Skylake (Mainstream) CORE count Speculation Poll

[SOFTengCOMPelec]

That's nice, but subjective feelings aren't useful when talking about absolute, measurable facts. Maybe a phone might still be very slow but software has been optimized and/or simplified to make the experience satisfactorily.

The smartphone market is a fast growing market where a lot of money can be earned. Since the CPU is one of the distinguishing features of a phone and faster CPUs were necessary, there have been a lot of investments in making mobile CPUs faster. Before that, both volume and ASPs were lower, resulting in inferior CPUs.

I agree.
There are two different ways of looking at the arm processors current performance. I was effectively saying that it was "good enough" for what I wanted to do, an a battery powered, handheld tablet device.
But on the other hand, if benchmarked and/or used to calculate very high demanding software (e.g. A comicated Chess position), it would have behaved relatively badly, compared to a current, high end, mainsteam Haswell desktop processor.

Today, every effort is made by Qualcomm, Nvidia, Apple and Intel to squeeze the most performance out of a given TDP. So in the end, the CPU with the best transistors will win, independent from ISA.

Yes and no.

Analogy:
There (in theory, at least) have been times when a particular company has been able to produce the "best" single, discrete transistors, e.g. In the 1960s.

BUT, in some cases they have flatly refused to actually put on the market the "best" transistors that they can make, (in the lowest cost range), because they are worried that it would compete with their own, upper market (high profitability, high margins) higher end transistors. (N.B. I am putting this argument through, as a theoretical possibility, to avoid arguments about this, so if I am wrong, then sorry!).

I.e. The Intel manufacturing process advantage, would be artificially hindered, if they stick to only quad cored CPUs.

E.g. Eventually the competition, may end up with a worse process, but one which is not that bad, compared to Intel's.
So if Intel continue to refuse to release sensibly priced, more than 4 cored processors, and yet the competition releases 6/8 or whatever cored processors, then although they would still have worse instructions per watt ratings, and be slower (single thread), they could sail past the Intel quad core offerings, in multi-threaded benchmarks.

I think we actually had this happen with the FX-8350, which for one or two highly multithreaded tasks, benchmarked slightly faster than the Sandy bridge.
If the Fx-8350 had been much closer to the Intel process of that time, then it might have been able to be more of a success.
(Sadly, in truth, the Intel chip was so much faster, even for most multi threaded stuff and because the FX-8350 used so much power, the Intel chip was the thing to go for).

Tl:dr;
If eventually Intel is 14 nm, quad core, Skylake
And Competition are 16 nm, 8 core, somewhat-near-skylake
Then multi-threaded benchmarks may show the competition beating Intel. (At least performance wise, BUT TDP/efficiency would probably suffer).

But I agree we are NOT there yet, and it may never happen.

 

[Bolshoi Booze]

probably 4 cores but 10-15% better IPC than haswell, though i believe skylake will be a more exciting desktop cpu than haswell if they manage to improve OC

 

[Khato]

Definitely an interesting question. Really, I'd expected to see mainstream 6-core processors from Intel with Broadwell thanks to 14nm density scaling, but that didn't happen due to it being realigned to a mobile focus. Hence I'd be quite surprised if it doesn't happen with Skylake, but it's certainly possible that it won't...

 

[BrightCandle]

I suspect just 4. Many people are happy with dual cores and with Intel's focus being on mobile and laptops they are reducing power consumption rather than increasing performance with those process changes.

 

[Sweepr]

I think Intel will focus on reducing power + improve iGPU performance + improve per-core performance in mainstream LGA115x platforms for the time. On the other hand, Haswell-E is a 8C/16T chip and Broadwell-E is rumoured to have a 10C/20T desktop version, so they could (and I wish they will) lower prices for 6C/12T LGA2011 parts in 2014/2015. Lots of users want something better than Intel quads but are not willing to pay close to $600 for 6-core goodness.

 

[SOFTengCOMPelec]

I think Intel will focus on reducing power + improve iGPU performance + improve per-core performance in mainstream LGA115x platforms for the time. On the other hand, Haswell-E is a 8C/16T chip and Broadwell-E is rumoured to have a 10C/20T desktop version, so they could (and I wish they will) lower prices for 6C/12T LGA2011 parts in 2014/2015. Lots of users want something better than Intel quads but are not willing to pay close to $600 for 6-core goodness.

I was thinking exactly the same thing, before creating this thread.

Ideally make an affordably priced >4 cored Skylake, but if Intel are NOT happy to do this, then at least make a lower cost LGA2011 hex core part(s) available, even if it is priced MORE THAN existing I7-4771's, BUT significantly less than the existing extreme hex core processors. (Just like you just mentioned above).

Intel should see that by doing so, it will allow LGA2011 to become somewhat more mainstream (due to significantly lower prices at the bottom end), which should allow the prices of LGA2011 motherboards to drop with the larger sales volume, potentially leading to Intel making more higher end LGA2011 sales as well.

I do concede that my lack of intricate knowledge about the "true" manufacturing/development costs of the LGA2011 platform, might mean that it may well need to cost significantly more than the mainstream devices.

Although I do remember they use to sell (they may still do), a quad cored LGA2011 part, at about the same price as the mainstream I7 counterpart.
So maybe it is feasible to have a similar hex cored device at the bottom, with more expensive 8 (or even more later) cored parts at the higher end, of the extreme range.

EDIT: Although I may seem all for the hex cored LGA2011, usually when I look into the viability of it for myself, the combination of extra costs, IPC losses as it is usually a generation behind the current mainstream I7 architecture (or a new arch is just round the corner), and its relatively high TDP, make it NOT worthwhile compared to say a quad cored Skylake.

 

[Ajay]

If Skylake-EP starts @ 10 cores, or more, then it may not make sense to cut it down to less than 8 cores. In that case, Intel may want a six core single socket solution for small business servers. Since the single socket server CPU comes from the mainstream desktop solution, there would be would be the possibility of a a market driven mainstream hexacore CPU (and probably new entry point for Intel's HEDT line). Ultimately, I think this dynamic will play out and we'll get > 4 core CPU.

IMHO.

 

[ShintaiDK]

The longer the mainstream stays on quadcores, the harder it will be for nVidia/AMD to sell dGPU/APUs.

And we still dont have any killer apps on the desktop/mobile that need more than 4 cores. Not even close. And the lack of software development due to the natural limitations is the one setting the barriers.

The 99% crowd wants faster IGPs and faster CPU cores with an ever better performance/watt ratio and lower consumption, not more cores and higher consumption.

 

[SOFTengCOMPelec]

If Skylake-EP starts @ 10 cores, or more, then it may not make sense to cut it down to less than 8 cores. In that case, Intel may want a six core single socket solution for small business servers. Since the single socket server CPU comes from the mainstream desktop solution, there would be would be the possibility of a a market driven mainstream hexacore CPU (and probably new entry point for Intel's HEDT line). Ultimately, I think this dynamic will play out and we'll get > 4 core CPU.

IMHO.

I hope you are right!

 

[SOFTengCOMPelec]

The longer the mainstream stays on quadcores, the harder it will be for nVidia/AMD to sell dGPU/APUs.

And we still dont have any killer apps on the desktop/mobile that need more than 4 cores. Not even close. And the lack of software development due to the natural limitations is the one setting the barriers.

The 99% crowd wants faster IGPs and faster CPU cores with an ever better performance/watt ratio and lower consumption, not more cores and higher consumption.

Ironically more cores may lead to LOWER power consumption, especially for mobile application.
I.e. More cores at a lower clock frequency can (in theory) produce the same performance (if software can efficiently use the increase in core number), using less power, than a similarly powerful, higher clocked part, with correspondingly fewer cores.

E.g. This is what seems to have happened with modern games consoles, as they may have got fed up with the heat related failures of the previous generation.

I agree (as you have previously said in other threads) that it is easier said than done, to make software efficiently use more cores, over clock speed and IPC improvements.

http://en.wikipedia.org/wiki/Amdahl's_law

EDIT: re no killer apps.
What about the trend for modern and upcoming games, to usefully use hex cored CPUs ? (In my opinion, but I'm not really enough into modern gaming to have a reliable opinion here).

 

[Bateluer]

The quad core versions will expand further into the lower markets, pushing dual cores into even lower end. Likewise, there'll be performance mainstream parts with hex cores.

 

[NostaSeronx]

I voted for four cores and eight threads but meant; Four cores and sixteen threads.

Moar Therds!

 

[SOFTengCOMPelec]

I voted for four cores and eight threads but meant; Four cores and sixteen threads.

Moar Therds!

Mindbogingly, I think there are already IBM processors, which do just that (many, many more threads, than there are cores).

Yes an Intel 16 core/256 thread CPU will do just fine (until the 64C/4096T one comes out, the week after).

EDIT: I think graphics cards are like massively high thread count CPUs, with significantly less flexibility than conventional desktop CPUs.

 

[ShintaiDK]

Ironically more cores may lead to LOWER power consumption, especially for mobile application.

I assume you talk about big.LITTLE? You do know Qualcomm and Apple rejected it, right?

E.g. This is what seems to have happened with modern games consoles, as they may have got fed up with the heat related failures of the previous generation.

Console manufactors didnt pick Jaguar for that. They did it due to cost and nothing else. MS and Sony would have loved something better if it was possible. Both consoles struggle badly with multiple 30FPS releases due to massive CPU bottleneck. MS even had to up the core speed in the last minute to try remove some of the headaches.

EDIT: re no killer apps.
What about the trend for modern and upcoming games, to usefully use hex cored CPUs ? (In my opinion, but I'm not really enough into modern gaming to have a reliable opinion here).

How much difference is there in those couple of games with 4C/4T and 4C/8T vs 6C/12T and 8C/16T?

 

[SOFTengCOMPelec]

I assume you talk about big.LITTLE? You do know Qualcomm and Apple rejected it, right?

Sorry, I did not explain it well enough. (I should have used later processors in the following example, but it illustrates what google searches came up with, and should still be valid).

E2643 has 4 cores @3.3 GHz, and has a TDP of 130W
E2658 has 8 cores @2.1GHz, but uses less power, with a TDP of only 95W

So if the software can usefully (and efficiently) utilize double the number of cores, it can use less power, and have more performance at the same time.

http://ark.intel.com/products/64587/Intel-Xeon-Processor-E5-2643-10M-Cache-3_30-GHz-8_00-GTs-Intel-QPI

http://ark.intel.com/products/61428/Intel-Xeon-Processor-E5-2658-20M-2_10-GHz-8_0-GTs-Intel-QPI

Console manufactors didnt pick Jaguar for that. They did it due to cost and nothing else. MS and Sony would have loved something better if it was possible. Both consoles struggle badly with multiple 30FPS releases due to massive CPU bottleneck. MS even had to up the core speed in the last minute to try remove some of the headaches.

I remember reading about concerns about the overheating (issues with the old generation consoles) being (part?) of the reason they chose the high number of cores, at a lower frequency.
BUT I am happy to fully accept YOUR explanation, and agree with it, because I do not know much about the finer details of the latest gaming consoles.

How much difference is there in those couple of games with 4C/4T and 4C/8T vs 6C/12T and 8C/16T?

That's a very good question, and I'm NOT sure of the answer. But someone with considerably more (latest) games knowledge on these forums, may well be able to answer it.

I remember reading (on our forum) that dual core (2 thread) cpus are NOT really the best these days for gaming, and at the very least a dual core (4 thread), hyper-threaded processor, if not a full fledged Quad core cpu, should be used.

 

[HurleyBird]

So if the software can usefully (and efficiently) utilize double the number of cores, it can use less power, and have more performance at the same time.

And if software doesn't you can throttle the unused cores and turbo the ones that are needed.

More cores will always increase energy efficiency if you decide to use them in an efficient manner.

 

[SOFTengCOMPelec]

And if software doesn't you can throttle the unused cores and turbo the ones that are needed.

More cores will always increase energy efficiency if you decide to use them in an efficient manner.

Good points.
Also I find that a computer with many cores is usually smoother and feels more powerful (as things running in the background usually don't effect the apparent speed of the foreground application), compared to single/small number of cores cpus.

 

[rgallant]

I wonder if 6 fast cores ,with no ht would work out ?
use the ht tech for better true multi core processing.
-and would not venture into the 6c/12t server/2011 chip space.

 

[ShintaiDK]

I wonder if 6 fast cores ,with no ht would work out ?
use the ht tech for better true multi core processing.
-and would not venture into the 6c/12t server/2011 chip space.

HT basicly comes for free. It take up something like 1% of the total diesize. And less than 5% of a core die.

 

[ShintaiDK]

E2643 has 4 cores @3.3 GHz, and has a TDP of 130W
E2658 has 8 cores @2.1GHz, but uses less power, with a TDP of only 95W

So if the software can usefully (and efficiently) utilize double the number of cores, it can use less power, and have more performance at the same time.

If is the key word. Also the diesize of the 8 core is much greater. And you know Amdahls law as well as I do. In the wast majority of applications that 4 core will be 50% faster. If everything was so perfect, then we would all use Xeon Phis as CPUs.

I remember reading about concerns about the overheating (issues with the old generation consoles) being (part?) of the reason they chose the high number of cores, at a lower frequency.
BUT I am happy to fully accept YOUR explanation, and agree with it, because I do not know much about the finer details of the latest gaming consoles.

Try compare the coolers and you know why. And the GPU is much weaker compared to last time as well.

But again, new consoles overheat too:
https://www.psu.com/a021674/PS4-blinking-red-light---How-to-stop-your-PS4-from-overheating
http://www.playstationlifestyle.net...hida-talks-about-the-ps4s-red-light-of-death/

 

[VirtualLarry]

HT basicly comes for free. It take up something like 1% of the total diesize. And less than 5% of a core die.

But it's not free, in terms of power-consumption. Otherwise, we would see 4-way HyperThreading in Core chips by now.

 

[ShintaiDK]

But it's not free, in terms of power-consumption. Otherwise, we would see 4-way HyperThreading in Core chips by now.

4 way(thread?) doesnt make sense if the unused execution resources isnt there.

 

[know of fence]

A lot of people seem to get the Intel Tic-Toc and expect nothing else, which is a safe bet IMO. So following are two less likely scenarios:

The new 14-nm node will allow to bring 4-core/8-threads to all their offerings (including Celerons) thus finally creating the egg for the multi-threaded software chicken. Intel can bin and fuse things off to increase yields and create a range of SKUs.
A similar thing has been somewhat the case since the Introduction of the Core i brand since all the Core i CPUs come with at least 4 threads already, before that core 2 created widespread adoption of dual core CPUs.

We actually get a 6 core part and finally very intuitive and neat core number based SKUs:
i3 -> 2 core
i5 -> 4 core
i7 -> 6 core

 

[VirtualLarry]

We actually get a 6 core part and finally very intuitive and neat core number based SKUs:
i3 -> 2 core
i5 -> 4 core
i7 -> 6 core

That's WAAAY too logical for Intel. How would they confuse their potential customers otherwise?

Edit: But count me in for a quad-core Celeron. How cheap can they go?

 

[NTMBK]

We actually get a 6 core part and finally very intuitive and neat core number based SKUs:
i3 -> 2 core
i5 -> 4 core
i7 -> 6 core

Still too confusing to the average Joe. I demand a 7 core CPU!