm not really interested.

I think I represent a good chunk of the consumer computing public these days. While fanless is not the normal requirement, those of us without heavy mobile computing needs are probably the majority these days. In this context, battery life is far more important than raw performance, esp. since it's difficult for us to make good use of those four cores simultaneously.

In fact, my primary laptop is a Core 2 Duo right now. I want a new laptop, but not really because of the performance. I want a new one for lighter weight, longer battery life, and a Retina screen. The main performance related thing I want is strong h.265 support going forward, but even that won't be an issue since 14 nm Kaby Lake gets robust h.265 support in hardware, so fanless dual-core designs would be OK for that too.

The question for me is just how much extra power utilization would four cores be in that context, and for how much money. If not much, then great, lay it on me. However if it significantly affects battery life and cost, I'm not sure I'd be interested vs. dual-core.

BTW, I was originally planning on buying Skylake Core M, but because of h.265, decided to wait for Kaby Lake. Now I'm wondering if I shouldn't just wait and see for Cannonlake.

 

[cbn]

Quad core sounds in an ultraportable laptop interesting if they can keep the single-core (and dual-core) clockspeed and performance up, and higher than previous years' dual core designs.

However, if it isn't a design that can be run fanless, I'm not really interested.

If they start off with octocore, then they have a better chance to get four good low power ones.

So maybe Core M fanless becomes achievable at 10nm with an octocore die, but only by using the choicest four cores.

Not sure what the downside (s) to this strategy would be? Obviously one downside is that it takes more wafers to yield X amount of mobile quad cores , but then maybe has figured out the extra waste is worth it due to the extra mark-up they can give the chips?

 

[Idontcare]

Here is where I found the info:

http://forums.anandtech.com/showpost.php?p=34198524&postcount=1

Intel's 10nm is finfet.

 

[Fjodor2001]

The main performance related thing I want is strong h.265 support going forward, but even that won't be an issue since 14 nm Kaby Lake gets robust h.265 support in hardware

Skylake already has HW accelerated HEVC/H.265 support, but only 8 bit. KabyLake will have 10 bit support too. Is that why you want to wait for KabyLake?

 

[ShintaiDK]

Skylake got hardware accelerated 10bit support. But only full hardware decode of 8bit.

 

[davygee]

Quad core sounds interesting in an ultraportable laptop if they can keep the single-core (and dual-core) clockspeed and performance up, and higher than previous years' dual core designs.

However, if it isn't a design that can be run fanless, I'm not really interested.

I think I represent a good chunk of the consumer computing public these days. While fanless is not the normal requirement, those of us without heavy mobile computing needs are probably the majority these days. In this context, battery life is far more important than raw performance, esp. since it's difficult for us to make good use of those four cores simultaneously.

In fact, my primary laptop is a Core 2 Duo right now. I want a new laptop, but not really because of the performance. I want a new one for lighter weight, longer battery life, and a Retina screen. The main performance related thing I want is strong h.265 support going forward, but even that won't be an issue since 14 nm Kaby Lake gets robust h.265 support in hardware, so fanless dual-core designs would be OK for that too.

The question for me is just how much extra power utilization would four cores be in that context, and for how much money. If not much, then great, lay it on me. However if it significantly affects battery life and cost, I'm not sure I'd be interested vs. dual-core.

BTW, I was originally planning on buying Skylake Core M, but because of h.265, decided to wait for Kaby Lake. Now I'm wondering if I shouldn't just wait and see for Cannonlake.

But we have to wait until the 2nd half of 2017 for Cannonlake and I agree that fanless in Laptops is the way things should be going.

 

[VirtualLarry]

Intel's 10nm is finfet.

Thanks for the confirmation, IDC. I was really scratching my head over that.

 

[Eug]

Skylake already has HW accelerated HEVC/H.265 support, but only 8 bit. KabyLake will have 10 bit support too. Is that why you want to wait for KabyLake?

Yes. I tend to keep my laptops a very long time, so I buy my hardware accordingly. I choose the form factor first, and then wait for a CPU/GPU that suits my computing needs (or projected future needs) to fit that form factor. My Core 2 Duo lasted this long partially because I chose to wait for full hardware h.264 decode support before buying.

 

[Sweepr]

2. iGPU die size will increase

Source? GT2 should be pretty small @ 10nm.

4. they may need to integrate the eDRAM (TSV)

In non-Iris GT2 parts? I don't think so.

With all those above, the die size will be close to current Skylake but at more expensive 10nm process. Adding 4 more cores will make it even more expensive (bigger die, lower yields etc).

Dont believe we will see 8-cores for the client at 10nm.

Cannonlake cores should be ridiculously small. I don't think there will be a big impact by adding 2 or 4.

What I do agree with you is, AMD better hope this is fake. Otherwise Intel will cover any possible niche for 6C/12-8C/16T mainstream client chips.

Quad core sounds interesting in an ultraportable laptop if they can keep the single-core (and dual-core) clockspeed and performance up, and higher than previous years' dual core designs.

+1
Skylake will bring the first 25W quad-cores. 4C/8T 'U' processors are definitely a possibility for 10nm parts. I'd love a more agressive roadmap now that Apple is shaking things up in mobile.

I agree with Ashraf (the author of the article we are discussing) in that the argument exists to expand core count at 10nm because of how tiny a quad-core die will be at that point. And doubling CPU cores does not double the silicon die area, probably will only be a 15% adder to go from 4 core to 8 core at 10nm.

Fingers crossed. Adding up to that, Xeon D-1541 packs 8 Broadwell cores @ 2.1-2.7GHz (no iGPU) at 45W.
Reasonable clocks + GT2 iGPU should be possible @ 95W at 10nm.

 

[dark zero]

But we have to wait until the 2nd half of 2017 for Cannonlake and I agree that fanless in Laptops is the way things should be going.

SCREW FANLESS! Heatsinkless is the future!

 

[VirtualLarry]

SCREW FANLESS! Heatsinkless is the future!

Now that's just... kind of a stupid comment. Smaller chips, with more dense lithography, means that heat density of CPUs is going UP, not down. It's basic physics, that you're going to need a heatsink, even moreso.

Edit: Unless you're speaking of graphene or carbon nanotube-based CPUs, where the chip can act as its own heatsink, somewhat. But it would still need some mass attached, as a sort of "thermal reservoir".

 

[Fjodor2001]

Yes, and you lose a lot of features. Including memory above 2133 and such.

So its pretty pointless for your "extreme gamer" that want 6/8 cores. Now we are over in the budget/mainstream user.

Non-Z170 chipsets are totally fine unless to intend to OC. Not nearly all gamers OC you know, not even "extreme" gamers. They'd be much better served by a doubling of core count.

 

[ShintaiDK]

Non-Z170 chipsets are totally fine unless to intend to OC. Not nearly all gamers OC you know, not even "extreme" gamers. They'd be much better served by a doubling of core count.

On DDR4 2133? And what about the rest of the features?

Its funny how your definition of extreme gamer is a core count thing only

How many games support more than 4 cores? Even better, above the benefit of HT with 4 cores.

 

[Fjodor2001]

And it shows what? People dropping quadcores for dualcores as gamers?

Since you say the Average Joe does not need more than 2 cores for most scenarios, doesn't that mean Intel should kill off quad cores on the mainstream platform, and move all SKUs with more than two cores to the HEDT platform?

 

[AtenRa]

Source? GT2 should be pretty small @ 10nm.



In non-Iris GT2 parts? I don't think so.



Cannonlake cores should be ridiculously small. I don't think there will be a big impact by adding 2 or 4.

Haswell 4+2 at 22nm process is 170mm2

Skylake 4+2 at 14nm process is 122mm2

At 10nm it should be close to half, that is 60mm2.

Now, add the larger iGPU + the PCH and we are looking at close to 80-90mm2

Skylake single core + L3 is roughly 10,5mm2.
4x cores + L3 = 22mm2 at 14nm.

That means close to 20mm2 at 10nm for an additional 4x cores.

So at the end we are looking at 100-110mm2 for a 8+2 Cannonlake.

Well, i dont believe Intel will bring something like that to the client market, especially not in Laptop space.

Im expecting Intel to double the iGPU with Cannonlake, especially for the mobile market.
So they will devote more and more die space for larger iGPUs, like they do 4-5 years now with every new process. Cannonlake will be their first real SoC specifically oriented for the Mobile market. You dont need 8 cores for the mobile but you need a faster iGPU.

What I do agree with you is, AMD better hope this is fake. Otherwise Intel will cover any possible niche for 6C/12-8C/16T mainstream client chips.

I dont expect AMD to release any 6C/12T-8C/16T ZEN for the mainstream client market. They will have the Quad core ZEN APUs for that.

 

[Fjodor2001]

On DDR4 2133? And what about the rest of the features?

Its funny how your definition of extreme gamer is a core count thing only

How many games support more than 4 cores? Even better, above the benefit of HT with 4 cores.

What chipset "features" are you talking about that affect performance? More USB ports and similar is N/A. The only thing I can think of is more PCI lanes, but if you intend to use a GFX card with 16 PCI lanes you'll be fine.

 

[ShintaiDK]

Since you say the Average Joe does not need more than 2 cores for most scenarios, doesn't that mean Intel should kill off quad cores on the mainstream platform, and move all SKUs with more than two cores to the HEDT platform?

Did you provide your documentation yet? Oh right, your personal speculation.

 

[ShintaiDK]

What chipset "features" are you talking about that affect performance? More USB ports and similar is N/A. The only thing I can think of is more PCI lanes, but if intend to use a GFX card with 16 PCI lanes you'll be fine.

Look back in the thread and compare the chipset features. And remember memory locked at 2133 as well.

Why dont you show the countless examples of extreme gamers needing more cores but none of the Z170 features.

 

[Fjodor2001]

Did you provide your documentation yet? Oh right, your personal speculation.

What documentation? It's you that previously (in other threads) has said that the Average Joe only needs two cores. So are you retracting from that statement now?

 

[Fjodor2001]

Look back in the thread and compare the chipset features. And remember memory locked at 2133 as well.

I've compared the chipset features and apart from lack of OC I did not find any differences affecting performance other than the PCI lanes and locked 2133 memory. The former is N/A if you use a 16 lanes PCI card, the latter does not affect performance more than a few percent.

So what performance affecting chipset features are you talking about that the non-Z170 chipsets lack?

 

[ShintaiDK]

What documentation? It's you that previously (in other threads) has said that the Average Joe only needs two cores. So are you retracting from that statement now?

No I didn't.

Not so tiny. See this:
http://www.bbc.com/news/technology-34117120

"Intel claims something like 10% of the PC market is extreme gamers."

And that's just the "extreme" gamers, whatever that means. But it is safe to assume they are enthusiasts with really high performance hardware, that likely will benefit from and desire more cores in the years ahead. Add to that other users that use their desktop for video editing, compiling code, etc.

Remember this? You linked extreme gamers and more cores together. Yet you couldn't provide a single fact.

Also you even noted them with really high performance hardware. Yet now you try and showhorn them down into cheap limited builds. Because else you got caught by the HEDT platform in price.

 

[ShintaiDK]

I've compared the chipset features and apart from lack of OC I did not find any differences affecting performance other than the PCI lanes and locked 2133 memory. The former is N/A if you use a 16 lanes PCI card, the latter does not affect performance more than e few percent.

So what performance affecting chipset features are you talking about that the non-Z170 chipsets lack?

H170 lacks 4 lanes, no memory above 2133, less M.2 slots, less USB3 ports, no PCIe splitting for SLI/CF. And with this you save a whooping 20$ max over Z170. All for your extreme gamer who is willing to pay ~400$ for a 6 core. Not to mention how low clocked said 6 core CPU would be at 95W for the extreme gamer of yours.

Want H110 and B150 too? Since you really depend on these for any real price benefit.

 

[Sabrewings]

You had 12t/6c options when you bought the 4790K though. Why didn't you buy one. :biggrin:

I see what you're doing there.

In my case Haswell-E did exist when I bought my 4790k. Four things: cost of the chip, cost of the motherboard, cost of the RAM, and the clock speeds are lower. Until games are more threaded than they are now, the extra cores don't matter over the existing four as much as the speed of those cores. This is a primarily gaming machine, so game considerations won.

However, all else being equal (DDR3's cost, roughly equivalent motherboard cost) I would've been all over a 4.0GHz 6C/12T Haswell with no IGP for $500 or less. Alas, no such beast exists. The IGP might prove useful with DX12, but for the moment I have it completely disabled.