Skylake Core Configs and TDPs

[Enigmoid]

Just to add another graph,
Total GPUs shipped in 2013 are more than the combined total shipments of Desktop and Laptop PCs.
We can also see that both Desktop and Laptop shipments are decreasing from 2010 onwards and that makes dGPUs shipments also decreasing.

How do you sell that many more desktop PCs than desktop GPUs?

 

[24601]

How do you sell that many more desktop PCs than desktop GPUs?

End User Upgrades.

End User Built PCs .

Double Counting iGP and dGP since everything has an iGP for a while now.

Etc.

 

[mrmt]

I agree with the only high-end not being feasible point.

But I actually thought Nvidia would go down first.
If IGPs are the future, then AMD's GPU division would still live.
Their APUs can match Intel's IGP performance, they just can't sell enough because of lackluster CPU performance.
If AMD could get its CPU IPC competitive again then it has a better chance than Nvidia because Nvidia doesn't make any APUs, so they have nowhere to go.

Both players already agreed that the dGPU is a dead end. They aren't investing more funds on graphics, but trying to develop new business lines around other technologies and using their GPU IP as a competitive advantage and even on the desktop they are trying to make their GPU chips something more of a co-processor, and not a dedicated graphics processing unit anymore.

That said, if both business were doomed to failure, AMD's would go first. Nvidia has a healthy and profitable professional market, while AMD has a a couple of embedded contracts that gets them peanuts profits.

 

[Enigmoid]

End User Upgrades.

End User Built PCs .

Double Counting iGP and dGP since everything has an iGP for a while now.

Etc.

The number of user build PCs is small and end user upgrades is miniscule. Counting igps and dgpu twice is disingenuous to say the least. When intel ships every piece of consumer silicon (excepting E series and server SKUs) with an igp on die, the physical chip which you can not remove, counting it as a separate GPU shipment is pretty silly.

Trying to put "total graphics" shipments on that chart is silly and a misuse of statistics.

 

[Fjodor2001]

This is why they make multiple different GPU sizes. On Haswell they already have GT1, 2 and 3, with GT3 being 4 times larger than GT1. And Skylake brings in GT4!

The problem is that you cannot choose CPU performance and iGPU performance freely with the current Intel models. If you get a top end CPU (3770/4770-K) you have to get the highest performance iGPU model along with that.

They could of course change that policy, so you freely can select to get the Skylake model that will correspond to 4770K with either GT1, GT2, GT3 or GT4. And then you should be able to make the same iGPU choice for all other CPU models. But just think about the number of CPU models Intel would have to maintain then. It will not be economically possible to keep that many versions in production and in stock.

So in the end, they'll have to settle for some "reasonably high iGPU performance that correspond roughly to the CPU performance" for each CPU model. And that will never be even close to top end dGPU performance levels, since for most users that is unnecessary and would hence be wasted silicon.

 

[NTMBK]

The problem is that you cannot choose CPU performance and iGPU performance freely with the current Intel models. If you get a top end CPU (3770/4770-K) you have to get the highest performance iGPU model along with that.

They could of course change that policy, so you freely can select to get the Skylake model that will correspond to 4770K with either GT1, GT2, GT3 or GT4. And then you should be able to make the same iGPU choice for all other CPU models. But just think about the number of CPU models Intel would have to maintain then. It will not be economically possible to keep that many versions in production and in stock.

So in the end, they'll have to settle for some "reasonably high iGPU performance that correspond roughly to the CPU performance" for each CPU model. And that will never be even close to top end dGPU performance levels, since for most users that is unnecessary and would hence be wasted silicon.

Not so. 4770k is only GT2, for GT3 you need 4770R. And look a the Ivy Bridge model. Intel offered both i3-3220 and i3-3225- identical CPU performance, but one had GT1 and the other had GT2.

 

[Fjodor2001]

Not so. 4770k is only GT2, for GT3 you need 4770R. And look a the Ivy Bridge model. Intel offered both i3-3220 and i3-3225- identical CPU performance, but one had GT1 and the other had GT2.

Ok, so please let us know what Intel CPU has an iGPU that matches a GTX 780.

 

[ShintaiDK]

Ok, so please let us know what Intel CPU has an iGPU that matches a GTX 780.

Why would they need that? nVidia is not going to sell GTX780 only. Then they are out of business long time ago.

 

[Fjodor2001]

Why would they need that? nVidia is not going to sell GTX780 only. Then they are out of business long time ago.

The point is that Intel does not have any CPU with an iGPU that even comes close to the higher end dGPUs. And there are economic reasons for that, which I mentioned before (wasted silicon for the average user ...). So things will stay that way, and there will be a market for dGPUs going forward too.

The shift from lower end dGPUs to iGPUs made sense, because there was still a need for the average computer user to get better GPU performance so they might as well "bundle" it with the CPU. But that shift has already happened. It will however not happen for high end dGPUs since most users don't need that kind of GPU performance.

 

[NTMBK]

That is why most users will get a smaller die with less GPU on it. Intel doesn't just produce one die and fuse parts of it off any more, they produce about 5 different Haswell dies with different core counts and GPU sizes.

 

[ShintaiDK]

The point is that Intel does not have any CPU with an iGPU that even comes close to the higher end dGPUs. And there are economic reasons for that, which I mentioned before (wasted silicon for the average user ...). So things will stay that way, and there will be a market for dGPUs going forward too.

The shift from lower end dGPUs to iGPUs made sense, because there was still a need for the average computer user to get better GPU performance so they might as well "bundle" it with the CPU. But that shift has already happened. It will however not happen for high end dGPUs since most users don't need that kind of GPU performance.

When there is no ROI left in dGPUs their development will stop, while IGP will continue to progress. And AMD and nVidia cant just develop and sell highend GPUs only and make a profit.

Also IGP development goes a lot faster than dGPU already. Not to mention the node progress.

Then it doesnt matter what some people on this forum wants or not. The only thing that matters is what is profitable.

 

[Insert_Nickname]

The shift from lower end dGPUs to iGPUs made sense, because there was still a need for the average computer user to get better GPU performance so they might as well "bundle" it with the CPU. But that shift has already happened. It will however not happen for high end dGPUs since most users don't need that kind of GPU performance.

I'd go so far as to say that shift happened 12 years ago with the Intel "Extreme Graphics". They weren't extreme (except how extremely bad they where), but they did provide a low-cost way to put a picture on screen. Lets face it, the average PC user just needs something to do that and accelerate video. Nothing more and nothing less.

Discrete has been more-or-less a niche market since, and as far as I can see that's not going to change any time soon.

When there is no ROI left in dGPUs their development will stop, while IGP will continue to progress. And AMD and nVidia cant just develop and sell highend GPUs only and make a profit.

The IP for the IGPs still have to be developed. That development is expensive, and dGPUs provide an additional way to capitalize on that IP. Why not do that?

 

[mrmt]

When there is no ROI left in dGPUs their development will stop, while IGP will continue to progress. And AMD and nVidia cant just develop and sell highend GPUs only and make a profit.

Both AMD and Nvidia already have iGPUs. NVidia was so keen to get Tegra to work not only for new profits streams, but to get more volume and spread R&D costs; Tegra on cars won't get much profits, but will spread R&D costs. Same with AMD, they are trying to develop an embedded business that will need GPU IP, and that will help them to spread R&D costs.

 

[ShintaiDK]

Both AMD and Nvidia already have iGPUs. NVidia was so keen to get Tegra to work not only for new profits streams, but to get more volume and spread R&D costs; Tegra on cars won't get much profits, but will spread R&D costs. Same with AMD, they are trying to develop an embedded business that will need GPU IP, and that will help them to spread R&D costs.

The problem is neither of them is having any success in that. Both of them are having MPU divisions in free fall.

 

[mrmt]

The problem is neither of them is having any success in that. Both of them are having MPU divisions in free fall.

They don't need the profits, they need the volumes.

 

[ShintaiDK]

They don't need the profits, they need the volumes.

The revenue stream for both is declining fast.

 

[Fjodor2001]

When there is no ROI left in dGPUs their development will stop,

As Insert_Nickname also concluded the shift to iGPUs for low end dGPU -> iGPU already happened years ago, and dGPUs are still around.

The same people buying dGPUs now will be the same ones buying them 5-10+ years from now. They need higher performance and will not settle for iGPUs. There are sufficient number of people like that to sustain a healthy dGPU market with ROI to be made.

 

[ShintaiDK]

As Insert_Nickname also concluded the shift to iGPUs for low end dGPU -> iGPU already happened years ago, and dGPUs are still around.

The same people buying dGPUs now will be the same ones buying them 5-10+ years from now. They need higher performance and will not settle for iGPUs. There are sufficient number of people like that to sustain a healthy dGPU market with ROI to be made.

The entire dGPU market in 2013 was 65 million units, it havent been this low in this century. And that number continues to decline. But again we already have 3 year cycle for GPUs to offset cost.

 

[voodoo7817]

So if I get this straight, it doesn't appear that there will be any 6-core variants of Skylake (non-e), right? Or is it too early to tell? If it will only have 4 cores, I'm now leaning toward upgrading my current i7-860 to a 4790k rather than waiting another whole year for Skylake.

When I originally bought my 860 in late 2009, I was thinking that my next upgrade would have 6 cores and DDR4 ram, as without that I couldn't imagine there would be much of a performance increase. I know Haswell-E has both, but I don't think that platform will offer great price/performance for me, as I mostly game and platform costs (motherboard and DDR4) will be expensive, at least initially.

My current setup has served me very well, bridging my 8800gt to my current 560ti, and I also added an SSD along the way. I think my setup could probably last another GPU upgrade, and I'm waiting on the next set of GPUs (Maxwell or whatever AMD releases) for that.

But if Skylake won't have 6 cores, maybe I should just upgrade to Devil's Canyon now. With CPU improvements seemingly dwindling over time (i.e. IPC and ability to hit higher clocks), that's perhaps another reason to make the jump sooner rather than later. Thoughts?

 

[AtenRa]

The entire dGPU market in 2013 was 65 million units, it havent been this low in this century. And that number continues to decline. But again we already have 3 year cycle for GPUs to offset cost.

Since you are talking about dGPUs you are talking about desktop and laptops.
Desktop and Laptop CPUs are already in a three year cadence and same goes for iGPUs.

Intel has just released Haswell refresh which have the same iGPU as last year.
Same in Laptops, Intel released Haswell refresh one/two months ago and that will be the main Laptop CPU and iGPU for the next 12 months (2014 and early 2015).

Intel iGPUs are not progressing faster than Desktop dGPUs after all

Also, Both AMD and NVIDIA use their GPU IP in different markets (Consoles, Automotive etc) to spread R&D and raise volumes. GCN 1 and 1.2 have been used in Desktop/Laptop dGPUs, in Desktop/Laptop APUs, in Consoles, in Embedded etc etc. GCN is already more than 2 years in the market in various segments and products. Just because Desktop/Laptop dGPUs are decreasing in volume doesnt mean they are going to be extinct.

 

[jpiniero]

So if I get this straight, it doesn't appear that there will be any 6-core variants of Skylake (non-e), right? Or is it too early to tell? If it will only have 4 cores, I'm now leaning toward upgrading my current i7-860 to a 4790k rather than waiting another whole year for Skylake.

Yes, it is only 4 cores max. Cannonlake might have more but I would deem it as unlikely.

But if Skylake won't have 6 cores, maybe I should just upgrade to Devil's Canyon now. With CPU improvements seemingly dwindling over time (i.e. IPC and ability to hit higher clocks), that's perhaps another reason to make the jump sooner rather than later. Thoughts?

If the only reason is games, I see no reason to upgrade. The 860 is going to be more than enough for several more years, esp if you are getting a 4 Ghz overclock.

 

[frozentundra123456]

So if I get this straight, it doesn't appear that there will be any 6-core variants of Skylake (non-e), right? Or is it too early to tell? If it will only have 4 cores, I'm now leaning toward upgrading my current i7-860 to a 4790k rather than waiting another whole year for Skylake.

When I originally bought my 860 in late 2009, I was thinking that my next upgrade would have 6 cores and DDR4 ram, as without that I couldn't imagine there would be much of a performance increase. I know Haswell-E has both, but I don't think that platform will offer great price/performance for me, as I mostly game and platform costs (motherboard and DDR4) will be expensive, at least initially.

My current setup has served me very well, bridging my 8800gt to my current 560ti, and I also added an SSD along the way. I think my setup could probably last another GPU upgrade, and I'm waiting on the next set of GPUs (Maxwell or whatever AMD releases) for that.

But if Skylake won't have 6 cores, maybe I should just upgrade to Devil's Canyon now. With CPU improvements seemingly dwindling over time (i.e. IPC and ability to hit higher clocks), that's perhaps another reason to make the jump sooner rather than later. Thoughts?

Depends on your budget and if you want the absolute best performance. I think you should wait for haswell E to see the pricing and how well it overclocks. If the rumors of 400.00 haswell E six cores hold up (which I doubt), and if it over clocks well, it could be an attractive platform. This would also allow us to see a few more next Gen games and see if the high CPU usage like watchdogs is the coming trend or just a poorly optimized outlier.

OTOH, a 4790k system will for sure give better bang for the buck, and be more than adequate for the foreseeable future.

 

[Insert_Nickname]

My current setup has served me very well, bridging my 8800gt to my current 560ti, and I also added an SSD along the way. I think my setup could probably last another GPU upgrade, and I'm waiting on the next set of GPUs (Maxwell or whatever AMD releases) for that.

But if Skylake won't have 6 cores, maybe I should just upgrade to Devil's Canyon now. With CPU improvements seemingly dwindling over time (i.e. IPC and ability to hit higher clocks), that's perhaps another reason to make the jump sooner rather than later. Thoughts?

That's a difficult question. Currently it doesn't look like Skylake is bringing much more then support for AVX 3.2, the actual CPU performance increase doesn't seem to be that much better then Haswell. Skylake appears to be much more focused on the IGP and power efficiency then CPU performance. Both of which are negligible from a desktop perspective.

If you feel the upgrade urge, I see no harm in going with Devils Canyon now, rather then wait another year for perhaps 5-10% more IPC. It also gets you to a current generation chipset, the 5x-series is a bit long in the tooth. It doesn't have native SATA3 f.x.

 

[ShintaiDK]

Intel has just released Haswell refresh which have the same iGPU as last year.
Same in Laptops, Intel released Haswell refresh one/two months ago and that will be the main Laptop CPU and iGPU for the next 12 months (2014 and early 2015).

Intel iGPUs are not progressing faster than Desktop dGPUs after all

And the last real uarch, if you exclude Maxwell on the GPU front was in early 2012..

As for IGP, there will be a new IGP uarch later this year. And another next year.

 

[AtenRa]

And the last real uarch, if you exclude Maxwell on the GPU front was in early 2012..

And ?? that only shows how far behind Intel iGPU architectures are.

As for IGP, there will be a new IGP uarch later this year. And another next year.

There will be no new arch in Desktop/Laptops in 2014, Broadwell Y will only be used in Tablets this year. Laptop Broadwell SKUs will be release in Q1 2015, more than 6 moths from now.
Broadwell Desktop SKUs with new iGPU arch will only be released one year from now, in Q2 2015.
20nm dGPUs will also be release around this time or even earlier. So, i dont see why people believe that Intel iGPUs will progress faster than dGPUs.