What will happen when Intel brings core counts greater than four to mainstream?

[cbn]

So for a low end gaming desktop that would normally be powered by a Core i3, I would hope a 65W Intel 6C/12T could be used instead.

This 6C/12T with a 2C turbo equivalent to the clocks on the Core i3 it replaces.

Then Intel can develop better Xeon-D based consumer SKUs in such a way that conflicts with HEDT are reduced or eliminated.

Ultimately we end up with the following choice:

Additional cores vs. iGPU

 

[biostud]

So are you questioning whether there is use for more performance in mainstream computers? Otherwise the purpose of more cores would be just that, to provide more performance.

I think that once mainstream need more CPU power than quad cores, we will have an efficient cloud computing up and running.

 

[Systems analyst]

The "famous" ARM is going to take on x86 anytime "soon" in traditional x86 segments?

ARM is already losing tablets. Its a oneway battle.

One year ago Intel reported $2 .8 billion earnings for 2014 Q1 PC group. This year, they reported $1 .4 billion earnings in the combined PC + mobile group for 2015 Q1. Some people think that they combined PC and mobile to hide the latter' s losses, but maybe they are disguising PC losses. We will see in the later quarters, but if a fall in earnings of $1 .4 billion is repeated over four quarters, it will be a fall of $5.6 billion for this year. Predictions are for a 6% fall in PCs shipped, and a move towards cheaper form-factors.

 

[cbn]

I think that once mainstream need more CPU power than quad cores, we will have an efficient cloud computing up and running.

Cloud sometimes has its place, but for home automation (which is something I mentioned in the OP) I don't think requests should have to go through a cloud server:

http://www.anandtech.com/show/9174/home-automation-systems-a-consumer-checklist

 

[Systems analyst]

OpenCL, HSA, CUDA whatever would also solve all the worlds problems. They didnt.

You assume all code can be parallelized and its simply wrong. Even servers with "endless money and cores" are plagued by massive amount of serial code. And the only way to use more cores is more users. This is also why ARM chips are too slow for transaction servers and a lot of web frontends.

The ARM model is based on a very diverse 'sea of SoCs', where the ARM cores play a role, but are not always the main number - crunchers; accelerators are used in the SoCs to do the heavy lifting; see for example, the HP Moonshot servers with TI SoCs, in use by PayPal for real-time financial transaction monitoring.
The French have developed servers using ARM chips, in conjunction with Intel chips. Other servers configurations using ARM and Intel jointly have been tested.
At an HPC conference in Germany last year, AMC SoCs used in conjunction with Nvidia accelerators were discussed; they were reported to have remarkable efficiency compared with Intel solutions.

 

[TheELF]

The price of Intel hexcores is already dropping (example: i7 5820K @ $389 which is the first Intel hexcore to drop below $400).

Yeah,cause thats what I was talking about right?
The 5820 has no igpu and still is $50 more expensive than the 4970 ,cores are expensive and companies are gonna chrage you for them.

 

[cbn]

cores are expensive and companies are gonna chrage you for them.

That used to be true.

Here is an example (i7-4930K 6C/12T @ $583: http://ark.intel.com/products/77780/Intel-Core-i7-4930K-Processor-12M-Cache-up-to-3_90-GHz)

Now the difference is only $50 for an additional 2C/4T (comparing price of i7-5820K vs. i7-4790k).

And eventually I expect certain Intel 6C/12T to be priced below the top 4C/8T processors.

 

[sxr7171]

Cloud sometimes has its place, but for home automation (which is something I mentioned in the OP) I don't think requests should have to go through a cloud server:

http://www.anandtech.com/show/9174/home-automation-systems-a-consumer-checklist

Tell me about it. The new Insteon systems are cloud based and lack any decent response. Lights turn on and off with a lag of 2-10 seconds. Thankfully there is a way to address the hub directly using the old IP address method. They really need to rethink moving everything to the cloud.

 

[Fjodor2001]

Now the difference is only $50 for an additional 2C/4T (comparing price of i7-5820K vs. i7-4790k).

You'll have to add the cost of a discrete GFX card and a more expensive socket 2011 motherboard to that. So perhaps $200-250 price difference, which is quite a lot.

 

[Alan2020]

I don't think we are very far away, considering the 6 core 4930K costs around $250 more than the highest end mainstream i7. Maybe in 1 or 2 years we will see 6 core mainstream i7s.

 

[cbn]

Tell me about it. The new Insteon systems are cloud based and lack any decent response. Lights turn on and off with a lag of 2-10 seconds. Thankfully there is a way to address the hub directly using the old IP address method. They really need to rethink moving everything to the cloud.

Other concerns mentioned in the Anandtech article were data mining and security.

 

[CriticalOne]

It doesn't make sense to include more than four cores on a mainstream platform.

We have to remember that the exact same die that constitutes a i7 4790k is what is being used to make mobile processors. Only the most highest end processors on mobile use a four core setup due to the heat output. If the mainstream die size had six cores, that would make mobile CPUs more expensive to make for something they wont ever have (6+ cores).

Also, its a mainstream series of processors. Its known that most users are satisfied with a 2C/4T configuration. Gamers go with a 4C/4T CPU if possible and many power users are happy to have a 4C/8T setup. Its only a really small minority of PC users who find themselves needing more than four cores for whatever they want. All of this push for more cores on the mainstream platform from, lets face it, enthusiats, is pretty selfish because it would increase the cost of the platform for most people, essentially making the masses subsidize the few so they can get cheaper hex cores.

Besides, its never just an simple increase in core count. You pretty much have to increase cache sizes and potentially more IMCs for more channels. That means you are going to need more connecting pins on the socket, making the socket and CPU package larger and more expensive. Small things tend to add up for servers and big businesses.

 

[Enigmoid]

It doesn't make sense to include more than four cores on a mainstream platform.

We have to remember that the exact same die that constitutes a i7 4790k is what is being used to make mobile processors. Only the most highest end processors on mobile use a four core setup due to the heat output. If the mainstream die size had six cores, that would make mobile CPUs more expensive to make for something they wont ever have (6+ cores).

Also, its a mainstream series of processors. Its known that most users are satisfied with a 2C/4T configuration. Gamers go with a 4C/4T CPU if possible and many power users are happy to have a 4C/8T setup. Its only a really small minority of PC users who find themselves needing more than four cores for whatever they want. All of this push for more cores on the mainstream platform from, lets face it, enthusiats, is pretty selfish because it would increase the cost of the platform for most people, essentially making the masses subsidize the few so they can get cheaper hex cores.

Besides, its never just an simple increase in core count. You pretty much have to increase cache sizes and potentially more IMCs for more channels. That means you are going to need more connecting pins on the socket, making the socket and CPU package larger and more expensive. Small things tend to add up for servers and big businesses.

With 14nm and aggressive turbo 6C would be no different than 4C was for sandy bridge on mobile on 14nm. Which had lower clocks but no trouble performing.

This also allows intel to bring more value to the xeon e3 line as well.

On 14nm adding two cores adds ~20 mm^2 for cores and cache which is really a drop in the bucket considering you add 50% cpu performance. It will not require more memory channels and will allow the smallest xeon dies to be a little bigger (don't have to cut down as much the big xeon dies).

Not everyone is proposing for intel to shift the entire lineup. Simply adding another die with 6C + HT and GT2 graphics would be great. Considering the number of dies intel makes (2+1, 2+2, 2+2U, 2+3U, 4+2, 4+3e) for the mainstream this really isn't a huge cost, especially when you consider intel doesn't need to sell so many heavily cut down x99 chips (can still offer a SKU for people who want more PCIe lanes or quad channel RAM).

 

[cbn]

All of this push for more cores on the mainstream platform from, lets face it, enthusiats, is pretty selfish because it would increase the cost of the platform for most people, .

1.) Intel already has a Broadwell 8C/16T die (Xeon-D) with dual channel DDR4 @ 160mm2. (It is actually smaller than the die for Broadwell 4C/8T + GT3e which is 169mm2). The measurement of 169mm2 for quad core + GT3 found in this document

2.) AMD's mainstream AM4 platform will have both 8C/16T (no iGPU) and quad core with iGPU on it.

 

[Pariah]

Until more mainstream software takes advantage of 4C CPU's, what's the point of pushing more cores into the mainstream? It makes no financial sense for Intel to move on when what we currently have is not being taken advantage of. AMD has 6 and 8 core CPU's. How many highend users are buying those? None. With so little software taking advantage of that many cores, Intel wins almost all the benchmarks with their 4 core CPU's.

 

[cbn]

Until more mainstream software takes advantage of 4C CPU's, what's the point of pushing more cores into the mainstream? It makes no financial sense for Intel to move on when what we currently have is not being taken advantage of. AMD has 6 and 8 core CPU's. How many highend users are buying those? None. With so little software taking advantage of that many cores, Intel wins almost all the benchmarks with their 4 core CPU's.

Games scale to eight threads. Notice the older design 4C/8T i7 2600K beating the modern 4C/4T i5-4670K in the following benchmark:

So 160mm2 Intel Broadwell 8C/16T with hyperthreading disabled (for 8C/8T) would be even faster clock for clock than the more expensive to produce 169mm2 4C/8T i7-5775C (which also needs a separate 80mm2 eDRAM die).

Of course, that is just a game (for a single user). What I referring to in the OP would involve multi-user.

 

[Enigmoid]

Until more mainstream software takes advantage of 4C CPU's, what's the point of pushing more cores into the mainstream? It makes no financial sense for Intel to move on when what we currently have is not being taken advantage of. AMD has 6 and 8 core CPU's. How many highend users are buying those? None. With so little software taking advantage of that many cores, Intel wins almost all the benchmarks with their 4 core CPU's.

People don't buy core for the sake of having more cores. They buy more cores for the sake of better performance. AMD's 8 cores perform comparably to intel's quads; given other factors intel wins out and people end up buying 4C/8T chips instead of 8 C chips (or 4M/8T). Intel having 6C $330 chips (on z97 equivalent not the significantly more expensive x99) is completely different than 8C FX chips.

Pushing 6C into the mainstream is completely different than introducing a new top end mainstream/enthusiast chip. Intel could easily move the i5 range to include both 4C/4T (i5 - 65xx) and 4C/8T (i5- 66xx) (same chips as HT is feature binned so this doesn't affect margins too much) and introduce a 6C/12T chip for high end z97 equivalent at the same $330-350. While margins would decrease for the $330 6C chip, intel would sell significantly more $330 chips (similar to how the 4970k vs. 4690k favours the i7 devils canyon more than the 4770k vs. 4670k did because of the significantly higher stock clock, especially for builds without OC). Margins on a 6C z97 ish product would be higher than a cut down xeon (die is significantly smaller).