Intel CPU Rant

[frozentundra123456]

They've done the math. It would cost too much to sell any quantity to consumers. That means they'd lose money.

No business will purposely set out to lose money.

Hasnt stopped them from dumping billions into the rathole of mobile. And how much have they spent on iris graphics, which is barely used.

But they know with no competition, they can sell 2 or 3 generation old server rejects and an expensive chipset because there is no alternative. I am by no means an AMD fan, but maybe Zen will change that.

In any case, sometimes a company just needs to sacrifice a tiny bit of margin (assuming you are correct, which you have not documented) to produce a quality, top of the line product. It certainly would have a minimal impact on their bottom line overall. Just devote some of that die space devoted to the igpu to a couple more cores and provide a tiny igpu or even none at all like the x99 chips.

 

[bunnyfubbles]

we basically hit a clockrate wall as far back as Sandy Bridge, from here on out it is looking more and more like we're stuck with more cores and architecture improvements

and considering that, if anything we should be complaining that we're not getting more cores, as the mainstream chips are still stuck with 4 cores (been that way since Q1'07), meanwhile we have to pay ~$1000 for the extreme edition chips to get more than 6 cores...of which original Sandy Bridge-E chips were 8 core native with 2 cores disabled, so all this time we've been artificially denied more cores even on the high end.

But they know with no competition, they can sell 2 or 3 generation old server rejects and an expensive chipset because there is no alternative. I am by no means an AMD fan, but maybe Zen will change that.

Its not that they don't have competition, its that the desktop market is shrinking. Most people have no need for bulky desktop computers, hell, even the laptop market is being pushed aside for the ever more convenient tablets and phones

 

[Sheep221]

Its not that they don't have competition, its that the desktop market is shrinking. Most people have no need for bulky desktop computers, hell, even the laptop market is being pushed aside for the ever more convenient tablets and phones

Quad core is still mainstream for sole reason: cheap nodes featuring more than 4 cores and more than 8 MB of cache are not yet developed - they do exist but they are expensive to manufacture so the supply is reserved for enterprise market which has required demand and capitalization for such a chips.

 

[JimmiG]

Hasnt stopped them from dumping billions into the rathole of mobile. And how much have they spent on iris graphics, which is barely used.

But they know with no competition, they can sell 2 or 3 generation old server rejects and an expensive chipset because there is no alternative. I am by no means an AMD fan, but maybe Zen will change that.

In any case, sometimes a company just needs to sacrifice a tiny bit of margin (assuming you are correct, which you have not documented) to produce a quality, top of the line product. It certainly would have a minimal impact on their bottom line overall. Just devote some of that die space devoted to the igpu to a couple more cores and provide a tiny igpu or even none at all like the x99 chips.

Yeah, look up some die shots of the 6700K:
http://abload.de/img/unbenanntopuco.png
The integrated GPU takes up a huge chunk of the total space. It's silly to pay for a large, integrated GPU that will never get used. Just having it there, even if disabled, probably impacts temperature, power and overclocking potential as well. Very few consumers who buy a 6700K use it with the Intel IGP. If Intel cared at all about the mid-range and high-end segments, they'd use those transistors for 2 extra cores instead. Really hoping Zen offers some competition and encourages Intel to release 6 or even 8 core mid-range CPUs on modern chipsets.

 

[ShintaiDK]

Yeah, look up some die shots of the 6700K:
http://abload.de/img/unbenanntopuco.png
The integrated GPU takes up a huge chunk of the total space. It's silly to pay for a large, integrated GPU that will never get used. Very few consumers who buy a 6700K use it with the Intel IGP. If Intel cared at all about the mid-range and high-end segments, they'd use those transistors for 2 extra cores instead. Really hoping Zen offers some competition and encourages Intel to release 6 or even 8 core mid-range CPUs on modern chipsets.

There isn't economics for it. Also clocks would be lower on a 6700K. How much? Around 500mhz lower if not more.

If you want 6 cores, buy HEDT.

 

[coercitiv]

The integrated GPU takes up a huge chunk of the total space. It's silly to pay for a large, integrated GPU that will never get used.

You know how mainstream desktop SKUs would look like without the iGPU?
Same number of cores as now, same prices as now, smaller dies.

Keep in mind the following 2 factors:

1. Even with the massive increase in GPU transistor budget, new mainstream chips are still tiny in comparison to the first chips with iGPU on die. (Skylake 4+2 is 122mm^2, Sandy Bridge 4+2 was 216mm^2)
2. Mainstream 4 core CPUs are still built with 90W+ TDP in mind. They still give all the performance possible within that power envelope.

When you buy a CPU nowadays, you're not paying for die area, you're paying for performance. If you don't believe that, compare Core M with Atom at 14nm (98mm^2 versus 87mm^2, difference in price is one order of magnitude).

Adding two more cores to mainstream does not require giving up the iGPU. Moreover, have you given any thought of what would happen to R&D and release schedule when Intel would design a new pure CPU for desktop, while still making the iGPU ones for mobile? If you're annoyed now, imagine how outraged you would be with 6700K "Pure" being 6 months late and just as expensive, because Intel would still make the mobile parts first. (quads with iGPU)

Can Intel lower prices and/or offer more? Sure, I can agree with that. But it's not the iGPU stopping them.

 

[NTMBK]

When you buy a CPU nowadays, you're not paying for die area, you're paying for performance. If you don't believe that, compare Core M with Atom at 14nm (98mm^2 versus 87mm^2, difference in price is one order of magnitude).

Core M also has the additional chipset die. And those parts probably have much lower yields than Atom- or rather, the parts that don't make Core M voltage get reused as higher voltage laptop parts.

 

[coercitiv]

Core M also has the additional chipset die. And those parts probably have much lower yields than Atom- or rather, the parts that don't make Core M voltage get reused as higher voltage laptop parts.

I used Core M as example, you can take Skylake U chips and the price difference is still there. In fact, if I really wanted to swing things my way, I would have used the Broadwell 4+2 core which is 82mm^2.

I'm definitely willing to pay Atom price for Broadwell 4+2 if it means I also have to pay for the chipset separately. But it doesn't work that way, I have to pay for Broadwell performance.

 

[Sheep221]

Yeah, look up some die shots of the 6700K:
http://abload.de/img/unbenanntopuco.png
The integrated GPU takes up a huge chunk of the total space. It's silly to pay for a large, integrated GPU that will never get used. Just having it there, even if disabled, probably impacts temperature, power and overclocking potential as well. Very few consumers who buy a 6700K use it with the Intel IGP. If Intel cared at all about the mid-range and high-end segments, they'd use those transistors for 2 extra cores instead. Really hoping Zen offers some competition and encourages Intel to release 6 or even 8 core mid-range CPUs on modern chipsets.

Processor graphics is one of the best inventions so far. Ideal as a backup once your video card malfunctions, is also good for playing older games, which many users still play and is provided at no additional cost. It's also cheaper to make CPUs one of which is same but unlocked rather than entirely separate design where iGP is replaced by computational cores.

You also provide no evidence whatsoever to backup your statement that disabled/unused iGP increases temperature and decreases overclocking abilities.
If iGP is unused at all, the remaining die space also gets abit warm from the main cores and this increased area causes heat to transfer much faster to IHS and cooler, temps won't be much different but I'm sure this also slightly benefits the overclocking.

 

[VirtualLarry]

Processor graphics is one of the best inventions so far.

I don't know if I would go that far. Sure, overall, in terms of BOM cost, integration is a good thing generally, and if OEMs can get away with building a system without a discrete GPU, then that's good for their bottom-line.

But as consumers, sometimes you get stuck with lackluster graphics, and as far as CPU costs versus performance goes, the R&D money spent on improving the iGPU cores ("EU"s, in Intel-speak) and the die area, could have gone towards improving the performance and number of actual computational CPU cores.

Sure, in a perfect world, we could use the iGPU processing elements in mainstream software. But has that happened yet? AMD launched a bold plan, with their HSA initiative, but that seems to be treading water lately, along with most of the rest of their company, until they can get Zen out the door, and see some significant profit again.

Edit to add: I've experimented with using the iGPU on a Skylake i3-6100, the HD 530, to do computation. Using BOINC, and attaching to the project "Collatz Conjecture", and logging into the account on the web site and configuring it correctly (to allow usage of Intel iGPUs), then you actually CAN utilize the compute features of the Skylake iGPU.

Only problem, and I'm not sure if this has to do with the limited number of compute elements in the HD 530, or what, but it bogs the entire OS UI down horribly. Clicking things takes like two seconds to respond.

I've successfully used GPU compute on my 7950 3GB GPU, on my other Skylake rigs, and it doesn't slow down the UI any noticeable amount at all. Scrolling still smooth, UI responsive, can still listen to internet radio though the card's HDMI audio output.

Whether this is simply down to the sheer number of compute units available on the 7950, or down to the app utilizing them (leaving some free for the UI), or even AMD's "async compute" feature, in which they can thread compute with graphics, I don't know.

All I know is, Intel is functional for GPU compute, on modern iGPUs (Haswell and especially Skylake), but it sucks, in practice. AMD APUs, and dGPUs, fare much, much better.

 

[Magic Carpet]

Processor graphics is one of the best inventions so far.

I quite agree. Processor graphics has never failed on me. Ever. Unlike discrete GPUs... (just way too many areas to fail). Can't wait for Intel/AMD better iGPU!!!!

 

[frozentundra123456]

There isn't economics for it. Also clocks would be lower on a 6700K. How much? Around 500mhz lower if not more.

If you want 6 cores, buy HEDT.

I guess not when you can like I said, sell server rejects for high prices, along with an expensive motherboard.

But if there is no economics for more than 4 cores, why is AMD bringing out their flagship new architecture as an 8 core chip?

 

[coercitiv]

as far as CPU costs versus performance goes, the R&D money spent on improving the iGPU cores ("EU"s, in Intel-speak) and the die area, could have gone towards improving the performance and number of actual computational CPU cores.

Are you saying CPU R&D at Intel is not funded properly?

 

[VirtualLarry]

Are you saying CPU R&D at Intel is not funded properly?

I'm just saying, in theory, if your overall R&D budget is for both X (cpu cores) and Y (iGPU cores), then if you cut out Y, you should have more money available for X.

 

[ShintaiDK]

I guess not when you can like I said, sell server rejects for high prices, along with an expensive motherboard.

But if there is no economics for more than 4 cores, why is AMD bringing out their flagship new architecture as an 8 core chip?

For the same reason they did with construction cores. AMD 8 cores is their HEDT. A staggering failure so far and Zen doesn't look like its going to change anything.

 

[poofyhairguy]

Processor graphics is one of the best inventions so far.

Meh.

Intel's GPUs

1. Aren't fast enough to matter

2. Are mismatched (aka a $400 CPU that will be paired with a 980 ti has the best GPU, not some celeron customer that is actually running games on integrated graphics)

They exist at the level they do because of Apple and retina screens.

 

[frozentundra123456]

Man, Shintai, I am not going to get into this anymore with you, I still dont understand why you so obsessively defend intel, but we are going to have to agree to disagree I guess.

I just have to say though, calling Zen "HEDT" is devious at best, because it is on the mainstream platform, while intel "HEDT" is not. So while they may be intended for the same market ultimately, they are far from an analogous market position.

 

[ShintaiDK]

Its you being obsessed with an entitlement claim in the shape of a product and price you wish.

 

[Sheep221]

Meh.

Intel's GPUs

1. Aren't fast enough to matter

2. Are mismatched (aka a $400 CPU that will be paired with a 980 ti has the best GPU, not some celeron customer that is actually running games on integrated graphics)

They exist at the level they do because of Apple and retina screens.

I know but it's still far better than Intel GMA IGPs on found on 775 motherboards and laptops, that's what I was trying to say is that it's useful for many things just not hardcore gaming or video rendering. For office stuff, playing old games, movies and so on it's very much fine.

 

[myocardia]

I know but it's still far better than Intel GMA IGPs on found on 775 motherboards and laptops, that's what I was trying to say is that it's useful for many things just not hardcore gaming or video rendering. For office stuff, playing old games, movies and so on it's very much fine.

Indeed. I have a Haswell i3 4160-based computer, and I was surprised at just how competent it happens to be for gaming @ 1080P. Don't get me wrong, I'd hate for it to be my primary gaming setup, but compared to what Intel was 'serving' up 15 years ago in the same price range, it's awesome.

 

[RussianSensation]

For the same reason they did with construction cores. AMD 8 cores is their HEDT. A staggering failure so far and Zen doesn't look like its going to change anything.

Do you have a source or statistics that HDET is a staggering failure? By definition, i7 3930K->i7 5960X are targeted at enthusiasts. Even if all of these CPUs represent just 3-5% of Intel's market share, doesn't mean they are a failure. That's like claiming that GTX280->GTX980Ti are all failures since <5% of the entire dGPU market buys them.

Very few people buy a Porsche 918, McLaren P1, and LaFerrari. Are those failures too since 99.9% of people who don't own those cars?

The 2 primary reasons Intel still hasn't bright 6-core to the mainstream are lack of competition and people keep paying $350/350 EUR for a glorified HT i5. It's very simple. As soon as there is competition in any viable form, Intel will bring a 6-core Icelake/Tigerlake to the mainstream. It's only a matter of when not if of a mainstream 6-core Intel CPU. With $389 i7 5820K and i7 6800/6820K, the price premium over Z170/270 and i7 6700/7700K is minimal. More and more people will move to HDET. Just because the average Joe noob who buys Lenovos, HPs, Dells, and Apples of the world will keep buying weak CPUs like Core-Ms, i3 and i5s, there is a growing number of PC enthusiasts who want and will buy 6-10 cores. With Zen, next gen PS5/XB2 likely going AMD again, the future is more cores. It's inevitable since IPC and clock speed gains are slowing down. It means the only other way to get more performance is spread the load across kofe threads. With DX12, it will be easier than ever once it matures and developers learn to load 6-8 cores well.

 

[ShintaiDK]

Do you have a source or statistics that HDET is a staggering failure? By definition, i7 3930K->i7 5960X are targeted at enthusiasts. Even if all of these CPUs represent just 3-5% of Intel's market share, doesn't mean they are a failure. That's like claiming that GTX280->GTX980Ti are all failures since <5% of the entire dGPU market buys them.

Very few people buy a Porsche 918, McLaren P1, and LaFerrari. Are those failures too since 99.9% of people who don't own those cars?

I wasn't talking about Intels HEDT line.

There is no way HEDT represent 3-5% of Intels volume either. You are going to be lucky if it even reach 1%. Its a product that cant exist without Xeons for the same reason.

 

[dmens]

Consumers want higher frequency less cores, that's what works best for desktop computing, games, simulations. Because this doesn't fit your marketing strategy isn't a justification to NOT provide what consumers really want ... more die space, higher frequency.

Consumers want good single thread performance and if there has been any effort to sabotage that aspect of performance to cram more cores into the available die space, I am not aware of it.

Single thread performance is also the most difficult thing to get speedup these days so maybe your expectations are out of whack.

 

[Cogman]

Few things are going on here.

First off, All of the easy architecture changes that can be made have pretty much been made. This is what is currently killing intel's performance increase per generation.

Next, the transistor energy math doesn't work out any more with die shrinks. We are no longer seeing power per transistor drop substantially per node shrink like we used to in the good ole days. That means, intel can't solve problems and increase performance by simply throwing more transistors at the problem.

And finally, Intel neglected the mobile market and it is desperately trying to stay relevant there for consumer goods. Probably the biggest concern Intel would have had was when microsoft made Windows RT (even though it flopped). Intel still has big problems if people decide that "You know what? Android makes a pretty good desktop. And look, I can turn my phone into a desktop by plugging in this one port".

Dedicated desktops are on the verge of extinction. The fact is, mobile CPUs have nearly caught up to their desktop counterparts in terms of performance (and most consumers don't care that much). Further, the advent of eGPUs could really take a big bite out of the market share as it would make it even more reasonable to just have a mobile device that plugs into an external GPU when needed.

Intel is trying to stay relevant in the consumer market. They have found that consumers want mobility so that is why they are focusing so heavily on it. The last thing they want is for people to start using non-intel cpus for their mobile devices with their mobile OSes, they would have done that by now had intel not focused so heavily on mobility (Remember the days when you got 2 hours of battery life out of a laptop and thought that was good?).

When it comes to performance, intel is in a pickle. About the only way forward for them is a complete uArch redesign. I doubt they will do that for a few years (they are just going to keep givings us ever diminishing performance percentage bumps). However, once ARM CPUs get to the point where they are starting to be competitive in performance to an Intel CPU, then we will see them do some pretty drastic things. AMD is so far behind and struggling so much that I doubt they will be able to push Intel ever again. They are in an even worse bind than Intel is right now. Nobody wants their CPUs in almost any market except for the low end consumer desktop market and that market is quickly drying up. AMD is really far behind when it comes to mobile.

 

[Sheep221]

Do you have a source or statistics that HDET is a staggering failure? By definition, i7 3930K->i7 5960X are targeted at enthusiasts. Even if all of these CPUs represent just 3-5% of Intel's market share, doesn't mean they are a failure. That's like claiming that GTX280->GTX980Ti are all failures since <5% of the entire dGPU market buys them.

Very few people buy a Porsche 918, McLaren P1, and LaFerrari. Are those failures too since 99.9% of people who don't own those cars?

The 2 primary reasons Intel still hasn't bright 6-core to the mainstream are lack of competition and people keep paying $350/350 EUR for a glorified HT i5. It's very simple. As soon as there is competition in any viable form, Intel will bring a 6-core Icelake/Tigerlake to the mainstream. It's only a matter of when not if of a mainstream 6-core Intel CPU. With $389 i7 5820K and i7 6800/6820K, the price premium over Z170/270 and i7 6700/7700K is minimal. More and more people will move to HDET. Just because the average Joe noob who buys Lenovos, HPs, Dells, and Apples of the world will keep buying weak CPUs like Core-Ms, i3 and i5s, there is a growing number of PC enthusiasts who want and will buy 6-10 cores. With Zen, next gen PS5/XB2 likely going AMD again, the future is more cores. It's inevitable since IPC and clock speed gains are slowing down. It means the only other way to get more performance is spread the load across kofe threads. With DX12, it will be easier than ever once it matures and developers learn to load 6-8 cores well.

1. i3 and i5 are VERY powerful CPUs
2. Owning LGA 2011 i7 or Xeon does not define someone as enthusiast, nor it means that after a purchase of this you become one, nor it means you are not one if you don't have such a CPU.
3. As mentioned before several times in this thread, no mainstream hex and oct cores are not result of competition or lack of it.