New Intel Core Processor Combines High-Performance CPU with Custom Discrete Graphics from AMD

[Axils]

In another news.. Nvidia teams up with AMD CPU... lol

 

[shad0w4life]

More AMD gpus out there isn't bad for them, get people more comfortable with their brand etc.

Wish stock went up more, seems really low considering push towards smaller tech

 

[IdiotInCharge]

So why is nvidia coming out with gsync hdr? Oh, that's right so it can compete with freesync 2.

To add HDR support. FreeSync2 is adding the same kind of buffer support that Nvidia put into G-Sync in order to keep input lag down.

FreeSync is better than nothing, but Nvidia fixed the whole problem before AMD even had a demo to show. Only with FreeSync2 will they potentially catch up to G-Sync, and it'll be more expensive than FreeSync to boot, cause they'll be following Nvidia's lead with tech to keep input lag down and a program to maintain the integrity of the brand.

 

[IdiotInCharge]

the way it looks is that the Open standards are starting to win out, 1/5th the price of gsync and Xbox, PS4, Laptops and desktop can use it while gsync is just some desktop computers, enthusiasts and consumers are two vastly different classes of consumers and 75% of market ain't gonna buy a 500$+ screen.

Nearly all of your examples use HDMI, which isn't open.

As for cuda\opencl - Performance is roughly the same and devs like opencl over cuda but Nvidia have poured tons of resources into Cuda thus Cuda got really big but Adobe is doing a switch and others are also following thus AMD\Intel seem to be winning that one but that one really remains to be seen.

Adobe already uses OpenCL- exceedingly poorly. I use their products.

And Nvidia put resources into CUDA because they pioneered GPU compute, and needed an API for their processors, because OpenCL didn't exist!

 

[oldmanbal]

Could practically fit this thing in a phablet, fkn unbelievable performance for it's footprint, and all the mobo space that it just cancelled the need for.

I would love to see this in something the size of a walkman (give me external power brick thing) that I could carry anywhere. Please god. . .make my dream come true. Find a way to bury the memory and an ssd in something about 3 times as thick as an ssd and I'd be first in line.

 

[IdiotInCharge]

The use case is all 15W notebooks will have a lot to gain from RR than sticking with Intel + HD-whatever-#, because Zen on low power leverage it's perf/w advantage for higher boost clocks. Without the clock speed advantage than Intel has on desktop, and a single CCX design, RR is superior to Intel's low-power Kabylake for raw performance.

So essentially it obsoletes (hardware perf wise) Intel from the Ultrabook or thin notebook category where price does not approach premium dGPU category, by offering higher CPU performance and much higher graphics performance.

Intel is gonna still be king in the premium Ultrabook and Notebook because of this solution they have now with AMD graphics on an MCM.

The hell are you talking about- Intel mobile isn't Intel desktop, mobile is where they've been investing. Enthusiasts have been bitching about it for years!

And 15w is 15w; it's a hard limitation in ultrabooks where battery life is king, and heat output must be controlled. AMD can't get away with the TDP rating halfassery that they do on desktops here. Any rational person would expect their 15w 4C8T CPU component to be slower and/or use more power (and thus burn battery faster) as the competing 15w 4C8T Intel part, as this is AMD's first try.

What we can expect is for AMD to provide a competitive solution for the price, that might play casual games a little bit better.

 

[Shintai]

This product isn't going to sell more AMD GPUs, its just going to replace Polaris 11 discrete AMD GPUs in the MacBook Pro line while increasing the price with another few 100$ on the SKUs. But then they can add more batteries that this will require for the performance uplift that they desperately need against other products like the Surfacebook.

And then an epic NUC disaster with 1000$ bare bone NUCs and an attempt to cool a 100W little box. Buyers, 0.

Dont expect anyone else to touch it because the cost structure is insane and offers pretty much 0.

Its an Apple product and its not going to sell more than 100-200K units in its lifespan. The benefit for both AMD and Intel is they can charge even more money for the parts. All paid by the top end MacBook Pro buyers.

 

[Anarchist4000]

Now the big question:

What is the future of RTG?

Is Raj leaving because this deal shoots AMD in the foot? I'm going to have to re-contemplate all this now.

If the deal shot AMD in the foot Lisa would have axed it. AMD has indicated they don't have a competing product for whatever reason. No HBM based APUs in the very near future, but perhaps that is a play on words for 2018.

A HBM2 based ~Polaris10 does appear to exist though as a somewhat discrete part. So they could very well be making Nanos out of them if not using Vega.

You can disagree but look at how RX Vega is going. If you are talking large scale production for Intel , that is something completely different. And today were still waiting on custom Vega boards which supposedly are still not here because the volume that

AIBs aren't getting the volume, but that doesn't mean it isn't going somewhere. GPU demand has been strong in all recent financials, so they're shipping GPUs somewhere. Drivers seem a bit behind, but there seems to be a lot of work with all these products. We've already seen Vega64 competing with 1080ti despite that. The Polaris vs Pascal comparisons are expectedly shifting as well.

 

[Shintai]

Raja is gone because someone needs to take the blame for running the GPU division into the ground with as always bad strategic decisions and misplaced R&D. Vega is just the current bottom of that downhill curve.

And this deal only moves RTG one step closer to being a glorified IGP maker. Not to mention not only does it say Intel cant advance its IGP line up for customers like Apple. It also says AMD cant move its CPU line up either.

KBL-H(Not KBL-R or CFL-H) and Vega 11.

 

[tybert7]

I'm not sure intel can't make more powerful integrated gpus, it might be that the important patents and IP are too locked up by nvidia and amd, but this deal will make it easier to not have to engineer around those (likely more streamlined) technical solutions to graphics problems.

 

[5150Joker]

I just don't see the point of this module because Vega is power hungry and can't compete against Pascal in efficiency so even if Intel squeezes it in some gaming notebooks as an integrated solution, who'd want this over a Pascal notebook? At the 1060+ level I don't see this thing offering much in the way of any value and at 1050 it will be eclipsed by NVIDIAs power efficiency.

GPU demand has been strong in all recent financials, so they're shipping GPUs somewhere. Drivers seem a bit behind, but there seems to be a lot of work with all these products. We've already seen Vega64 competing with 1080ti despite that.

AMD has moved GPUs only thanks to miners otherwise their overall share isn't moving. As for Vega 64 keeping up with 1080 Ti, not even close unless you're talking about Wolfenstein which was made in partnership with Bethesda and even then it still loses. So even with a crutch, AMD still can't compete.

 

[Shintai]

I just don't see the point of this module because Vega is power hungry and can't compete against Pascal in efficiency so even if Intel squeezes it in some gaming notebooks as an integrated solution, who'd want this over a Pascal notebook? At the 1060+ level I don't see this thing offering much in the way of any value and at 1050 it will be eclipsed by NVIDIAs power efficiency.

Vega 11 isn't the only failure on the package, KBL-H is the other.

And this is a Q2ish 2018 product. KBL-R already launched and CFL-H will too at that time.

Other OEMs besides Apple will ask twice about they should pair their lineup with a slow, outdated and overly powerhungry and hot package vs CFL-H+Volta at that time frame.

 

[Pieter3dnow]

It's all about the ramp up. If these become available Q1, the HBM2 supply issues might be resolved. But I do agree this is more a niche piece for upper end intel CPU's.
And I thought the custom board issues were due to how there were differences in the Vega die heights.

I read about the difference in height but do AIB engineers sit on their hands all day and give up because none of them are smart enough to solve the problem ?

Timeframe is important. This is a early 2018 product. There's no rush to market like RX Vega. HBM2 yields maturing as we speak. The Radeon chip supplied looks to be ~200mm2 so basically, a small chip, yields will be non-issue.

Both companies would not commit for this to be a low volume product. Logic 101, use it.

The problem is not HBM2 yields it is how you package the HBM2 on the die and that is why AMD is the only option .....

If the deal shot AMD in the foot Lisa would have axed it. AMD has indicated they don't have a competing product for whatever reason. No HBM based APUs in the very near future, but perhaps that is a play on words for 2018.

A HBM2 based ~Polaris10 does appear to exist though as a somewhat discrete part. So they could very well be making Nanos out of them if not using Vega.
AIBs aren't getting the volume, but that doesn't mean it isn't going somewhere. GPU demand has been strong in all recent financials, so they're shipping GPUs somewhere. Drivers seem a bit behind, but there seems to be a lot of work with all these products. We've already seen Vega64 competing with 1080ti despite that. The Polaris vs Pascal comparisons are expectedly shifting as well.

The volume of HBM2 supply is critical for it to be a consumer budget option, there are no indicators that this is any different for Intel as for AMD. secondly why would AMD help launch a product that will screw themselves with the new https://www.anandtech.com/show/1196...md-apus-for-laptops-with-vega-and-updated-zen

Makes no sense whatsoever ....

 

[chithanh]

This product isn't going to sell more AMD GPUs, its just going to replace Polaris 11 discrete AMD GPUs in the MacBook Pro line

Its an Apple product and its not going to sell more than 100-200K units in its lifespan.

When was the last time Intel made a product specially for Apple (or any other individual company, for that matter)?
Last I can remember is Core Duo (Yonah) which were used in the first Intel Macs. Itanium for HP and Pentium 4 EE for Dell maybe count too.

Make no mistake, Apple likely demanded such a product. The current Core m have a hard time against Apple A10X performance-wise and will probably be obliterated by A11X which pushes into Core i territory. Intel needs to give Apple a reason to keep buying Intel chips. But the H-Series will see quite some uses outside Apple, as the package is much smaller than having discrete graphics, which benefits every maker of compact high-end computers.

 

[IdiotInCharge]

When was the last time Intel made a product specially for Apple (or any other individual company, for that matter)?

You haven't heard of ultrabooks and the Macbook Air?

 

[Anarchist4000]

The volume of HBM2 supply is critical for it to be a consumer budget option, there are no indicators that this is any different for Intel as for AMD. secondly why would AMD help launch a product that will screw themselves with the new https://www.anandtech.com/show/1196...md-apus-for-laptops-with-vega-and-updated-zen

Makes no sense whatsoever ....

HBM goes into such high margin parts that getting supply likely isn't the issue. I still think AMD has been selling everything into the pro markets. Possibly directly to financial institutions and not AIBs for mining. Wasn't it the Hynix CEO that stated customers were willing to pay 2.5x for it?

Oddly it isn't competing with any AMD product. At least not yet and that comparison is likely Polaris vs Vega(perhaps Navi?). As is Intel integrated a HBM2 based Polaris. Not all that different from a AIB combining Intel CPUs with discrete graphics. Only difference being the Intel chips locked to AMD.

The HBM2 connection will be interesting to see however. The big gain was the CPU being able to use the HBM as main memory, but that doesn't appear to be the case.

 

[Shintai]

HBM goes into such high margin parts that getting supply likely isn't the issue. I still think AMD has been selling everything into the pro markets. Possibly directly to financial institutions and not AIBs for mining. Wasn't it the Hynix CEO that stated customers were willing to pay 2.5x for it?

Oddly it isn't competing with any AMD product. At least not yet and that comparison is likely Polaris vs Vega(perhaps Navi?). As is Intel integrated a HBM2 based Polaris. Not all that different from a AIB combining Intel CPUs with discrete graphics. Only difference being the Intel chips locked to AMD.

The HBM2 connection will be interesting to see however. The big gain was the CPU being able to use the HBM as main memory, but that doesn't appear to be the case.

Selling everything into the highly paying pro market? Where is the money then?

The HBM2 connection is the same as always, replace interposer with EMIB. GPU attached to the CPU via an x8 interface. And guess what's missing on the package, something U and Y chips sport on theirs. The benefit besides cost that EMIB provides in this case is lower height compared to an interposer design.

 

[Deleted member 93354]

I just don't see the point of this module because Vega is power hungry and can't compete against Pascal in efficiency so even if Intel squeezes it in some gaming notebooks as an integrated solution, who'd want this over a Pascal notebook? At the 1060+ level I don't see this thing offering much in the way of any value and at 1050 it will be eclipsed by NVIDIAs power efficiency.



AMD has moved GPUs only thanks to miners otherwise their overall share isn't moving. As for Vega 64 keeping up with 1080 Ti, not even close unless you're talking about Wolfenstein which was made in partnership with Bethesda and even then it still loses. So even with a crutch, AMD still can't compete.

Once you start ramping down the volts on Vega it actually becomes quite efficient. And having direct access to that hbm2 might have side benefits for Intel. You can't package a Pascal the way you can Vega.

 

[Shintai]

Once you start ramping down the volts on Vega it actually becomes quite efficient. And having direct access to that hbm2 might have side benefits for Intel. You can't package a Pascal the way you can Vega.

There is no direct access to that HBM2 for the CPU. CPU<PCIe x8>GPU<EMIB/1024bit bus>HBM2

The NUC will be up to 100W and still losing to a mobile 1060. Power efficient isn't what it is. And Volta will be out by the time this product hits the market. The KBL-H instead of KBL-R or CFL-H is another blunder in power efficiency.

If this product was out 9-12 months ago it would be interesting. But in 6-9 months from now it sure aint.

 

[Deleted member 93354]

There is no direct access to that HBM2 for the CPU. CPU<PCIe x8>GPU<EMIB/1024bit bus>HBM2

The NUC will be up to 100W and still losing to a mobile 1060. Power efficient isn't what it is. And Volta will be out by the time this product hits the market. The KBL-H instead of KBL-R or CFL-H is another blunder in power efficiency.

If this product was out 9-12 months ago it would be interesting. But in 6-9 months from now it sure aint.

I meant it will certainly have better access than gddr5 access. When you encode movies most of it's done with intels quick sync and stays off the GPU. The reason being it's faster to access and pull the block data on board the CPU then to constantly shift back and forth across the bus and gddr5 for short quick ops. This may change that.

And Vega is more power efficient than you think at low power envelopes. Isn't Intel aiming for a 35W tdp?

So maybe between an Iris pro and 1060? But with other benefits like a smaller package and quicker vid mem access

 

[Shintai]

I meant it will certainly have better access than gddr5 access. When you encode movies most of it's done with intels quick sync and stays off the GPU. The reason being it's faster to access and pull the block data on board the CPU then to constantly shift back and forth across the bus and gddr5 for short quick ops. This may change that.

And Vega is more power efficient than you think at low power envelopes. Isn't Intel aiming for a 35W tdp?

So maybe between an Iris pro and 1060? But with other benefits like a smaller package and quicker vid mem access

Its faster than GDDR how? And the CPU is still connected to an old fashion PCIe x8.

35W TDP? The CPU part itself with cTDPdown is 35W. The NUC version goes up to 100W and its a mobile chip. There seems to be 2 TDP versions of this package, 65W and 100W. The 65W seems to have 1.4Ghz HBM2 and the 100W part 1.6Ghz HBM2. The top bin have a peak compute of 3.3Tflop and losing by quite a margin to mobile 1060.

 

[Uvaman2]

I think... we will know performance when it gets reviewed... everything else is just a guess.
Also, if Nvidia is so far ahead in mobile, why would Intel even consider AMD.. I mean is Nvidia that terrible to work with that Intel would say... "fuck it!, I am going with a vastly inferior product, I don't give shit, I would rather gouge both my eyes than to work with Nvidia"
I mean to me there is not much to this, its a bit of a binary decision.
Neither looks good for Nvidia honestly

 

[Shintai]

I think... we will know performance when it gets reviewed... everything else is just a guess.
Also, if Nvidia is so far ahead in mobile, why would Intel even consider AMD.. I mean is Nvidia that terrible to work with that Intel would say... "fuck it!, I am going with a vastly inferior product, I don't give shit, I would rather gouge both my eyes than to work with Nvidia"
I mean to me there is not much to this, its a bit of a binary decision.
Neither looks good for Nvidia honestly

This is an Apple product. Intel is just selling an EMIB packaging instead of an interposer because of the height issues with interposer in laptops. Its amazing tho they got away selling the way outdated KBL-H with it, specially in 2018 when its been replaced twice!

This is about as big of a win as the Polaris chips in current Macbook Pros.

 

[Pieter3dnow]

Selling everything into the highly paying pro market? Where is the money then?

Isn't that for Intel to figure out ?

So what is the problem? First Intel touts this as a big win for EMIB, a silicon interposer strip put in to their package. This saves cost over a full interposer like AMD has used but adds significant expense to the package cost. Unlike what Intel says, it is not cheap, not even close, only cheap*ER* than a full interposer if the package substrate with EMIB yields well. Intel will probably say it is incredible, best yields ever, but like their processes of late, big grain of salt until it is independently verified. Ironically the CPU to GPU connection isn’t an EMIB based trace, it is just PCIe pulled off the die and routed through the normal substrate.

The following is even more telling:

Intel is claiming that Kaby-G will enable thinner, smaller laptops. It will meet half of those claims vs a discrete GPU, with consequences. The questionable half is the thinner part, SemiAccurate can see the smaller side, Kaby-G will reduce board area, but thinner is a stretch.

https://semiaccurate.com/2017/11/06/intel-gpus-officially-dont-work-amd-rescue/

 

[Shintai]

Isn't that for Intel to figure out ?

I think you missed the context.

 

[IdiotInCharge]

Also, if Nvidia is so far ahead in mobile, why would Intel even consider AMD.. I mean is Nvidia that terrible to work with that Intel would say... "fuck it!, I am going with a vastly inferior product, I don't give shit, I would rather gouge both my eyes than to work with Nvidia"
I mean to me there is not much to this, its a bit of a binary decision.
Neither looks good for Nvidia honestly

Doesn't really look good or bad; these will not likely be high-volume parts. Nvidia could have just as easily said 'screw that volume, not worth it' and moved on; AMD isn't really in the position to turn away any potentially profitable business, though. This is also better for Intel than outright paying AMD not to die as they continue to work to avoid being broken up as a monopoly in the future.

 

[_mockingbird]

There goes the advantage that Raven Ridge once would have had.

Obviously, this isn't a huge surprise with the turncoat Raja in charge until recently.​

 

[Shintai]

There goes the advantage that Raven Ridge once would have had.

Obviously, this isn't a huge surprise with the turncoat Raja in charge until recently.​

RR was never an option. Also besides a couple of MacBook Pro SKUs and a couple of NUC SKUs with no buyers. Who will buy it? Its certainly not for gamers. Its ~3000$ devices from Apple and something along the line of a ~1000$ bare bone NUC to get some meh performance in both CPU and GPU that's a couple of generations behind.

 

[Nightfire]

Shintai, calm down. You are working overtime to discredit this product.
Are you actually bashing KBL-H in order to bash an unreleased product that uses AMD?

You are clearly NOT pro intel but instead ANTI Amd.

Ease up on the Haterade. Life is too short!

 

[Teenk9]

Well this thread was a nice read but that didn't last long. Usual AMD thread crappers are here now. Sigh.

 

[Shintai]

Shintai, calm down. You are working overtime to discredit this product.
Are you actually bashing KBL-H in order to bash an unreleased product that uses AMD?

You are clearly NOT pro intel but instead ANTI Amd.

Ease up on the Haterade. Life is too short!

I dont get your problem. Is it so important for you that this product is praised beyond what it is?

Do you like outdated technology in a fancy package sold as new? Specially when at launch it will be 2 generations behind? Would you buy a Skylake laptop with medio Maxwell or Tonga today for ~3000$?

The only people that will buy this is Apple consumers.

If the product had launched 9-12 months ago it would be interesting. But there is 6-9 months before you see it. Meaning 15-21 months after it was relevant. And you get to pay top premium as well.

 

[kac77]

And Nvidia put resources into CUDA because they pioneered GPU compute, and needed an API for their processors, because OpenCL didn't exist!

What the?!?! They didn't pioneer GPU compute. They BOUGHT GPU compute and all of the required SDKs. Big damn difference. CUDA was no where until the Ageia acquisition. Furthermore GPU Compute was being done long before the advent of CUDA. There are how to's to show how to do it with DX10 for crying out loud.

 

[Nightfire]

No but there will be plenty buying a console with Jaguar/Polaris today...

Are we pretending like Intel made huge leaps in the last couple of years?

 

[IdiotInCharge]

What the?!?! They didn't pioneer GPU compute. They BOUGHT GPU compute and all of the required SDKs. Big damn difference. CUDA was no where until the Ageia acquisition. Furthermore GPU Compute was being done long before the advent of CUDA. There are how to's to show how to do it with DX10 for crying out loud.

Yeah, I'm not talking about PhysX- I'm talking about big-iron, and I am talking about DX10-level hardware (and so on). You know, where CUDA is basically the only (good) option.

 

[IdiotInCharge]

No but there will be plenty buying a console with Jaguar/Polaris today...

Are we pretending like Intel made huge leaps in the last couple of years?

In terms of process efficiency? Absolutely.

They haven't pushed performance on the desktop (well, IPC), because in reality they're fast enough. But mobile battery life, server farm power bills? Those matter quite a bit more, and that's where their efforts have been focused.

 

[_mockingbird]

RR was never an option. Also besides a couple of MacBook Pro SKUs and a couple of NUC SKUs with no buyers. Who will buy it? Its certainly not for gamers. Its ~3000$ devices from Apple and something along the line of a ~1000$ bare bone NUC to get some meh performance in both CPU and GPU that's a couple of generations behind.

Do you know what AMD should have done?

Put its best integrated GPU on its Ryzen Mobile processors and sell the whole thing to Apple.

 

[kac77]

Yeah, I'm not talking about PhysX- I'm talking about big-iron, and I am talking about DX10-level hardware (and so on). You know, where CUDA is basically the only (good) option.

CUDA wasn't released until 2007. Direct X 10 came out before then. Further more you need unified shaders / shader model 4 to do it in any kind of sane way. Remember the whole Direct X 10 debacle? That was because Nvidia didn't have the hardware to do it initially. The first card with unified shaders for Nvidia was the 8000 series / Tesla. AMD had those long before then and was the primary go to initially for doing GPU compute on the card. Their first card that could do it was the 1800XL. Yep waaay back in 2005. Two years before CUDA / Nvidia hardware that could do it. OpenGL surprisingly was the first mainstream API to allow it to be done and that came out in 2004. Although I don't know of anyone who would want to do it with that. Most recommend at least 3.3.

Nvidia popularized GPU compute and made it accessible/easier for programmers. However, they in no way pioneered it. Not even close. Big-iron or otherwise.

 

[Uvaman2]

I dont get your problem. Is it so important for you that this product is praised beyond what it is?

Do you like outdated technology in a fancy package sold as new? Specially when at launch it will be 2 generations behind? Would you buy a Skylake laptop with medio Maxwell or Tonga today for ~3000$?

The only people that will buy this is Apple consumers.

If the product had launched 9-12 months ago it would be interesting. But there is 6-9 months before you see it. Meaning 15-21 months after it was relevant. And you get to pay top premium as well.

How to you claim to know what it is, up until yesterday only Kyle knew this might be happening, and no one believe him.
This chip might be a freaking performance monster... A PS4 or Xbox whatever on steroids for all we know.

 

[lcpiper]

What kind of thinking? If you can't comprehend that things get more features i don't know what to tell you. I quoted idiot and what did I say was wrong? He thinks gsync is superior just because AMD is moving to freesync 2.

Perhaps I misunderstood what you meant when you said "so NVidia can compete with AMD"? G-Sync has been superior in performance to Freesync, I can't say it will be superior to Freesync2, I haven't seen the numbers and use cases on it yet. Cost wise, I also feel that G-Sync is a superior option unless a person is already invested in Freesync capable components or unless you scale your performance needs for an "affordable solution" similar to when I build a machine specifically for 1080P gaming wishing to avoid the higher costs associated with more advanced display options.

So to be clear, when someone says "compete", it would be normal to think the speaker is referring to price points and performance. HDR is a feature but I don't know anyone who would make a major buying decision based solely on this feature, it's just not that ground breaking compared with adaptive sync, high refresh rates, etc which have tremendous impact on game play and not just how pretty the lighting is. HDR is a fad in comparison.

Of course some people must have the latest even if it is a relatively minor feature that requires full support of the display, the display adapter, and the software.

BTW, does idiot know that you quoted him?

 

[Deleted member 93354]

Whatever the hate for AMD's graphics

They dominate in the setup top and console marketplace. The integrated performance on XBOne X is pretty damn surprising given it's a modified jaguar core. And RTG is selling every last GPU they can thanks in part to miners. They are in a good position for now despite what others might claim. How long that holds out is a mystery. But I will stick by my statement that AMD needs Navi to be on par with NVIDIA in terms of cost to performance ratio or they are out of it.

Can Raj be blamed for this? Well a lot of RTG's problems were set in stone before Raj took the reigns. He did the best he could with what was handed to him. The next generation stuff will tell us how much of a future is left for RTG.