[WCCF] Intel Skylake 2015 Platform Details Revealed

[blackened23]

Not true, since both use the same manufacturing process and same architecture they are progressing at the same speed at best.

Edit: Also, any 90mm2+ 28nm dGPU is faster than 22nm IRIS PRO. Same will continue to hold true in future products.

Why the hell are you still talking about desktop products? Jesus christ man. MOBILE. Ultrabooks. AIOs. Brix Pro type devices. Performance per watt. There are mobile dGPUs that are faster, but at the expense of performance per watt. Maxwell is very very promising as an architecture and right now it boasts incredible performance per watt, but at some point I do think intel will be able to converge with mobile dGPUs. When? I don't know. But at some point, there will be a convergence between iGPU and medium range mobile DGPU. And even if they don't, the performance per watt for a high end mobile dGPU will be far greater. As far as AMD goes, they're not even a real competitor in mobile dGPUs - their mobile dGPU performance per watt is a joke compared to even Kepler and especially maxwell on the mobile ultrabook front. Hell the mobile R9 is a rebadged pitcairn isn't it.....

This doesn't change that:

1) iGPU was designed for ultrabooks, macbooks, AIO, etc. NOT desktop.
2) all ultrabooks/macbook buyers are not gamers, therefore viewing everything through the lens of PC gaming, WRONG ANSWER.
3) despite #2, iGPU performance has made incredible strides. Iris Pro performs pretty incredibly for the given die size, even though it is expensive.
4) continued developments on the iGPU front will eventually be able to give us 1080p high detail gaming. Heck, with iris pro you can already game at 1080p although you will obviously have to turn settings down.

Given the level of improvements intel has made with iGPU, I don't know when , but at some point they will be competetive with the best mobile dGPUs. Not sure when, but they're progressing faster than anyone in terms of mobile graphics performance within a performance per watt and die size constraint - without sacrificing PPW or CPU mobile performance. Obviously you can get an alienware gaming laptop with 30 minutes of battery life and a mobile dGPU in SLI. But, the performance per watt suffers.

 

[AtenRa]

dGPUs are 1+ process node behind Intel.

Lowest 28nm dGPU (Cape Verde 123mm2) released on 15th February 2012.

Intel HD4000 with IvyBridge was released on the 23rd of April 2012, two months later.
Intel Iris Pro with Haswell was released in June 2013 and they are still more than 2x slower than Cape Verde.

If Intel catches up with ~Gen8, they can get more performance out of a given die area.

Broadwell and Intel Gen 8 Graphics will not even come close to HD7770 performance at 1080p close to 3 years later.

In 2015 dGPUs will transition to 20nm, dGPU performance will once again become way faster than iGPU. It will be 2012 once again.

Also, products like TK1 show that lower TDPs have higher efficiency.

Efficiency is good but performance is what we are talking about here.

BTW, he was talking about mobile GPUs.

Mobile dGPUs are also faster than iGPUs, so same applies here as with Desktop.

 

[AtenRa]

Why the hell are you still talking about desktop products? Jesus christ man.

Where did you see me talking about Desktop products when i was replying to your comment ??

What applies to Desktop applies to Laptops etc. They use the same Architectures and Processes in both Desktop and Mobile. Mobile Kaveri is not faster than 512 SP Mobile dGPU. Next year there will be 20nm Desktop and Mobile dGPUs, iGPUs will continue to be inferior in performance at the same die area.

Also, performance per Watt is only relevant when we are talking for the same performance. One product may have 2x times the performance per watt of another product but less than half its performance.

 

[witeken]

Lowest 28nm dGPU (Cape Verde 123mm2) released on 15th February 2012.

Intel HD4000 with IvyBridge was released on the 23rd of April 2012, two months later.
Intel Iris Pro with Haswell was released in June 2013 and they are still more than 2x slower than Cape Verde.

Cape Verde has a TDP of 80W. Haswell: 84W for the whole chip with processor. 2x slower? I don't want to search benchmarks, but according to Wikipedia, Iris pro has about the same FLOPS as your GPU, so there are probably bottlenecks beside the shaders, which could impact things like performance and efficiency.

Broadwell and Intel Gen 8 Graphics will not even come close to HD7770 performance at 1080p close to 3 years later.

In 2015 dGPUs will transition to 20nm, dGPU performance will once again become way faster than iGPU. It will be 2012 once again.

Efficiency is good but performance is what we are talking about here.

Mobile dGPUs are also faster than iGPUs, so same applies here as with Desktop.

You seem to know more about Gen8 then we do, could you tell us more about the architecture? Since your other conclusions depend on this unconfirmed statement, I wouldn't attach much value to them. Will have to wait till there are benchmarks to continue that discussion.

 

[Pheesh]

Not sure what exactly you are railing against here, this is standard supply-chain management type stuff that I'm discussing.

If that is tantamount to a "conspiracy theory" for you, then I'm guessing you don't have much experience with Intel's supply chain?

AMD does this as well, coordinating with OEMs regarding their launch timelines for their next gen products by market segment within the context of managing existing inventory at the OEMs.

Yes, thinking that Intel keeps delaying broadwell /14nm process all due to customers suddenly wishing to clear out additional inventory is tantamount to a conspiracy theory for me. Pardon my skepticism.
Yes there is coordination regarding launches to a degree, but you are essentially saying OEM wishes are to delay (which wouldn't even make sense in the critical mobile space) and are causing a 14nm push. This is not 10 or even 5 years ago in a stable desktop/laptop market with a captive consumer. The pressure is on Intel and all OEM's in the mobile sector to iterate as quickly as possible to gain market share. Why would Intel always be talking about the need to move faster in mobile and then turnaround and just arbitrarily delay it?

 

[Fjodor2001]

I don't know what is worst; Intel delaying 14 nm intentionally to clear out old inventory, or Intel having actual 14 nm process tech problems?

 

[tential]

Maybe you should sharpen your reading comprehension skills. The iGPU is designed for ultrabooks and super small form factors (such as the Brix Pro), AIOs, and portable ultrabooks. They were never meant to replace desktop GPUs. At some point, iGPUs will converge will mobile dGPUs because the pace of progress with iGPU is far faster than that of mobile dGPUs. But if you feel you can throw a GTX 780ti into a Brix Pro within the same size constraints as the Iris Pro die size contained within, well okay if you say so. You're just oblivious to the bigger picture for which the iGPU is designed. It was designed at the suggestion of Apple for their mobile macbooks, that is why intel invested into the iGPU originally. And they've made significant strides with their mobile graphics performance, in fact they have the fastest mobile graphics performance at this time. The fact that Iris Pro as a mobile form factor is trading blows with a desktop level 7850k? Mobile chip versus desktop chip, roughly same performance, trading blows. What does that imply in terms of how well intel is progressing in graphics performance?

But if you live in the desktop only world where AIOs, SFF, ultrabooks, macbooks, and mac mini/brix pro devices don't exist, yeah, a dGPU is obviously better. Thanks for reminding us again, no one said otherwise. But good luck fitting your dGPU into a mac mini or Brix Pro inside of that size constraint. That's why intel designed the iGPU. They didn't design it for your 3 foot tall ATX case, even though some SKUs have it for free. It was designed for mobile at the suggestion of Apple, and apple was willing to pay the big bucks for integrated graphics that did not compromise CPU performance or performance per watt.

This is completely beside the point that most people buying macbooks and ultrabooks aren't gamers at all. It's neat that you view everything and anything through the lens of gaming. I assure you, the bigger market for mobile macbooks and ultrabooks aren't 100% gamers that expect to play crysis 3 at 1080p/60 fps. And they aren't 100% desktop users.

As a huge fan of the ultra portable/ultrabook, I never would truly care about iGPU performance. If I want to game on my laptop, then I get a laptop with a mobile GPU. iGPU performance is good for things that the average user probably doesn't know it's doing.

Either way though, I find it highly amusing that AMD "Fanboys" continue to harp on GPU performance. It's a nice BONUS to have on your laptop to be able to utilize your GPU. Chances are, the average user who is picking up an Ultrabook isn't worrying about iGPU performance. At least it hasn't been a factor in any of the recent Ultrabook's I've seen picked up. At the end of the day, power consumption and the capability to do your basic laptop functions are far more important and AMD just isn't going to win at that.

To me, AMD's platform works best as a "Console fighter" but how big is the market for that? And is it profitable for them to compete against themselves?

 

[Ajay]

The answer is NEVER.

Say you get a Kaveri or IRIS PRO today, you have the fastest iGPU on the market. In 2 years you want to upgrade only the GPU. At the time, a low end ($100) GPU will be 2x times or more faster than your iGPU.

Also, $100 dGPUs will always provide more performance than any $100 APU. Even if Kaveri had GDDR-5, it would not even reach HD7750 level of performance because it would need to raise the iGPU frequency to 900-1000MHz. That would make the TDP to raise to more than 95W.

Well, AMD faces a series of challenges. As a consumer, I hope they succeed (new uArch should help* + licensing deal with Samsung). But as Homeles mentioned - AMD really needs to be first here (dGPU).

Intel probably needs to manufacture their on die DRAM @ 14nm or smaller to get the cost low enough (right now their 128MB LLC is 22nm). Though 14nm only gets them to ~256MB for an IRIS Pro part. 10nm should be pretty impressive, maybe they have enough thermal headroom to go to 1GB with a very wide bus.



* Keller came in in 2012. ~5 years for a new uArch means 2017/18 for a new start for AMD visa vi x86. If AMD keeps hanging on like they have been, its possible. AMD might have started a new uArch before Keller, but even if he liked the general direction, R&D has been pretty tight. Of course, I really have no idea what is actually happening, other than Keller being upbeat publicly.

 

[Pheesh]

I don't know what is worst; Intel delaying 14 nm intentionally to clear out old inventory, or Intel having actual 14 nm process tech problems?

Clearly the former, as if that was actually the case the board would give the CEO a boot. The latter is kind of expected when going to a new node.

 

[jpiniero]

Maybe a little of all of the above and more? It's not like Intel's desktop and higher end laptops have any competition, and I still think they are a little afraid of making Atom too good and kill off a good portion of Core sales even though that is their destiny.

* Keller came in in 2012. ~5 years for a new uArch means 2017/18 for a new start for AMD visa vi x86. If AMD keeps hanging on like they have been, its possible. AMD might have started a new uArch before Keller, but even if he liked the general direction, R&D has been pretty tight. Of course, I really have no idea what is actually happening, other than Keller being upbeat publicly.

I wouldn't really expect much out of this theoretical new x86 big core from AMD. The ARM K12 might end up being something, but we'll just have to wait and see.

 

[Homeles]

The answer is NEVER.

Never say never.

One of these days, I'll get around to graphing the gap between IGPs and dGPUs. I've started my own website, so that'll definitely be one of the articles I write for it.

As a huge fan of the ultra portable/ultrabook, I never would truly care about iGPU performance. If I want to game on my laptop, then I get a laptop with a mobile GPU. iGPU performance is good for things that the average user probably doesn't know it's doing.

I game on my IVB Ultrabook just fine. Not pretty, but it gets the job done. Not long from now, ultrabooks will be good enough for gaming at high resolutions at decent settings

 

[blackened23]

There will always be a few mobile dGPUs which are faster. But the obvious caveat is performance per watt and battery life drain. Yes, you can get a GTX 880M for your ultrabook. Rest assured the battery life will be acceptable, but will not be as good as an iGPU.

So that's the point. When iGPUs hit the performance per watt area at a performance level that is very high - there will be a convergence between low end mobile dGPUs and iGPU. Already the Iris Pro in mobile form trades blows with the desktop 7850k (a full blown LGA chip) and the GT650m. It's already happening, but this obviously doesn't allow full detail 1080p gaming yet. But you can still game 100% fine. You just can't max every slider in every game.

Intel is leapfrogging their own graphics performance every generation, so it will be interesting. Maxwell is an amazing architecture for mobile, so i'm curious to see how the upcoming Maxwell GPUs do for mobile later this year (or next) and how Broadwell mobile will do. Should be interesting.

I'm not saying iGPU will converge in the near term. But I do think it will happen. Yes, there will be faster mobile dGPUs. But they will not have the same level of performance per watt; performance per watt/battery life means everything for ultrabooks.

 

[Fjodor2001]

Clearly the former, as if that was actually the case the board would give the CEO a boot. The latter is kind of expected when going to a new node.

Even if the former would increase return on investment for the company? Why would they fire the CEO then?

 

[witeken]

Intel probably needs to manufacture their on die DRAM @ 14nm or smaller to get the cost low enough (right now their 128MB LLC is 22nm). Though 14nm only gets them to ~256MB for an IRIS Pro part. 10nm should be pretty impressive, maybe they have enough headroom to go to 1GB with a very wide bus.

According to AnandTech, 128MB is more than enough. Intel started with 32MB being necessary for games, doubled that to be future proof, and doubled that again just to be sure.

 

[Pheesh]

Even if the former would increase return on investment for the company? Why would they fire the CEO then?

I dunno, perhaps because he would be blatantly lying in public statements? If it was in the best interest of the shareholders then the CEO wouldn't need to make materially false statements to say it was something else. Can you imagine the scandal if something like that happened?

 

[podspi]

According to AnandTech, 128MB is more than enough. Intel started with 32MB being necessary for games, doubled that to be future proof, and doubled that again just to be sure.

I never understood that, since it isn't like the hardware itself is future-proof.

Maybe to make it easier to write drivers across gens?

 

[Revolution 11]

How will Skylake support PCI-E SSD booting? Devil Canyon says that it does this but we have no idea which drives will work or how seamless this support is.

Will PCI-E SSD booting on Skylake be as easy as SATA HDD booting has been for 20 years? I sure hope so.

 

[Homeles]

How will Skylake support PCI-E SSD booting? Devil Canyon says that it does this but we have no idea which drives will work or how seamless this support is.

Will PCI-E SSD booting on Skylake be as easy as SATA HDD booting has been for 20 years? I sure hope so.

I'd imagine the majority of the PCI-E boot issues will be worked out this year, and certainly by the time Skylake launches.

 

[Revolution 11]

Is the Skylake chipset have more SATA/USB/PCI ports than the Haswell/Broadwell chipset(s)? Will there still be PCI-E 2.0 lanes on the chipset itself or is Intel moving to all PCI-E 3.0 in time?

I/O connectivity and chipset features seems to be the biggest improvement in desktop CPU platforms recently. Which I don't mind at all.

 

[Fjodor2001]

I dunno, perhaps because he would be blatantly lying in public statements? If it was in the best interest of the shareholders then the CEO wouldn't need to make materially false statements to say it was something else. Can you imagine the scandal if something like that happened?

The thing though is that the CEO has made somewhat vague public statements, like "We have confidence the problem is fixed because we have data it is fixed" (Intel CEO regarding the 14 nm production issues). They can also quite easily make statements that sound like they have an obvious meaning when you hear them, but can be interpreted another way if you really think about it. So then they are not really making any false public statements. How about the famous "I did not have sexual relations with that woman" statement from a famous president for example.

 

[Ajay]

According to AnandTech, 128MB is more than enough. Intel started with 32MB being necessary for games, doubled that to be future proof, and doubled that again just to be sure.

I was looking at it from the perspective of main graphics memory (MCM, HBM). I think that by 10nm Intel could be in a position to do this, but it is really a guess. 1GB would clearly be low end GFX. The largest impediment for desktop graphics would be CPU price. The higher end Laptop CPUs are already more expensive, so that where we would probably see this technology first. An important enabling technology will probably be 450mm wafers, which will accelerate the decline of xtors/$. After that, I would expect a fast uptick in consumer grade CPUs with HBM. I expect we'll see compute AIBs and server CPUs with HBM sooner.

On a side note, I wonder if there is an opportunity here for AMD's APUs to add HBM sooner than Intel, as their CPU prices are lower. I don't see it helping margins much, but the improvement in GFX performance (as they already have a lead) could drive higher volume at least.

 

[Pheesh]

The thing though is that the CEO has made somewhat vague public statements, like "We have confidence the problem is fixed because we have data it is fixed" (Intel CEO regarding the 14 nm production issues). They can also quite easily make statements that sound like they have an obvious meaning when you hear them, but can be interpreted another way if you really think about it. So then they are not really making any false public statements. How about the famous "I did not have sexual relations with that woman" statement from a famous president for example.

Hah, yes there is some dancing that can happen, but at the end of the day it doesn't make any sense to delay advances in the mobile space and throw away any competitive edge when you're trying desparately to gain market share there (to the tune of subsidizing your BOM's just to stay competitive). OEM's in this space are iterating way too fast, particularly with ARM. All of Intel's comments about the tablet/mobile market are about how they need to move faster there.

 

[witeken]

I was looking trough an Intel presentation of 2011 about Tri-Gate, and what interested me was that multiple times a slide explicitly said that Intel has a 2 year cadence for going to smaller nodes, so that gives Intel's statement of volume production of 10nm in 2015 more credibility.

I was looking at it from the perspective of main graphics memory (MCM, HBM). I think that by 10nm Intel could be in a position to do this, but it is really a guess. 1GB would clearly be low end GFX.

I guess 1GB will be obsolete in 2016.

 

[AtenRa]

I was looking trough an Intel presentation of 2011 about Tri-Gate, and what interested me was that multiple times a slide explicitly said that Intel has a 2 year cadence for going to smaller nodes, so that gives Intel's statement of volume production of 10nm in 2015 more credibility.

You know very well that 14nm started production in Q1 2014. If they will manage to keep the 2 years cadence, then the earliest they will start production of 10nm will be Q1 2016.
Since 22nm and 14nm have taken more than 24 months each, i will say that 10nm will also slip beyond 24 months. I would even speculate that 10nm will not even start production in 2016 at all.

 

[Fjodor2001]

Hah, yes there is some dancing that can happen, but at the end of the day it doesn't make any sense to delay advances in the mobile space and throw away any competitive edge when you're trying desparately to gain market share there (to the tune of subsidizing your BOM's just to stay competitive). OEM's in this space are iterating way too fast, particularly with ARM. All of Intel's comments about the tablet/mobile market are about how they need to move faster there.

Yeah, for mobile I guess it does not make sense to intentionally delay products (since Intel is desperately trying to gain market share in that area), but for desktop it can make sense (since Intel has a close-to-monopoly situation in that area).