Coffee Lake Coming After Kaby Lake with 6 Cores

[AllBodies]

So apparently Intel is going to have 4 products on 14nm in the end, not just 3.

Coffee Lake is reportedly the same arch as Cannon Lake but slightly early, and on 14nm instead. And will go up to 6 cores on desktop and large laptop!

They must be worried about Zen!

Also, importantly, the Z270 motherboard series was said to support up to 10nm Cannon Lake. So it must be going to support 14nm Cannon Lake (Coffee Lake) too. What's that? 3 years of chipset support from Intel?! Hell is freezing over folks.

Article: http://wccftech.com/intel-roadmap-kaby-lake-coffee-lake-cannonlake-leak/


Picture of roadmap with 6 cores stated: http://cdn.wccftech.com/wp-content/uploads/2016/09/Intel-2016-2018-Roadmap.jpg

 

[AllBodies]

I am just wondering how the 6 cores coffee lake would impact the HEDT 6 cores. Coffee Lake 6 cores is bound to be faster than Skylake-X 6 cores.

Well yeah, it should be a nice little chunk faster in theory.

The IPC difference should be along the lines of Haswell compared to Sandy Bridge. Where Kaby Lake is Ivy Bridge. Or in other words, around 10% higher IPC.

Maybe it'll mean Skylake-X will be a minimum of 8 cores? And at least the Chipset will have caught up.

If it was 6-core on Z270 vs slower 6-core on X99, the HEDT platform would get decimated.

 

[AllBodies]

Hmmm, makes me wonder if Intel will abolish K unlocked CPU's for CL... Skylake-X will no doubt have larger cache levels but CL could muddy the waters between the mainstream and HEDT platforms.

Don't give them ideas!

 

[AllBodies]

More cache more cores, I predict a drop in clock rates.. To the point the new CPU's are only 10% faster than the previous gen for CPU's in the same price bracket

Perhaps, but their 14nm+ process seems to be pretty great.

The 7700k is going to have 4.5 GHz boost at stock, and the process will be even more mature when 6 core Coffee Lake comes.

 

[AllBodies]

That "roadmap" looks like it was made in paint.

>wccf

Why? It has been proven time and again they do nothing but make up fake bs for clicks. Anyone still posting them is doing everyone else a major dis-service.

The original source came from Overclock3D or something. I couldn't re-find the original, so just posted WCCF. But they didn't make the slide.

 

[AllBodies]

Indeed. For example, I do some development and database work. My FX-8350 with its 8 real cores is better than an i5 Skylake and competitive (though not equal) with some i7 Skylakes. And to be honest, I'd have to spend a lot more on a graphics card (I have a 480) before it was the bottleneck in most games as well.

It's not really quite 8 cores.

It's 4 pairs of 2 cores, each core having less resources to itself than a 'true' single core.

So it's more like 6-ish cores, with very low IPC but high clockspeed.

 

[AllBodies]

To be fair it's 8 cores. If you want to look at cores and FPU's you could say it's a 12 processor chip, but that would be equally as wrong as saying the 8 core chip is 6 cores.

The FX 6 range is 6 cores if you are getting mistaken with that. The FX 8 is 8 cores. Clue's in the name.

I meant 6-ish as a way of comparing it to 'true' individual cores, like Intel's or AMD's pre-bulldozer architectures.

It is technically 8 cores, but starved of resources. Each core must share some die resources with another core in its 'pair'.

 

[AllBodies]

It's not an 8 core CPU on a technicality. The FX8 is a true 8 core chip in the truest sense of the definition.

Lot of CPU's share resources across dies. Some even with other cores on other CPU's altogether. Pairing up and sharing resources is what parallel computing is based on.

No, it's eight cores. Are you going to claim that older single core processors had zero cores if they didn't have an integrated FPU? Once there were CPUs and then there were CPUs that had a "maths co-processor". Then the maths co-processor got integrated into the CPU. And all was well. But FP calculations make a small part of most use cases so CPUs becoming multi-core it no longer makes sense to have every core have its own dedicated FP unit, so they share them. That's efficient. There's no "more like 6-ish cores". It's eight cores with four FPUs.

Unless you're talking about how cache or pipelines are implemented for some reason in which case it still doesn't change that it's eight cores.

You're both reading too literally into my point.

The simple situation is it's a 4-pair design, rather than 8 'standalone', resulting in 8 weak cores with high clockspeed and very good parallelism. And cannot be compared to 8 Intel cores in terms of performance equivalence.

That's all I was saying.