Intel to Supply Apple with Special High-End Haswell Processors for MacBook Pro

Intel Intentionally Killing dGPUs?

Has anyone else here read the semiaccurate rumors concerning Intel killing dGPUs by limiting access to the PCI-e?

The details presented are a bit speculative, but the general thrust of the articles makes sense and would explain why the next rMBP might only include an iGPU.

Did Intel restrict anandtech's testing of the Iris Pro because it would confirm the speculation by Charlie Demerjian?

Curiosities mine. Insights?

NVIDIA GeForce GT 720M Graphics (2GB)

Found this:

http://www.bhphotovideo.com/c/produ...30821-_-Body_Retail_Acer-Aspire-V7-582PG-9478

So if a $950 Acer ( Haswell 1.8 dual-core) computer can have Haswell and a NVIDIA GeForce GT 720M Graphics (2GB), surely a souped up next gen MacBook Pro Retina can also.

Sandy Santra said:

Worth waiting until October for.

Click to expand...

I read that apple may put intel hd 5100 to 13 in b rmbp hd 5100 are less powerful the. The 5200 but are good and casual gamers could like them

MacRumors said:

[url=http://cdn.macrumors.com/im/macrumorsthreadlogodarkd.png]Image[/url]


According to SemiAccurate, Intel will provide Apple with an ultra-high performance version of its Haswell for Apple's upcoming MacBook Pros. The report claims that the chips will include a special version of Intel's highest-end GT3e (Iris Pro 5200) integrated graphics.

Apple has requested a special top bin cream-of-the-crop GT3e selection from Intel, with "as much GPU power as possible." With the top chips going solely to Apple, that leaves the slightly less desirable remains for Intel's other customers.

It was widely assumed that Apple would release its next generation Haswell-based MacBook Pros at WWDC alongside the Haswell MacBook Airs, but that did not happen. Both the 13-inch and the 15-inch MacBook Pros have since shown up in benchmark results, however, suggesting that a release is coming soon.

The benchmarked 15-inch MacBook Pro revealed a Core i7-4950HQ chip running at 2.4 GHz, along with Iris Pro 5200 integrated graphics, which, as noted, are the best that Intel has to offer. Intel's promotional materials have suggested that the Iris offers 2-2.5x the performance power as the i7-3840QM chip with HD Graphics 4000 in the current high-end Retina MacBook Pro, but the new Retina MacBook Pros could see even greater performance boosts.

Image​

Earlier this week, KGI Securities analyst Ming-Chi Kuo suggested that the updated MacBook Pros will launch in mid-September. It is unclear if Apple will refresh both the Retina MacBook Pros and the standard MacBook Pros, but many believe that Apple is planning to phase out the non-Retina varieties.

Article Link: Intel to Supply Apple with Special High-End Haswell Processors for MacBook Pro

Click to expand...

Let's see a graph comparing Intel's GPU to the competition. Intel wouldn't stand a chance with real graphs, not the shiny starry "2X" throwaway garbage.

richeyty said:

And then apple will supply us with macbook pros

Click to expand...

No, Apple will SELL us MacBook Pros, and if we choose to buy them. I'll wait to read customers' feedback before buying, since the 2010-2012 models have had various issues. For a $2000+ product, people should have no problems - or, rather, the same problems repeatable by a large quantity of customers.

What I'm confused about is:

Current MacBook Pro Retina (Early 2013, 2.8GHz version).
- Intel-3840QM with 8 MB on-chip L3 cache.
Haswell
- Intel-4950HQ has only 6 MB on-chip L3 cache.

Current MacBook Pro Retina (Early 2013, 2.8GHz version).
- Intel HD Graphics 4000 with DDR3L SDRAM shared with main memory and
- Nvidia GeForce GT 650M with 1 GB GDDR5 memory.
Haswell
- Intel Iris Pro graphics 5200.
….and that's it?

I don't get it?

Val-kyrie said:

Has anyone else here read the semiaccurate rumors concerning Intel killing dGPUs by limiting access to the PCI-e?

The details presented are a bit speculative, but the general thrust of the articles makes sense and would explain why the next rMBP might only include an iGPU.

Did Intel restrict anandtech's testing of the Iris Pro because it would confirm the speculation by Charlie Demerjian?

Curiosities mine. Insights?

Click to expand...

Those articles from semiaccurate are so stupid it hurts reading them. I don't know how Charlie who is supposed to follow recent hardware developements can make such a huge mistake but he completely misundertood some info he read.
Yes the south bridge in all Haswell, Broadwell, Ivy Bridge ... Chipsets have only a bit of PCIe 2.0 lanes but those ARE NOT for the GPU. The GPU gets 16 lanes 3.0 which are in the northbridge on the cpu die. Those on the southbridge are only for some compatibility. Since that external chip is only connected via DMI which is only about as fast as Thunderbolt, there is absolutely no point in adding many fast PCIe lanes. They'd be fake anyway considering how bottlenecked by DMI they are.

Besides being completely wrong about the main thing he is ranting on about in those three articles, PCIe bandwidth is also not really a big issue in mobile. Most of the time GPUs are only connected via 8 lanes and not the max of 16 just because it saves power and more isn't needed. You need a lot of bandwidth with ridiculously fast desktop GPUs especially in Crossfire/SLI configurations but not for 650M class stuff. Mainstream mobile GPUs can barely compete with entry level desktop stuff.

For GPGPU stuff like OpenCL and its adoption it is really more the latency that hurts and not the bandwidth. That is why Intel and AMD APUs will eventually just pull ahead. Quicksync can do better and faster than a Nvidia Titan. The reason is you can use GPU resources and immediately alter the result still in the L3 cache with CPU Integer pipes then go back to GPU EUs. With dGPU you need a series of steps that the gpu can handle completely to make it worth transferring the work over to another chip.
AMD thinks if you put the GPU close enough to the CPU you don't need AVX2 instructions and waste millions of transistors on that. Dedicated simple don't allow for a processing model that this new gen of integrated GPUs does. This is why GPGPU computing never really got of the ground outside of HPC. In the long run an Iris Pro will offer more then a 650M for professional application performance.
And with AMD Huma coming and the PS4/Xbox One APU the hardware companies seem to expect more interest from developers to accelerate their code with gpu help.
CUDA cannot compete and Nvidia knows that. They are also trying to add a CPU core into their upcoming GPU designs. They are just going the reverse route of Intel. In the mobile and HPC space they have a chance. In between I doubt they will be able to compete with Intel or x86 in general.

In the first article he also seem slightly irrational. First he says Intel tries to force dedicated GPUs out of business and then he complains about Iris Pro costs. If Intel makes them expensive, they are throwing Nvidia/AMD a bone. If they made them as cheap as a HD 4600, then the would really force dGPUs out of business.

rainbowsofwhite said:

What I'm confused about is:

Current MacBook Pro Retina (Early 2013, 2.8GHz version).
- Intel-3840QM with 8 MB on-chip L3 cache.
Haswell
- Intel-4950HQ has only 6 MB on-chip L3 cache.

Current MacBook Pro Retina (Early 2013, 2.8GHz version).
- Intel HD Graphics 4000 with DDR3L SDRAM shared with main memory and
- Nvidia GeForce GT 650M with 1 GB GDDR5 memory.
Haswell
- Intel Iris Pro graphics 5200.
….and that's it?

I don't get it?

Click to expand...

It is also 128MB of 50GB/s both directions L4 cache which offers a 30% lower latency for the CPU too.
You also need to account for space savings and cost. Anything less then 2GB GDDR5 would be a joke next gen and you need to put that somewhere on the cramped logic board. The GPU needs space and the VRMs. Remove all that you have a smaller board or a much cheaper to manufacture board. (small logic boards are way more expensive)
You also save power by sharing resources like the memory bus.
The dGPU has to stay active when it might be used. The iGPU can go from lowest to highest power state in microseconds. Having photoshop open in the background won't mean many Watts wasted.
In raw power (excluding drivers and architecture) the Iris Pro is even slightly ahead of a 650M. It is not a little GPU anymore and in some workloads capable of pulling ahead. For the most part it is just good enough for anything other than games. A photoshop GPU accelerated rotation won't be any less fluid.

Anything other slower than a 760M 2GB GDDR5 GPU just wouldn't be worth the downsides in the next rMBP. The difference wouldn't be big enough for a 750M IMHO.

dusk007 said:

It is also 128MB of 50GB/s both directions L4 cache which offers a 30% lower latency for the CPU too.
You also need to account for space savings and cost. Anything less then 2GB GDDR5 would be a joke next gen and you need to put that somewhere on the cramped logic board. The GPU needs space and the VRMs. Remove all that you have a smaller board or a much cheaper to manufacture board. (small logic boards are way more expensive)
You also save power by sharing resources like the memory bus.
The dGPU has to stay active when it might be used. The iGPU can go from lowest to highest power state in microseconds. Having photoshop open in the background won't mean many Watts wasted.
In raw power (excluding drivers and architecture) the Iris Pro is even slightly ahead of a 650M. It is not a little GPU anymore and in some workloads capable of pulling ahead. For the most part it is just good enough for anything other than games. A photoshop GPU accelerated rotation won't be any less fluid.

Anything other slower than a 760M 2GB GDDR5 GPU just wouldn't be worth the downsides in the next rMBP. The difference wouldn't be big enough for a 750M IMHO.

Click to expand...

Thanks for your thorough explanation.

So if I was interested in playing the occasional game, the Iris Pro would suffice? Or would the dedicated GPU (current 650M) be the better option?

yup, sounds like they want to ditch the GPU and go with chipset video.
Maybe they have research showing that gamers, and content creators favor other Macs like the iMac and Mac Pro for "heavy" video.

rainbowsofwhite said:

Thanks for your thorough explanation.

So if I was interested in playing the occasional game, the Iris Pro would suffice? Or would the dedicated GPU (current 650M) be the better option?

Click to expand...

Of course it would suffice. To say it wouldn't, would mean that any 2011 MBP with 6750M was crap and all those that have one can forget about gaming. An Iris Pro beats that. It looses to a 650M but not by a big amount. You can benchmark the difference and it changes with each game but on average you won't notice it. There have been as big differences between AMD and Nvidia GPUs over the years, which people considered largely equal.
Some people here act as if the Iris Pro was crap like the early integrated GPUs. The 9400M or 320M of Nvidia never got that much criticism even if both a beaten by even a HD 4600.
The difference between a 650M and an Iris Pro with the settings that are actually likely to run fluid is not big. To say it doesn't cut it, is to say the last gen MBP is horribly outdated and needs to be thrown out immediately.

People complain because a nvidia GPU (like a 760M especially) would be (though barely) possible in the same chassis and that would offer more gaming performance. It is a "it could be better(in performance)" complaint not a "it is insufficient" complaint.
Apple could just let people the choice between an Iris Pro notebook and a 760M one. Because each has its benefit. I would get the Iris Pro even if it costs the same. I hate graphic switching in OSX. On a Windows notebook I would choose the dGPU because graphics switching works there and the difference in driver quality between the Nvidia maintained drivers and Intel's is much bigger. But Apple has never been about choice, especially not when it is not a clear one with very obvious price differences.
Maybe they reintroduce a 17" with a dGPU (765M) while they take the 15" Iris Pro only. That would make many people happy. I don't even want to imagine what prices would look like .

dusk007 said:

Of course it would suffice. To say it wouldn't, would mean that any 2011 MBP with 6750M was crap and all those that have one can forget about gaming. An Iris Pro beats that. It looses to a 650M but not by a big amount. You can benchmark the difference and it changes with each game but on average you won't notice it. There have been as big differences between AMD and Nvidia GPUs over the years, which people considered largely equal.
Some people here act as if the Iris Pro was crap like the early integrated GPUs. The 9400M or 320M of Nvidia never got that much criticism even if both a beaten by even a HD 4600.
The difference between a 650M and an Iris Pro with the settings that are actually likely to run fluid is not big. To say it doesn't cut it, is to say the last gen MBP is horribly outdated and needs to be thrown out immediately.

People complain because a nvidia GPU (like a 760M especially) would be (though barely) possible in the same chassis and that would offer more gaming performance. It is a "it could be better(in performance)" complaint not a "it is insufficient" complaint.
Apple could just let people the choice between an Iris Pro notebook and a 760M one. Because each has its benefit. I would get the Iris Pro even if it costs the same. I hate graphic switching in OSX. On a Windows notebook I would choose the dGPU because graphics switching works there and the difference in driver quality between the Nvidia maintained drivers and Intel's is much bigger. But Apple has never been about choice, especially not when it is not a clear one with very obvious price differences.
Maybe they reintroduce a 17" with a dGPU (765M) while they take the 15" Iris Pro only. That would make many people happy. I don't even want to imagine what prices would look like .

Click to expand...

Thanks for your reply.
You've cleared up a lot of my confusions

Looking forward to seeing some benchmarks when the MacBook Pro Retinas "finally" get released.
I think this has been the most painful wait I've ever experienced.

O.K I may be changing my mind about the Intel® Iris™ Pro graphics 5200.

This article describes the current and very old Nvidia GT 650M card not being able to handle 4K displays.
http://www.extremetech.com/gaming/1...-your-computer-isnt-powerful-enough-to-use-it

That could be a worry if Apple does decide to upgrade the current Thunderbolt display to 4K?

Compared to the Intel® Iris™ Pro graphics 5200
http://www.hardwarezone.com.ph/feature-intel-4th-generation-core-hd-graphics-performance-review

rainbowsofwhite said:

O.K I may be changing my mind about the Intel® Iris™ Pro graphics 5200.

This article describes the current and very old Nvidia GT 650M card not being able to handle 4K displays.
http://www.extremetech.com/gaming/1...-your-computer-isnt-powerful-enough-to-use-it

That could be a worry if Apple does decide to upgrade the current Thunderbolt display to 4K?

Compared to the Intel® Iris™ Pro graphics 5200
http://www.hardwarezone.com.ph/feature-intel-4th-generation-core-hd-graphics-performance-review

Click to expand...

Couple of issues with that:

1. According to the first article, the fact that the machine couldn't output 4K was a limit imposed by OS X, not the 650M. I would be surprised if Mavericks doesn't introduce 4K support to OS X. Consequently the card was having to scale massive resolutions, which would probably hammer the hardware.

2. The second article is basically Intel's own marketing materials. They're not ideal for providing a real world comparison.

Basically, my money is still on the 650M outperforming Iris across the board. The gap has certainly closed, but not completely.