Do you think Intel should offer dual core chips with GT4e iGPU?

[cbn]

Do you think Intel should offer dual core chips with GT4e iGPU?

 

[witeken]

Sure, why not?

 

[AtenRa]

You mean they should integrate dual x86 cores in to their GPUs, because at 14nm the GT4 is close to 3 times the size of those 2x x86 cores.

Well, for laptops yes if the price is right.

 

[cbn]

Sure, why not?

I think it would work great myself.

And with Broadwell only 6.8mm2 to 6.9mm2 per CPU core (without L3 cache), I'll bet the Skylake core is very small too.....so Intel could just use the 4C die and disable two cores economically.

Put it on a 5x5 board (with Thunderbolt III) and I would be very happy myself. This provided the price was right.

Ideally such a build with the dual core GT4e would be attractive enough that some folks would consider it instead of Xbox One.... and therefore become a type of lowest common denominator target for Linux game/Open CL developers. (Example: my own comparison of OC G3258 (dual core)/R7 250X vs Xbox One in BF4 here).

 

[NTMBK]

Sure, it would be great for mini gaming PCs. Expensive, though, with 2x128MB cache. And Linux gaming is still a terrible experience.

 

[Gryz]

In the future, Intel (and others) might only be able to sell CPUs with more cores. Because clock-speeds will never go up, IPC improvements will be minimal. But we might still have more transistors per euro/dollar.

The value of those future multi-core CPUs will only go up if there is software to do something with those extra cores. Today there is a lack of parallel software. And there is a huge lack of understanding and education amongst programmers of all kinds to do proper parallel programming. A better toolset or paradigm for programming might help. But I don't see one coming yet.

Developers (companies and individual programmers) will be more inclined to keep ignoring parallel programming if part of their customer-base is still running on 1 or 2 cores. "We can't really spend the resources to make our software multi-threaded yet, because X percent of our customers has only 2 cores. It would be too expensive, and we still have to make sure our software runs on only 1 or 2 cores".

The longer Intel keeps single- and dual-cores around, the worst the impact on their sales-numbers in the future will be.
They need to move everybody to 4 cores or more.
So better multi-threaded software will be developed.
So their future CPUs will be worth their price.

 

[cbn]

In the future, Intel (and others) might only be able to sell CPUs with more cores. Because clock-speeds will never go up, IPC improvements will be minimal. But we might still have more transistors per euro/dollar.

The value of those future multi-core CPUs will only go up if there is software to do something with those extra cores. Today there is a lack of parallel software. And there is a huge lack of understanding and education amongst programmers of all kinds to do proper parallel programming. A better toolset or paradigm for programming might help. But I don't see one coming yet.

Developers (companies and individual programmers) will be more inclined to keep ignoring parallel programming if part of their customer-base is still running on 1 or 2 cores. "We can't really spend the resources to make our software multi-threaded yet, because X percent of our customers has only 2 cores. It would be too expensive, and we still have to make sure our software runs on only 1 or 2 cores".

The longer Intel keeps single- and dual-cores around, the worst the impact on their sales-numbers in the future will be.
They need to move everybody to 4 cores or more.
So better multi-threaded software will be developed.
So their future CPUs will be worth their price.

Intel can always sell the quad core GT4e also, but it will be expensive.

What I think you are referring to is something like the following:

Consumer version of Xeon-D (Broadwell 8C/16T) @ 160mm2 over a 4C/8T Broadwell GT3e @ 169mm2. (ie, extra cores rather than extra iGPU at roughly the same die size)

 

[Gryz]

No.
What I am refering to, is simple.

If software does not become better at multi-threading, then soon nobody will buy new CPUs, except for replacement when hardware dies.

Software will not become multi-threaded, unless the software developers have zero excuses to delay going full multi-threaded. Having a significant number of 1-core and 2-core CPUs in the field would be such an excuse.

Intel pushing for less 1/2-core CPUs in the field -> force software developers to make better multi-threaded software -> make 4+ core CPUs more valuable -> ensure that Intel will keep selling CPUs, even when there is no progress possible, except more cores.

 

[lehtv]

I think Intel should offer as many different core count / HT / iGPU variants as possible without necessitating higher prices. However it doesn't really matter what I think, Intel doesn't make chips to please everyone, but to make money.

 

[cbn]

No.
What I am refering to, is simple.

If software does not become better at multi-threading, then soon nobody will buy new CPUs, except for replacement when hardware dies.

Software will not become multi-threaded, unless the software developers have zero excuses to delay going full multi-threaded. Having a significant number of 1-core and 2-core CPUs in the field will be such an excuse.

One thing that I notice Intel doing is offering a discount per core if the user either buys dual core or hexcore (eg, i7 5820K was only $50 more than i7-4790K). Quad core has the most expensive price per core (currently)

So I think we will see what you are referring to occur, but principally for the non-iGPU processors (at least for the immediate future). With that mentioned, it might be that Intel eventually releases a 8C/16 Server type APU with iGPU.

 

[throwa]

Absolutely.

If not on Pentium then definitely the Core i3 series. I mean people are paying $110-$120 for these Core i3's, the least Intel could do is provide an integrated GPU that is at least comparable to AMD APU's in the same $100-$140 price bracket.

 

[cbn]

I voted yes (without the caveat for the 2C/4T).

Reason: I think 2C/2T could potentially work with a GT4e without stuttering. This due to the fact the GT4e could still be a bottleneck to the CPU. However, with this mentioned if one of the variants is 2C/2T it should have high clocks.

 

[frozentundra123456]

I voted yes because more options are better. However, I would much rather see a mainstream hex core, or iris pro standard in laptops.

 

[Arachnotronic]

No; I don't think such a SKU would be very popular, especially at the prices that Intel would likely charge for it.

 

[cbn]

No; I don't think such a SKU would be very popular, especially at the prices that Intel would likely charge for it.

Intel really doesn't really charge much of a premium for increasing iGPU size though.

Moving from dual core CPU to quad core CPU....Yes.

Moving from GT2 to GT3....No

(Ironically the Value segments for consumer processors appear to be dual core, hexcore and large iGPU. Quad core + small iGPU commands highest price per mm2 silicon)

 

[Yuriman]

I think it would work great myself.

And with Broadwell only 6.8mm2 to 6.9mm2 per CPU core (without L3 cache), I'll bet the Skylake core is very small too.....so Intel could just use the 4C die and disable two cores economically.

Put it on a 5x5 board (with Thunderbolt III) and I would be very happy myself. This provided the price was right.

Ideally such a build with the dual core GT4e would be attractive enough that some folks would consider it instead of Xbox One.... and therefore become a type of lowest common denominator target for Linux game/Open CL developers. (Example: my own comparison of OC G3258 (dual core)/R7 250X vs Xbox One in BF4 here).

Those mm² figures per core - do those include cache and interconnects required for each core? How big is the memory controller, and how much dedicated extra space is required per additional core?

 

[cbn]

Those mm² figures per core - do those include cache and interconnects required for each core?

The measurement is for the core and the L1 and L2 cache, but not the L3 cache.

 

[Ken g6]

With Skylake, even laptop i5's are going to be quad-core. So, no, I don't think GT4e with dual-core is necessary.

 

[cbn]

With Skylake, even laptop i5's are going to be quad-core. So, no, I don't think GT4e with dual-core is necessary.

I think for desktop, GT4e would be more desirable with the dual core crowd though. (re: quad core desktop users would be more likely to disable a large iGPU and use dGPU instead....but for a dual core user GT4e is actually appropriate)

 

[NTMBK]

I think for desktop, GT4e would be more desirable with the dual core crowd though. (re: quad core desktop users would be more likely to disable a large iGPU and use dGPU instead....but for a dual core user GT4e is actually appropriate)

The "dual core crowd" are extremely price sensitive. This is not going to be for them. GT4e is going to be a massive 14nm die, with two 22nm cache dies. It's a high cost, high perf/W monster, specialised for premium compact systems. It has more in common with the Fury Nano than with the G3258.

And building such a massively expensive part and then turning off Hyperthreading is just crazy.

 

[mohit9206]

Yeah i would love a desktop i3 with gt4e graphics. But price should not exceed $150. Gt4e is more powerful than my amd 7750 so i wouldn't need to buy a discrete gpu as gt4e would be a good upgrade.

 

[NTMBK]

Yeah i would love a desktop i3 with gt4e graphics. But price should not exceed $150. Gt4e is more powerful than my amd 7750 so i wouldn't need to buy a discrete gpu as gt4e would be a good upgrade.

Hah, good luck with that. It's going to be a considerably larger chip than the i7-6700k, and you want it to sell for half the price?

 

[mohit9206]

Hah, good luck with that. It's going to be a considerably larger chip than the i7-6700k, and you want it to sell for half the price?

Yes i want intel to sell it for half the price as a gesture of goodwill.

 

[DrMrLordX]

Intel's pricing scheme doesn't seem to be based on die size, though. It's market segment/performance expectation. Example: a 5820k has a die size of ~355mm squared, which is the same size as the 5960x die. The price on the two CPUs is certainly not the same.

The 6700k, with a much smaller die, has a price that is close to that of the 5820k. Yeah I know different processes, different price/transistor etc.

Anyway, point being, Intel charges as much as they do for the 6700k with the understanding that it is at the top of it's product heap. There is no other LGA1151 or LGA1150 processor which is meant to be faster than that chip. If 2c/4t i3 + GT4e winds up at the top of its market segment - that is, the fastest core i3 with the fastest iGPU - then expect Intel to price it according to the price patterns already established for i3 processors. I would expect a price in the range of $200-$250.

 

[ShintaiDK]

2+4e doesnt make sense. 2+3e is what we gonna get in the U segment. But there is also a power saving element there.

Just get 4+4e, its less than 20mm2 bigger and offers the full monty when you already went all the way for GT4e.