[Fudzilla] Apple 2016 laptops will have AMD GPUs

[maddie]

As I said in that post, they could well have done it, AFTER 04/21/16, but, there is no way for us to know until the next quarter is out.
AMD isn't flush with $$$, so, it would be a huge, huge risk to have another fab agreement and they most certainly would have made a report to the SEC about it, which they didn't, at least, I couldn't find anything.


I don't disagree that they didn't have engineering samples out at this time last year, I just don't agree that they have been in production since then.
This is the first time using FinFets for AMD, and it takes a long, long time to get working silicon.
The latest SEC filings on 4/28/2016 basically say this, and it wasn't until the first quarter that OEMs/AIBs have had a chance to have some sort of quantities.


Yes, we will see them at the end of the month, no idea on how many we will see, but, something should be seen.

Is this a compulsory requirement? I'm ignorant about this.

 

[monstercameron]

They are not locked out, but it is hard for Apple to build Metal drivers for undocumented architectures.

this isnt discussed very much. Are they actually building it themselves or is it a reskin of another api?

 

[Rakehellion]

I don't think it was known that nVidia had been completely locked out though.

Apple switches back and forth every few generations. It's no big deal.

 

[monstercameron]

I'd say CUDA is more likely to disappear if it doesn't get open sourced than anything.

cuda's compiler is opensource.

 

[Rakehellion]

I thought Metal was their iOS graphics API they were using instead of OpenGL since they only used PowerVR GPU tech for their SoCs.

Metal is graphics and compute. The main advantage over OpenGL is the low driver overhead like DirectX 12 and Mantle.

 

[Rakehellion]

Metal uses OpenCL for compute

Source?

 

[sirmo]

Is this a compulsory requirement? I'm ignorant about this.

No, they can absolutely bundle it with other expenses and not list it as an itemized item. In fact they are more likely to do this in order to protect trade secrets like this one, than not.

The whole argument is kind of pointless too, because nothing is stopping AMD from using Samsung since it's the same 14nm LPP process. Even if the reported contract is not real. They can go out and get it signed at any time.

AMD also has enough cash on hand. $600m-$1b is how much they are keeping available. They just closed $600m of China ventures.

 

[BlitzWulf]

http://globalfoundries.com/docs/def...lobalfoundries-14nm-collaboration---final.pdf


http://www.globalfoundries.com/docs...ries-14nm-collaboration-pr-final.pdf?sfvrsn=0




Not sure if this is relevant but it seems to me to clearly outline a dual sourcing model for AMD from Samsung/Glofo it say's they were ready to ramp up by end of 2014.
Am I being obtuse?It seems logical Samsung would be their secondary source.

 

[Elixer]

Is this a compulsory requirement? I'm ignorant about this.

Yeah, anything that can adversely affect them must be disclosed ASAP if it wasn't mentioned in the quarterlies.
http://www.nasdaq.com/symbol/amd/sec-filings
They now are reporting a ton of stock stuff (selling/buying), a 'Conflict Minerals Disclosure and Report' and other stuff.

 

[jpiniero]

Keep in mind the only current laptop Apple sells that has a dGPU is the upgraded 15" MBP. Which starts at $2499. It's probally not a big % of sales, but of course anything that's not 0 is a win for AMD.

 

[3DVagabond]

I don't think they've been locked out.

But the performance/watt advantage of nvidia is mainly with graphics, with opencl the advantage is much smaller. Apple doesn't really care about graphics anyway, so why pay more for nvidia chips?

I'm assuming you read both AMD's and nVidia's offers? Because that's the only way you could make that statement about paying more.

Past efficiency performance doesn't matter for current products. I don't think nVidia is any more efficient in graphics. Nor in OpenCL maybe ever? You can't count Windows and DX performance. So it remains to be seen where the current products fall efficiency wise. Polaris looks to be pretty good though.

Although if nVidia tried to charge them $700 for a card that was $200 (GTX 460) <6 years ago, (retail pricing, of course) I suppose that would upset most people without a pre biased opinion. Or, possibly it's because AMD embraces open source compared to an nVidia ecosystem where nVidia holds all the cards and pulls all the strings? I can't see Apple preferring that independent of pricing. Perhaps AMD just goes that bit extra to keep their customers satisfied? We know that they don't just offer off the shelf product but will make custom silicon. Or maybe it's because AMD's product is better with up to date work loads allowing Apple's customers to move forward with the industry?

Without being privy to any info at all, except AMD got the design wins, we can't even make an educated guess why.

 

[3DVagabond]

They are not locked out, but it is hard for Apple to build Metal drivers for undocumented architectures.

I don't know if you would know, but I'm assuming Metal is another offshoot of Mantle? Again, I don't know this to be a fact, but it sure appears that way.

If this is the case it's also likely that without any input for Metal from nVidia that Metal runs like dog doo on nVidia. Do AMD or nVidia have drivers certified for Metal even?

 

[sirmo]

http://globalfoundries.com/docs/def...lobalfoundries-14nm-collaboration---final.pdf


http://www.globalfoundries.com/docs...ries-14nm-collaboration-pr-final.pdf?sfvrsn=0




Not sure if this is relevant but it seems to me to clearly outline a dual sourcing model for AMD from Samsung/Glofo it say's they were ready to ramp up by end of 2014.
Am I being obtuse?It seems logical Samsung would be their secondary source.

Nice find. So they wouldn't even need to go to Samsung directly to dual source.

 

[ThatBuzzkiller]

To not mention NVIDIA that first invented bindless resources a long long time ago.

Except they didn't, at least not hardware wise ...

They may have came up with the patent first but it was AMD's GCN that launched with the first GPU microachitecture to feature bindless resources before Nvidia did with Kepler ...

AMD is arguably closer to having fully bindless resources (they already do with their "unlimited resources" hardware feature) than Nvidia is and it still holds true today since the latter still imposes limits on the number of CBVs and UAVs that you can dynamically index at any given shaders including their latest Pascal architecture!

 

[Enigmoid]

I think it's more than 5%. Mac Pros are pretty popular. And so are iMacs. You have to understand all the macs with retna display had an AMD GPU. And retna was a pretty big selling point.

Mac Pros are tremendously low volume.

Keep in mind the only current laptop Apple sells that has a dGPU is the upgraded 15" MBP. Which starts at $2499. It's probally not a big % of sales, but of course anything that's not 0 is a win for AMD.

Exactly. The cheapest laptop is $2500 and the cheapest imac (27") $1800 (significantly more in other countries).

Discrete GPUs in apple macs is absolutely <5% of sales.

 

[Silverforce11]

They are not locked out, but it is hard for Apple to build Metal drivers for undocumented architectures.

I think you called it right here.

OSX has moved to Metal, a Mantle derivative API.

OSX productivity apps have native OpenCL acceleration.

There's no point to an NV GPU in any Mac moving forward, until NV's architecture is adept at those features.

I suspected as much for many months now, but as soon as I saw AMD show casing the RX 480 as 2304 SP, I knew it was Apple. Only they can demand such volumes met from AMD that means full Polaris is not being showcased as a product.

This is both good and bad for AMD, good that they have a guaranteed buyer of chips they make. Bad because the full Polaris 10, 2560 SP @ 1.4ghz or so, would potentially match the 1070.

The RX 480 looks and smells like any previous gen, like the 7850, 7950. Compared to it's bigger brother, the 7870, 7970, it has a deficit in cores and clocks.

Edit: Apple pushing for Metal in OSX is them getting ready for the Vulkan/DX12 era. Where AAA PC developers can have an easy time porting over the OSX. So having a large install base of capable dGPUs, is a win for moving OSX into gaming.

 

[Silverforce11]

Keep in mind the only current laptop Apple sells that has a dGPU is the upgraded 15" MBP. Which starts at $2499. It's probally not a big % of sales, but of course anything that's not 0 is a win for AMD.

I think we will see a different structure.

Given the small die size and TDP of Polaris 11, this is how I think it's gonna happen.

MacBooks smaller than 15 inch will have Polaris 11. The MBP gets Polaris 10.

5K iMAC gets Polaris 11/10. MacPros obviously Polaris 10.

Apple's volumes are huge. 2016 is going to result in AMD winning back a ton of dGPU market-share.

 

[topmounter]

Apple switches back and forth every few generations. It's no big deal.

Buying GPU's from AMD and helping to keep AMD a viable going concern is good for Apple since it should in theory help to keep Intel honest with regards to their CPU pricing.

 

[jpiniero]

MacBooks smaller than 15 inch will have Polaris 11. The MBP gets Polaris 10.

The new MBP is said to be thinner - if anything the 13" could drop to the 15 W Kabylake? U from the Broadwell U 28 W they are using now. The 15" will be thinner too, but would obviously still be enough to take Polaris (possibly a decently cut 10 or maybe 11)

 

[Silverforce11]

The new MBP is said to be thinner - if anything the 13" could drop to the 15 W Kabylake? U from the Broadwell U 28 W they are using now. The 15" will be thinner too, but would obviously still be enough to take Polaris (possibly a decently cut 10 or maybe 11)

Think full chips running much lower clocks and less vcore, the TDP will be very low.

Polaris 10 @ 850mhz for example, will sip power on FF.

The same for Polaris 11.

This is why the full chips are more important to Apple, more shaders, less clocks, lower vcore results in more perf/w at a lower wattage.

Think Tonga, cut-down 180W on PC, full-chip 125W on MacPros.

 

[parkerface]

Surely the 480's to the masses will outnumber whatever Apple buys up (assuming there is a 2560 part).

Who knows -- maybe we'll get lucky and find some 480's with those extra CU's artificially disabled? :sneaky:

 

[monstercameron]

Mac Pros are tremendously low volume.



Exactly. The cheapest laptop is $2500 and the cheapest imac (27") $1800 (significantly more in other countries).

Discrete GPUs in apple macs is absolutely <5% of sales.

Source please.

 

[renderstate]

Except they didn't, at least not hardware wise ...

They may have came up with the patent first but it was AMD's GCN that launched with the first GPU microachitecture to feature bindless resources before Nvidia did with Kepler ...

AMD is arguably closer to having fully bindless resources (they already do with their "unlimited resources" hardware feature) than Nvidia is and it still holds true today since the latter still imposes limits on the number of CBVs and UAVs that you can dynamically index at any given shaders including their latest Pascal architecture!

Are you sure?
According wikipedia the first GCN parts were released in 2011, while NVIDIA introduced bindless graphics in 2009: http://developer.download.nvidia.com/opengl/tutorials/bindless_graphics.pdf

 

[jpiniero]

Think full chips running much lower clocks and less vcore, the TDP will be very low.

You underestimate how much Apple loves thinness. Seriously, they are anorexic.

 

[swilli89]

Surely the 480's to the masses will outnumber whatever Apple buys up (assuming there is a 2560 part).

Who knows -- maybe we'll get lucky and find some 480's with those extra CU's artificially disabled? :sneaky:

And if AMD really wants to try true viral marketing.. then almost make that happen by design. I've heard multiple times in the past that these days, cut down dies are usually artificial. In order for a harvested die to be a physical one, you must have defects ONLY in the CUs and none in the memory interface and other many parts of the chip, I/O, ROPs, internal buses and pathways have to 100% intact. If you are planning on 50% of your volume to be the cut down version, there is no way in hell that you'll come close to that amount of die harvested chips that weren't artificially cut.

AMD once had a card based on RV300 called the Radeon 9700 Pro. It was the first time they ever truly held the graphics crown up to that point. The 9700 Pro had a little brother chip, the 9500 and 9500 Pro. If you bought a certain 9500 card they were almost 100% unlockable for both the the 128-bit to 256-bit memory interface and a doubling of graphics pipelines (before shaders were unified) from 4 to 8 pipelines, so you essentially got a downclocked 9700 Pro.

I see this as pure win for AMD. How many people out there would really want to go through a warranty voiding process even if it was relatively safe? Maybe 10% at the very very most? What it would really do is generate a whole bunch of hype and people would see a tangible wave of excitement and "rush to the shelves" type situation if they heard there was some magic unlocking ability.