Skylake Core Configs and TDPs

[ShintaiDK]

Intel increase its GPU market share only because it is increasing its CPU market share over AMD, not because less dGPUs are sold. Since the market as a whole is growing, there's still money to be made for AMD and nVidia, even if Intel grabs a larger section of the complete combined dGPU/iGPU market share.

See this post for more details.

Lets try some more recent numbers instead of 2010:

A lot changed since 2010 as you use as latest numbers. Your numbers exclude the entire SB series and up. And while somewhat stagnant numbers from 1998 to 2010. Its gone downhill since:
http://jonpeddie.com/press-releases...-market-down-in-q1-nvidia-holds-market-share/

And IC design, R&D and wafer cost is only going up.

dGPU shipments are dropping fast.

 

[ShintaiDK]

What numbers are those ??

http://jonpeddie.com/press-releases...-market-down-in-q1-nvidia-holds-market-share/

I can see Intel increase its GPU market share the next years BUT only because they are flooding the Tablet market with ATOMs, not because they decrease the dGPUs in x86 Windows products.
How many times people upgrade there dGPUs ??? more than they change CPUs. Even if you get a iGPU PC, a lot of people will install a new dGPU later on as an upgrade.
So a $100 dGPU in 2016 will be much faster than todays Intels HD5200 iGPU and that dGPU will be a nice upgrade for that user.

Even in Laptops, dGPUs will always provide more performance for gaming. People buying laptops with dGPUs today will continue to buy them in the future.

So no, dGPUs are not dying

Intel rapidly increased its share in 2013 for example and dGPU shipments drop fast. Like it or not, dGPUs are dying.

There is a reason why dGPU makers are desperately trying to enter other segments. They already saw the writing on the wall.

We have been tracking AIB shipments quarterly since 1987—the volume of those boards peaked in 1999, reaching 114 million units, in 2013 65 million shipped.

Thats what today, 1 of 6 CPUs get paired with a dGPU? And if people upgrade, that number will be even lower.

 

[Erenhardt]

Seems like a prime time for kaveri then.

 

[frozentundra123456]

Well, the discrete gpu market is not going to stand still either. Maxwell already offers about twice the performance per watt, and will greatly improve on 20 nm. I still think for the high end and gaming, there will be sufficient market for discrete. Just look at nvidias workstation cards. They have a huge margin there to help fund development of new cards. Actually, I see amd in more danger for discrete, since they don't have as strong a presence in the workstation market. Even if workstation priced cards go up, l think the users would just bite the bullet and pay it if the performance warrants it.

I also think think igps will have a hard time reaching the performance some are speculating about, and who knows how delayed 10nm and below will be. There are also the thermal and bandwidth problems to be overcome. Even if ddr4 or stacked memory overcomes the bandwidth problem, you have thermal issues. I question how Intel will be able to cram 2 or 4 times the EU in a CPU die and effectively cool it.

 

[ShintaiDK]

Seems like a prime time for kaveri then.

But its performance metrics are so poor that people select Core based CPUs instead.

 

[ShintaiDK]

Well, the discrete gpu market is not going to stand still either. Maxwell already offers about twice the performance per watt, and will greatly improve on 20 nm. I still think for the high end and gaming, there will be sufficient market for discrete. Just look at nvidias workstation cards. They have a huge margin there to help fund development of new cards. Actually, I see amd in more danger for discrete, since they don't have as strong a presence in the workstation market. Even if workstation priced cards go up, l think the users would just bite the bullet and pay it if the performance warrants it.

I also think think igps will have a hard time reaching the performance some are speculating about, and who knows how delayed 10nm and below will be. There are also the thermal and bandwidth problems to be overcome. Even if ddr4 or stacked memory overcomes the bandwidth problem, you have thermal issues. I question how Intel will be able to cram 2 or 4 times the EU in a CPU die and effectively cool it.

Its obvious that AMD gets thrown of the train before nVidia. Not just due to workstation and HPC segment. But also the twice as high volume. And AMDs MPU division is in free fall too and cant offset anything there for the GPU R&D.

However, remember IGPs dont need to be as fast as dGPUs to kill them. They only need to make dGPUs unprofitable. And dGPU volume is shrinking, while IC design, R&D and node cost goes up. Broadcom for example as an ARM MPU designer already jumped of the 20nm wagon.

dGPU shipments are already low. And Skylake+Broadwell with more GTe models. Not to mention Skylakes GT4e models will erode that volume even further.

 

[frozentundra123456]

Exactly. Even with all the graphics expertise of amd, their apu cannot match the performance of a more than 2 year old low end hd7750, due to the inherent limitations of an igp. Despite their process lead, Intel has much less expertise in graphics, so I fail to see how their igps will even match a mid range discrete card. They may be able to throw enough transistors at the problem to catch or surpass amd's apus , but that is a far cry from matching a mid or high end discrete card.

 

[ShintaiDK]

Forget highend and enthusiast. If midrange gets eaten by IGPs its long past the unprofitability border for dGPUs.

 

[frozentundra123456]

Its obvious that AMD gets thrown of the train before nVidia. Not just due to workstation and HPC segment. But also the twice as high volume. And AMDs MPU division is in free fall too and cant offset anything there for the GPU R&D.

However, remember IGPs dont need to be as fast as dGPUs to kill them. They only need to make dGPUs unprofitable. And dGPU volume is shrinking, while IC design, R&D and node cost goes up. Broadcom for example as an ARM MPU designer already jumped of the 20nm wagon.

dGPU shipments are already low. And Skylake+Broadwell with more GTe models. Not to mention Skylakes GT4e models will erode that volume even further.

Well, let's wait and see what kind of pricing and availability high end igp parts have on broad well and sky lake. So far every new generation from Intel has supposedly been the one with killer igp performance. In actual fact, mainstream igp performance has gone up only modestly, while the truly "high performance" igps have been limited to extremely expensive and very low volume models.

 

[witeken]

LOL
Someone needs a reality check. Last time I checked intels igps were good for browsing internet and not demanding office stuff.

Could you please show me numbers proving me wrong or point out errors in my calculation.

 

[rtsurfer]

I think you both forget the business model. Neither nVidia or AMD can make a profit only selling and developing highend cards. And AMD will exit it a lot sooner than nVidia for the same reason.

I wouldnt be surprised if Intel commands 75%+ graphics share in Q1 2016 for example with the much broader range of GTe products and GT4 models. And with rising IC design costs, wafer cost increase and Intels continual node lead. Its simply gonna be a point where dGPUs cant make money. Then it doesnt matter if dGPUs are still faster. Because there will be no new development. Its all about ROI. The only question left is, where is that point.

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.

Intel'Intel might be able to eliminate the $150 GPU market but that's it.
But the low-end is always the biggest part of a market, so Intel would really do a lot of damage. Could Nvidia and AMD survive from high-end parts only?


What does this have to do with Intel's ability to make a big part of the dGPU market obsolete?

As far as mobile is concerned, which where the main battle is, nobody gets a dGPU to browse the Internet, its for games.
Games tend to get more demanding with time.

What I was trying to say is tjat supposedly if Intel increases its iGPU performance by 15% every generation then they still shouldn't be able to catch Nvidia/AMD because GPUs have large performance increases between subsequent gens. Intel isn't chasing a standing target, its a moving target & unless they literally double their IGP performance every gen, dGPUs will still offer more performance.
If there were 5-7% increase between subsequent dGPU gens like how it is on CPU market, than Intel could eliminate them fast.

Let us consider a hypothetical situation.
Intel could be targeting gtx 770m performance.
They could be thinking that suppose the current Haswell IGP is 15% slower than the 770M. With Broadwell they can achieve a 30% higher performance in IGP. But then Nvidia releases Maxwell which is 30% faster than 770M.
We are back where we started.

If Intel is getting 40% faster every gen, then it would still take them 3 more generations to catch up.

Finally, TSMC needs to get their s*** together. If new nodes keep getting delayed, like how 20nm has been then, Intel might be able to eliminate dGPU faster. Then what is TSMC going to do, they loose both AMD/Nvidia GPU business, Apple can't compensate for that.

 

[Insert_Nickname]

Exactly. Even with all the graphics expertise of amd, their apu cannot match the performance of a more than 2 year old low end hd7750, due to the inherent limitations of an igp. Despite their process lead, Intel has much less expertise in graphics, so I fail to see how their igps will even match a mid range discrete card. They may be able to throw enough transistors at the problem to catch or surpass amd's apus , but that is a far cry from matching a mid or high end discrete card.

This^^

...even with Intel's manufacturing lead, the 28nm IGP of the 7850K easily trumps Intel's HD4600 and can even match Iris Pro HD5200.

Why is everyone assuming that discrete performance will be stationary? Lets see how that HD5200, or Skylake GT4e, will compare with a second generation Maxwell/Pirate Islands. Or Pascal for that matter. Besides there is a market outside the top-of-the-line products, Skylake GT4e is unlikely to come cheap. I'm not writing off discrete GPUs just yet.

Wake me when there is an IGP that do photo-realistic gaming at 60FPS at 4K, then we can talk.

 

[frozentundra123456]

This thread has gotten pretty off topic, actually. But I do agree that dgpus really need to step up node improvements since igps are becoming more competitive. Ironically, shrinks are coming faster in cpus but showing little performance increase, while being very slow in dgpus where they should show good improvement.

 

[witeken]

Exactly. Even with all the graphics expertise of amd, their apu cannot match the performance of a more than 2 year old low end hd7750, due to the inherent limitations of an igp. Despite their process lead, Intel has much less expertise in graphics, so I fail to see how their igps will even match a mid range discrete card. They may be able to throw enough transistors at the problem to catch or surpass amd's apus , but that is a far cry from matching a mid or high end discrete card.

But the point is that the gap is rapidly shrinking. In just a few years, we went from HD3000 (Sandy Bridge) with 12EUs and 130 GFLOPS at 1350MHz (H1'12) to a full-blown GT4e IGP with 128EUs and competitive Gen9 microarchitecture ('15).

 

[Erenhardt]

I like how superior amd igps are being dismissed by some posters as not good enough, yet the same posters praise worthless intel igps.

 

[witeken]

What I was trying to say is tjat supposedly if Intel increases its iGPU performance by 15% every generation then they still shouldn't be able to catch Nvidia/AMD because GPUs have large performance increases between subsequent gens. Intel isn't chasing a standing target, its a moving target & unless they literally double their IGP performance every gen, dGPUs will still offer more performance.
If there were 5-7% increase between subsequent dGPU gens like how it is on CPU market, than Intel could eliminate them fast.

Let us consider a hypothetical situation.
Intel could be targeting gtx 770m performance.
They could be thinking that suppose the current Haswell IGP is 15% slower than the 770M. With Broadwell they can achieve a 30% higher performance in IGP. But then Nvidia releases Maxwell which is 30% faster than 770M.
We are back where we started.

If Intel is getting 40% faster every gen, then it would still take them 3 more generations to catch up.

But CPUs are a lot different from GPUs. Increasing GPU performance is easy: just ass more EUs, CUs, SMMs. Adding more cores won't increase single threaded performance.

Intel is doing some things Nvidia and AMD can't to catch up. They're improving their outdated microarchitecture to dGPU levels. Making a bad architecture good is easier and higher rewarding than making a good architecture very good. And they're using their process advantage to brute force their way into performance territories that would otherwise have been impossible.

 

[Erenhardt]

Intel is doing some things Nvidia and AMD can't to catch up. They're improving their outdated microarchitecture to dGPU levels.

well.. technically geforce gt410 is dgpu.... as is radeon X800xt.
nv and amd can do it because they are already past that...

 

[witeken]

This^^

...even with Intel's manufacturing lead, the 28nm IGP of the 7850K easily trumps Intel's HD4600 and can even match Iris Pro HD5200.

They do have a big manufacturing lead, but they don't have a meaningful density lead. A10-7850K has a 245mm² die size versus i7 4770's 177mm². An i7 4770R has a much bigger CPU and 4MB more cache, it has a 260mm².

So the fact that a 7850K can match Iris Pro isn't surprising or good at all.

Why is everyone assuming that discrete performance will be stationary? Lets see how that HD5200, or Skylake GT4e, will compare with a second generation Maxwell/Pirate Islands. Or Pascal for that matter. Besides there is a market outside the top-of-the-line products, Skylake GT4e is unlikely to come cheap. I'm not writing off discrete GPUs just yet.

I'm not assuming that, but 3 or more TFLOPS isn't bad at all for an IGP, certainly not for Intel. That is pretty high-end and that won't change substantially within 1 year that it's low-end. Today Intel is competing for the low-end, within 2 years, Intel will be competing for the mid-end.

 

[witeken]

well.. technically geforce gt410 is dgpu.... as is radeon X800xt.
nv and amd can do it because they are already past that...

You know what I mean.

 

[ShintaiDK]

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.

Oh I fully agree. But you could also say the same about nVidia with their Tegra. However neither AMD or nVidia will ever reach that point vs competition. That train simply left long ago for both of them.

 

[Sweepr]

I like how superior amd igps are being dismissed by some posters as not good enough, yet the same posters praise worthless intel igps.

I like how you completely ignore the fact that Intel's IGP have improved a lot in a few years. While you may call Iris Pro worthless for desktops because of its BGA-only form and higher price, I bet many people would laugh if someone told them in 2011 (Llano vs SB days) that Intel's fastest IGP would match the competitor's performance in lots of games at lower power levels in 2-3 years. It may surprise you, but not everyone is a hardcore gamer or care about HSA-accelerated stuff. In 1-2 years Intel will introduce two new graphics architecture (Gen 8 and Gen 9), starting with Broadwell and the GT3e LGA SKUs and continuing with Skylake GT3e (even @ 15W U SKUs) and GT4e (45-95W SKUs), and that's on top of great CPU performance and CPU perf/watt, an area the competition still has to catch up. Like it or not, they are improving fast.

 

[blackened23]

I like how superior amd igps are being dismissed by some posters as not good enough, yet the same posters praise worthless intel igps.

Well, that's interesting because mobile core CPUs with Iris Pro is actually faster than any mobile APU than AMD has. That makes it more clear that intel is winning the integrated graphics battle for mobile.

Furthermore, OEMs want BALANCED products. Good graphics are nothing to write home about if the CPU portion is crap. Yes, CPU matters for mobile. Performance per watt matters. Battery life matters. Instead, what AMD gives you is decent graphics with crap CPU, or decent CPU with crap performance per watt and terrible battery life. By the time OEMs tweak clockspeeds to get decent battery life, they end up crippling the performance of the mobile APUs to an extent that the balanced performance equation gets even further skewed.

Sorry, AMD just doesn't have a balanced offering yet. Intel has an entire range of IGPs with their Halo iris pro being a great performer for mobile products - that is why it is being used in quite a few rMBP SKUs. It has superior performance per watt and battery life, and has superior CPU performance to any competing AMD offering. Really, AMD has been focusing on their graphics performance and it's pretty good but not amazing. But it's still far worse than discrete. And all of the OTHER metrics that MATTER for mobile products, AMD is not on intel's level yet.

So that leaves AMD relegated to the bargain bin crap 400$ subnotebooks with platter HDs, and low screen resolutions and poor battery life. This is not a desirable position. If AMD wants to win at high end mobile, they need more than just "okay" graphics performance. They need a balance of everything. And they just don't have that. So that leaves AMD at the extreme low end bargain bin while intel wins all of the lucrative high end designs.

 

[cytg111]

Lets try some more recent numbers instead of

mkay ..

I only believe in statistics that I doctored myself

- Winston S. Churchill

You can chatter all the IGP up for world dominance that youd like however there is at least one thing that bothers me with this line of propaganda reasoning and that is the watts. As resent events has shown (i am sure you're gonna play a spec card here), we can put a 500 watt graphics / computation monster in a pcie slot.. Just thinking out loud here, do we se intel producing a 500 watt cpu/igp in the near future? Cause that would be awsome.

 

[Qwertilot]

They're not after the 500w monsters with iGP

People don't buy things on pure performance numbers. They buy things in order to do given things. Once they're doing iGPU's which can do 'reasonable' gaming at 1900*1080 (750ti/270 levels for instance), that's a huge threat to the market for dGPUs.

It actually doesn't terribly matter how much better dGPUs get in the meantime. The consoles will keep the requirements for games comfortably under control for a good while, and iGPU driven machines will be a lot simpler, smaller etc. Probably cheaper as a result too.

They could double performance and it just wouldn't matter for even most people with decent interest in games. To survive they need the requirements to go up - making 4k/VR etc rather well timed

 

[Insert_Nickname]

They do have a big manufacturing lead, but they don't have a meaningful density lead. A10-7850K has a 245mm² die size versus i7 4770's 177mm². An i7 4770R has a much bigger CPU and 4MB more cache, it has a 260mm².

So the fact that a 7850K can match Iris Pro isn't surprising or good at all.

+ the 77mm2 L4 cache chip. Considering the 7850K is pretty close in performance while on a slightly bigger node, I think its doing pretty good. CPU performance could be a lot better of course, but this was strictly considering the IGP.

The 4770R is also a lot more expensive and worse, at least for desktop, its BGA only.

I'm not assuming that, but 3 or more TFLOPS isn't bad at all for an IGP, certainly not for Intel. That is pretty high-end and that won't change substantially within 1 year that it's low-end. Today Intel is competing for the low-end, within 2 years, Intel will be competing for the mid-end.

I simply don't think so. The extra L4 cache and bigger die will add to cost, so GT4e will be high-end only. Intel is currently selling a lot more relatively simple 2C+GT2/GT3. As these paired with a discrete GPUs are easily able to match or exceed even the highest-end GT4e, I don't think there is much to worry about. The HD5200 hasn't even exceeded the performance of Nvidia's lowest-end desktop GT640 yet. That's a 2 year old chip.

Its exciting times though...