Intel's Future Chips: News, Rumours & Reviews

[goldstone77]

AVX-512 not a great feature to add for consumer chips... Read the discussion!

Ashraf Eassa@TMFChipFool

More Ashraf Eassa Retweeted Oscar Barenys

I hope not. AVX-512 is cancer for consumer chips.

https://twitter.com/TMFChipFool/status/987680634935107586

 

[juanrga]

I have several skylake servers at work. All of them include AVX512.

That is high-end Skylake server, which is a completely different architecture than Skylake client and low-end Skylake server. Some differences between both families

Skylake-SP/X/W/DE: 14nm+, 512bit SIMD units with AVX512, Mesh interconnect, 1MB L2, 2666MHz IMC,..

Skylake-Y/U/H/S/DT: 14nm, 256bit SIMD units with AVX256, Ring interconnect, 256Kb L2, 2400MHz IMC,..

 

[aldaia]

I have several skylake servers at work. All of them include AVX512.

That is high-end Skylake server, which is a completely different architecture than Skylake client and low-end Skylake server. Some differences between both families

Skylake-SP/X/W/DE: 14nm+, 512bit SIMD units with AVX512, Mesh interconnect, 1MB L2, 2666MHz IMC,..

Skylake-Y/U/H/S/DT: 14nm, 256bit SIMD units with AVX256, Ring interconnect, 256Kb L2, 2400MHz IMC,..

Intel considers them the same microarchitecture. Intel refers to them as configurations, not as different microarchitectures. The Skylake core is a single development project, making up a master superset core. The project resulted in two derivatives: one for servers and one for clients. All mainstream models use the client core configuration. Server models use the new configuration.

In the high-end and highest performance Xeons, Intel added 768 KiB L2 cache and the second AVX-512 VPU externally to the core.

 

[-Fran-]

I have to say, this is the first time I see that information, aldaia. For me they have always been different buggers.

Welp, you learn something new everyday it seems!

Cheers!

 

[juanrga]

I have several skylake servers at work. All of them include AVX512.

That is high-end Skylake server, which is a completely different architecture than Skylake client and low-end Skylake server. Some differences between both families

Skylake-SP/X/W/DE: 14nm+, 512bit SIMD units with AVX512, Mesh interconnect, 1MB L2, 2666MHz IMC,..

Skylake-Y/U/H/S/DT: 14nm, 256bit SIMD units with AVX256, Ring interconnect, 256Kb L2, 2400MHz IMC,..

Intel considers them the same microarchitecture. Intel refers to them as configurations, not as different microarchitectures. The Skylake core is a single development project, making up a master superset core. The project resulted in two derivatives: one for servers and one for clients. All mainstream models use the client core configuration. Server models use the new configuration.

In the high-end and highest performance Xeons, Intel added 768 KiB L2 cache and the second AVX-512 VPU externally to the core.

They are two different microarchitectures with two different cores. Your images are from wikichip. Precisely Wikichip has two entries, one for each microarchitecture

https://en.wikichip.org/wiki/intel/microarchitectures/skylake_%28client%29
https://en.wikichip.org/wiki/intel/microarchitectures/skylake_%28server%29

There are two different cores, but instead designing both cores from scratch in an independent way, engineers first designed a common baseline (the master core) and then used that baseline to design each final core.

The extra cache and the extra AVX unit aren't external to the 'server' core, but are part of the core; the yellow boxed image on the left is the 'server' core

Or if you want them by separate: 'Client' core

'Server' core

I am using quotes, because the distinction between "client" and "server" cores that wikichip does is misleading. I prefer small and large core, because the small core is used in both clients (e.g. i7-6700K) and servers (e.g. Xeon E3 1280 v5), whereas the big core is used in both clients (e.g. i9-7900X) and servers (e.g. Xeon Gold 6128).

 

[goldstone77]

https://twitter.com/TMFChipFool/status/988103788979544064

Looks like the 10nm might be the silver bullet that sinks Intel. What was once their greatest advantage, manufacturing, is becoming their greatest liability.

 

[juanrga]

What Will Intel Corp.'s Top Gaming Processor in 2019 Look Like?

https://www.fool.com/investing/2018/04/23/what-will-intel-corps-top-gaming-processor-in-2019.aspx

 

[goldstone77]

What Will Intel Corp.'s Top Gaming Processor in 2019 Look Like?

https://www.fool.com/investing/2018/04/23/what-will-intel-corps-top-gaming-processor-in-2019.aspx

Good article. We all knew that moar cores was coming as process technology slows down.

 

[8350rocks]

What Will Intel Corp.'s Top Gaming Processor in 2019 Look Like?

https://www.fool.com/investing/2018/04/23/what-will-intel-corps-top-gaming-processor-in-2019.aspx

Good article. We all knew that moar cores was coming as process technology slows down.

But, but...something something more cores takes years...

 

[juanrga]

What Will Intel Corp.'s Top Gaming Processor in 2019 Look Like?

https://www.fool.com/investing/2018/04/23/what-will-intel-corps-top-gaming-processor-in-2019.aspx

Good article. We all knew that moar cores was coming as process technology slows down.

But, but...something something more cores takes years...

8-cores Icelake was planned back for release in 2015. I am rather sure they had time enough to add extra cores for 2019 if they want.

 

[8350rocks]

Spoiler

What Will Intel Corp.'s Top Gaming Processor in 2019 Look Like?

https://www.fool.com/investing/2018/04/23/what-will-intel-corps-top-gaming-processor-in-2019.aspx

Good article. We all knew that moar cores was coming as process technology slows down.

But, but...something something more cores takes years...

8-cores Icelake was planned back for release in 2015. I am rather sure they had time enough to add extra cores for 2019 if they want.

Sure, on the extreme platform, it absolutely was.

 

[-Fran-]

I really hope Intel drops a 10-core bomb in the mainstream platform. That would be sooooooo damn sweet.

Cheers!

 

[juanrga]

Spoiler

What Will Intel Corp.'s Top Gaming Processor in 2019 Look Like?

https://www.fool.com/investing/2018/04/23/what-will-intel-corps-top-gaming-processor-in-2019.aspx

Good article. We all knew that moar cores was coming as process technology slows down.

But, but...something something more cores takes years...

8-cores Icelake was planned back for release in 2015. I am rather sure they had time enough to add extra cores for 2019 if they want.

Sure, on the extreme platform, it absolutely was.

8-core for the mainstream platform. The extreme platform was 16-core.

 

[goldstone77]

If, all these cores were planned for mainstream than one can reasonably assume that Intel has known about their manufacturing problems for a few years now.

 

[Gon Freecss]

If, all these cores were planned for mainstream than one can reasonably assume that Intel has known about their manufacturing problems for a few years now.

You make no sense here. Icelake will be on 10nm+. Also, Cannonlake was slated for a 2016 release with an 8 core model for the mainstream. Icelake would've followed it in 2017 with a presumably 8 core mainstream too.

They were increasing core counts in 2016+.

 

[goldstone77]

If, all these cores were planned for mainstream than one can reasonably assume that Intel has known about their manufacturing problems for a few years now.

You make no sense here. Icelake will be on 10nm+. Also, Cannonlake was slated for a 2016 release with an 8 core model for the mainstream. Icelake would've followed it in 2017 with a presumably 8 core mainstream too.

They were increasing core counts in 2016+.

It makes perfect sense! Adding cores is an easy way to improve "marketable" performance if you can not improve single thread performance.(ref. 1) Planning out architecture in advance is only half the story, because the reason for the addition of cores is the relevant fact. (theory)Failing to create new improved process technology is the other half of the driving force behind the increase in mainstream cores. And that is why I say (if) all these cores were planned for mainstream in advance than they known about their manufacturing problems for a few years now! Like I said, that makes perfect sense!


Ref. 1: What Will Intel Corp.'s Top Gaming Processor in 2019 Look Like?
Intel will probably go with more cores.
Ashraf Eassa (TMFChipFool) Apr 23, 2018 at 10:45AM
Easy and marketable
Driving up-sell

 

[Gon Freecss]

If, all these cores were planned for mainstream than one can reasonably assume that Intel has known about their manufacturing problems for a few years now.

You make no sense here. Icelake will be on 10nm+. Also, Cannonlake was slated for a 2016 release with an 8 core model for the mainstream. Icelake would've followed it in 2017 with a presumably 8 core mainstream too.

They were increasing core counts in 2016+.

It makes perfect sense! Adding cores is an easy way to improve "marketable" performance if you can not improve single thread performance.(ref. 1) Planning out architecture in advance is only half the story, because the reason for the addition of cores is the relevant fact. (theory)Failing to create new improved process technology is the other half of the driving force behind the increase in mainstream cores. And that is why I say (if) all these cores were planned for mainstream in advance than they known about their manufacturing problems for a few years now! Like I said, that makes perfect sense!


Ref. 1: What Will Intel Corp.'s Top Gaming Processor in 2019 Look Like?
Intel will probably go with more cores.
Ashraf Eassa (TMFChipFool) Apr 23, 2018 at 10:45AM
Easy and marketable
Driving up-sell

No, listen. Cannonlake and Icelake are on 10nm and 10nm+ respectively. Intel decided they'd up the core count with Cannonlake in 2016 from 4 cores (Skylake) to 8 cores (Cannonlake) before process problems hit them. When they knew they're gonna be with 14nm for the long haul, they made plan B. Kaby Lake, and Coffee Lake. They decided to up the core count to 6 cores with Coffee Lake for marketability as you said (they only pushed it earlier to combat AMD). Now, since Intel doesn't have the indisputable best mainstream processor out in the market yet, they've come up with a plan B.5, which is the 8 core Coffee Lake. That would make Intel take the indisputable best mainstream processor crown with a vice grip. This is where marketability comes in, not with Cannonlake and Icelake.

As for Icelake, I expect it to be 10C+ to combat AMD and to have better marketability. See where I'm coming from here?

 

[goldstone77]

No, listen. Cannonlake and Icelake are on 10nm and 10nm+ respectively. Intel decided they'd up the core count with Cannonlake in 2016 from 4 cores (Skylake) to 8 cores (Cannonlake) before process problems hit them. When they knew they're gonna be with 14nm for the long haul, they made plan B. Kaby Lake, and Coffee Lake. They decided to up the core count to 6 cores with Coffee Lake for marketability as you said (they only pushed it earlier to combat AMD). Now, since Intel doesn't have the indisputable best mainstream processor out in the market yet, they've come up with a plan B.5, which is the 8 core Coffee Lake. That would make Intel take the indisputable best mainstream processor crown with a vice grip. This is where marketability comes in, not with Cannonlake and Icelake.

As for Icelake, I expect it to be 10C+ to combat AMD and to have better marketability. See where I'm coming from here?

I do see what you are saying, and I agree with everything your saying. Added ~Xeon W processors to i9 segment to beat ThreadRipper. This also showed how fast they are able to bring a design to market, even if it was mostly a copy of the Xeon W design! They advanced timelines and created Coffee Lake to maintain a lead over AMD. I believe all that. But, I also think the reasoning behind the increase in cores on advanced architectures shows they knew many years ago about problems with their manufacturing process. Hence, the Cannon lake and Ice lake designs including 8 core main stream processors, and now rumored Coffee Lake 8 core. I say this for the reasons I mentioned above. Intel is being extremely secretive, and not really sharing information like Intel use too!

 

[gamerk316]

I really hope Intel drops a 10-core bomb in the mainstream platform. That would be sooooooo damn sweet.

Cheers!

In before we all complain it performs the same in games as the Quad core part! Because you know that is EXACTLY what people will complain about.

 

[-Fran-]

I really hope Intel drops a 10-core bomb in the mainstream platform. That would be sooooooo damn sweet.

Cheers!

In before we all complain it performs the same in games as the Quad core part! Because you know that is EXACTLY what people will complain about.

Not necessarily. Being a 10-core part does not mean the Turbo implementation (and speeds) will be the same. Look at TPB2 (Turbo Precision Boost 2... I think) and how it manages the cores speed. It works incredibly well.

I do think your point is not moot at all, but having a 6-8-10-12 core part being developed, I'd hope they also nail the Turbo for the right speeds they need given the workload types.

Cheers!

 

[juanrga]

Cannonlake and Icelake are on 10nm and 10nm+ respectively. Intel decided they'd up the core count with Cannonlake in 2016 from 4 cores (Skylake) to 8 cores (Cannonlake) before process problems hit them. When they knew they're gonna be with 14nm for the long haul, they made plan B. Kaby Lake, and Coffee Lake. They decided to up the core count to 6 cores with Coffee Lake for marketability as you said (they only pushed it earlier to combat AMD). Now, since Intel doesn't have the indisputable best mainstream processor out in the market yet, they've come up with a plan B.5, which is the 8 core Coffee Lake. That would make Intel take the indisputable best mainstream processor crown with a vice grip. This is where marketability comes in, not with Cannonlake and Icelake.

Exactly. Back in 2015 the plan was

i3 = 4 cores
i5 = 6 cores
i7 = 8 cores

before the 10nm node was delayed.

If a 10-core is coming now with Icelake, then either they update the i3 to six cores or launch an i9 line for mainstream.

 

[logainofhades]

If, all these cores were planned for mainstream than one can reasonably assume that Intel has known about their manufacturing problems for a few years now.

Or they really didn't have a reason to do it, due to lack of competition. Since Ryzen's release, Intel has had a fire lit under them, that we haven't seen in years. We kept getting minor improvements, but nothing substantial.

 

[YoAndy]

Skylake-X does support Intel's AES New Instructions, Execute Disable Bit SSE4.1, Intel® SSE4.2, Intel® AVX2 and Intel® AVX-512 instructions. https://ark.intel.com/products/123589/Intel-Core-i7-7800X-X-series-Processor-8_25M-Cache-up-to-4_00-GHz?q=7800x

 

[goldstone77]

If, all these cores were planned for mainstream than one can reasonably assume that Intel has known about their manufacturing problems for a few years now.

Or they really didn't have a reason to do it, due to lack of competition. Since Ryzen's release, Intel has had a fire lit under them, that we haven't seen in years. We kept getting minor improvements, but nothing substantial.

Lack of competition would give reason for them to innovate less, like they did and you mention, but that means they don't have to add more core either. What we do know is that their are reports of problems with Intel's manufacturing process. Other foundries have moved past Intel now, which is something Intel has led for ~20 years. ~3 years of delays is something they had to have known about for years. They would plan for that accordingly. AMD did the moar cores thing, and Intel has planned for moar cores many generations in advance. So, they wouldn't just let their manufacturing lead slip, because of lack of competition and lose all their foundry business in the process. The smarter play would be to continue to innovate the manufacturing process like they have done for 20 years to get their improvements. And not add more cores, which means less CPU's per wafer and less profit. So, that's why I say they would have just maintained the manufacturing lead if that was possible to maintain dominance, like they have for ~20 years. Just my 2 cents

 

[juanrga]

Bloomfield: 4 cores
Westmere-EP: 6 cores
SandyBridge-E: 6 cores
IvyBridge-E: 6 cores
Haswell-E: 8 cores
Broadwell-E: 10 cores

It is not like if Intel has not added moar cores systematically since 45nm!!!!