What Apple needs to do is refresh the price of the iMac...
Bring the prices down by about $200 on each iMac model and more on the Mac Pro...
This will help Apple sell more in the meantime
I agree Apple needs to increase value to customers, but remember, all apple needs to do is make the 20" and 24" have an LED backlight, use the Nvidia 9400M chipset with 9600GT discrete GPU, and Core2Duo Penryns at 2.5ghz and above. Boom! You have a great iMac you can sell at current prices, which have great value and are attractive, and much less re-engineering to be done, can use all the existing case designs! And yes, Core2Duo Quad mobile processor in the highest-end 24" iMac...
It's not technology. The lack of any iMac update so far is indecision on Apple's part. Core i7 can wait. In the current economic climate a Core i7 iMac just does not make sense unless it is that upgradeable "xMac". A super-sexy Core i7 iMac with liquid cooling is just far too "premium" even for Apple to push as their core desktop line.
Does that mean Apple will be forced to offer discrete GPUs on all Nehalem-based models to promote the adoption of OpenCL given that GMA is too slow and the fact that third party IGPs are not viable? That's good news then.
Well that is the hope since DMI is to low-bandwidth to hang an IGP off of and QPI equipped CPUs aren't the right market for IGPs.
The other option is to do like Sideport in the ATI 780G, 790G, and 790GX. Basically have an IGP with a small amount of dedicated memory, say 128MB, with the rest shared with system memory. Of course, by the time you start giving IGPs dedicated memory, you probably should consider whether you might as well go with a discrete GPU to avoid thermal and transistor budget limitations of sticking the IGP with the rest of the chipset.
Quote:
Originally Posted by nvidia2008
As someone pointed out above, Apple may be Intel's bitch... Well, just like Apple had a long running exit strategy with compiling for x86 (codename Marklar or something)... Apple had better keep in touch with AMD, ATI and PASemi etc. to avoid being painting themselves into a corner with Intel and/or Nvidia.
Well I don't view the decision to integrate the northbridge on the CPU as particularly malicious since it makes sense from a performance perspective. Using only low-bandwidth DMI-links for most chips also makes sense since high-bandwidth QPI is mainly needed for multi-socket setups. In terms of licensing, Intel is probably just holding out hoping to get concessions from nVidia. Afterall, nVidia has agreed to license SLI now which is important since PCIe controllers are going to be integrated on the CPU. If the bus license agreement said FSB, then nVidia will be in the wrong and will need to negotiate a new agreement for QPI and/or DMI.
I don't doubt that Apple will keep their options open. Realistically, ATI is a bigger supporter of OpenCL, since they've said they will support standards like OpenCL and DX11 Compute Shaders over proprietary GPGPU frameworks. nVidia however, is going to continue to promote their proprietary C for CUDA language alongside OpenCL and DX11 CS.
It should be noted too that while Apple may have a lot riding on Intel, Intel has a lot riding on Apple too. With Dell now selling AMD CPUs, Apple is actually the lone major OEM that is Intel exclusive for computer CPUs. With Apple's marketing power, Intel losing Apple as a customer would be a critical blow to Intel's image even if it doesn't directly effect their bottom-line. It's in Intel's best interest to keep Apple if not happy, then at least satisfied.
Quote:
Originally Posted by nvidia2008
Do we know if those Clarksdale and Arrandale on-chip GPUs come even close to a 9400M level of performance?.
The GMA X3100 was produced on a 90nm process and the GMA X4500 was produced on a 65nm process. The desktop GMA X3100 was clocked at 667MHz with 8 stream processors and the GMA X4500 was clocked at 800MHz with 10 stream processors with some internal redesign and optimization. Usually, a shrink from 90nm to 65nm would allow the doubling of stream processors, so of that was given up for higher clock speed, but Intel was definitely conservative with the GMA X4500's design. The GMA X4500 is still about twice faster than the GMA X3100.
The IGP on Clarksdale and Arrandale will be built on a 45nm process. I think the most likely configuration will be 16 stream processors clocked at 1066MHz (8 times Nehalem's CPU bus of 133MHz). That's probably conservative and achievable, which is good since this is the first time an IGP has been put on a CPU even if it is on package as a separate die from the CPU. This configuration won't beat the 9400M but will come close, especially if Intel is serious about their drivers which you'd think they will be in the lead-up to Larrabee. Being close to the CPU, and the raw power of Nehalem will only help performance. Intel was also on the working group for OpenCL and made a whole bunch of flowery talk about it in press releases so it's very likely future IGP's will have OpenCL support.
I'm sure if Apple put up enough fuss, Intel may even make versions with overclocked IGPs, like how current iMacs use overclocked Santa Rosa platform chipsets and CPUs specifically made for Apple rather than the newer Montevina platform chipsets and CPUs.
I agree that this argument isnt much about equipment that Apple will use.
However this is more troublesome for all PC users who like to build their own systems. The idea of not being able to use an nVidia motherboard for a core i7 system is disturbing as this rules out the use of core i7 for serious gamers and the like who want standard nVidia features like SLI. This would be a bad blow for nVidia at one of its core markets
That's exactly what I was thinking and maybe that's where Apple is sh*ting its pants right now. Because now that Apple is fairly committed to Snow Leopard, GPGPU/OpenCL, etc... Apple has to pretty much ensure decent GPUs across its *entire* product line. Which may mean even the lower end Macs having *discrete* GPUs (Nvidia or ATI) which may be more expensive and cut into margins compared to having a cost-effective, powerful integrated GPU like the 9400M.
As someone pointed out above, Apple may be Intel's bitch... Well, just like Apple had a long running exit strategy with compiling for x86 (codename Marklar or something)... Apple had better keep in touch with AMD, ATI and PASemi etc. to avoid being painting themselves into a corner with Intel and/or Nvidia.
@ltcommander.data Thanks for taking the trouble to write such an informative post (and winterspan likewise.) Much appreciated.
Thanks.. You should know that even short little comments like that really do make it worthwhile...
Quote:
Originally Posted by nvidia2008
Do we know if those Clarksdale and Arrandale on-chip GPUs come even close to a 9400M level of performance?
as Itcommander said, it should be close to the 9400, I don't think anyone really knows the details. The good part is that the chips not only have the iGPU on the same CPU package so the connection is much faster than the old FSB, but likewise the iGPU has direct access from the CPU to two channels of fast DDR3 memory.
Quote:
Originally Posted by nvidia2008
Apologies if I sound non-techie and stuff, brain not functioning so optimally today. We need more info from the 1337 forumers following this thread.
No not at all... I'm no expert either, just a geek.
Quote:
Originally Posted by ltcommander.data
..
Thanks for all your input. I do have some questions.
"Well that is the hope since DMI is to low-bandwidth to hang an IGP off of and QPI equipped CPUs aren't the right market for IGPs."
Wikipedia says the newest DMI is basically a modified PCIe x4 link running at 10Gb/s (1.25GB/sec). I'm wondering how much bandwidth an X4500-class integrated GPU would need to perform well? Is it really not feasible at all for them to connect one through the southbridge over DMI?
"In terms of licensing, Intel is probably just holding out hoping to get concessions from nVidia. Afterall, nVidia has agreed to license SLI now which is important since PCIe controllers are going to be integrated on the CPU. If the bus license agreement said FSB, then nVidia will be in the wrong and will need to negotiate a new agreement for QPI and/or DMI."
Thats what it looks like to me as well..
"I don't doubt that Apple will keep their options open. Realistically, ATI is a bigger supporter of OpenCL, since they've said they will support standards like OpenCL and DX11 Compute Shaders over proprietary GPGPU frameworks. nVidia however, is going to continue to promote their proprietary C for CUDA language alongside OpenCL and DX11 CS. "
Do you know how nVidia is going to be pursuing this? Are they writing a CUDA->OpenCL compiler, using an abstraction layer, etc? I haven't fiddled with programming any of the GPGPU stuff yet....
I agree that this argument isnt much about equipment that Apple will use.
However this is more troublesome for all PC users who like to build their own systems. The idea of not being able to use an nVidia motherboard for a core i7 system is disturbing as this rules out the use of core i7 for serious gamers and the like who want standard nVidia features like SLI. This would be a bad blow for nVidia at one of its core markets
I think you guys will be ok.. Nvidia will no doubt eventually give in and just pay Intel whatever they want.. They can't afford to lose all the integrated GPU platforms..
I agree that this argument isnt much about equipment that Apple will use.
However this is more troublesome for all PC users who like to build their own systems. The idea of not being able to use an nVidia motherboard for a core i7 system is disturbing as this rules out the use of core i7 for serious gamers and the like who want standard nVidia features like SLI. This would be a bad blow for nVidia at one of its core markets
nVidia has now licensed SLI to work on Intel chipsets just like ATI has long done with Crossfire. So now Intel's chipsets have the advantage of being compatible with both SLI and Crossfire, while ATI and nVidia's chipsets only support their own technology.
Quote:
Originally Posted by winterspan
"Well that is the hope since DMI is to low-bandwidth to hang an IGP off of and QPI equipped CPUs aren't the right market for IGPs."
Wikipedia says the newest DMI is basically a modified PCIe x4 link running at 10Gb/s (1.25GB/sec). I'm wondering how much bandwidth an X4500-class integrated GPU would need to perform well? Is it really not feasible at all for them to connect one through the southbridge over DMI?
An IGP is not the same as a discrete GPU connecting through PCIe x4. As I understand things, as GPU process data they need a place to hold information to serve as inputs or to temporarily store their outputs before they are reprocessed. Discrete GPUs have dedicated VRAM to store information and the PCIe link only serves to an input source from the CPU and HDD. In an IGP, it relies on system memory as a cache and it needs to be constantly reading and writing to it, which is why IGPs are slower since they are competing with the CPU for memory bandwidth.
Normally, the IGP and memory controller are both on the northbridge of the chipset so the IGP technically has direct access to the memory, while the CPU has to jump through the FSB to access memory. This has been the common complaint with the FSB, because it can often be the bottleneck for CPU performance. In current MacBook and MacBook Pros, the FSB is 1066MHz which equates to 1 channel of DDR3-1066 memory with dual channel largely being wasted since the FSB basically can't allow anymore information through. The whole point of integrating the memory controller on the CPU like in AMD chips and in Nehalem is to remove the FSB bottleneck and allow the CPU to take advantage of the full dual or triple channel bandwidth. (It's even more important in multiprocessor systems like the Mac Pro so that memory bandwidth scales linearly with CPU count since each CPU has their own memory pool.)
The disadvantage of integrated memory controllers is that an external IGP would now have to be 1 hop away from system memory and be bottlenecked by the bus interface, in this case DMI. Anytime the IGP reads or writes something it'll have to go through the DMI link and 10Gb/s or 1.25GB/s is not enough considering the current 9400M has direct access to the northbridge memory controller and has the full 17GB/s of dual channel DDR3-1066 available. What's more, the DMI link is used for all peripherals too like the HDD, USB, ethernet, etc to reach the CPU for processing. DMI just doesn't have the bandwidth if the IGP relies on it exclusively for memory access.
Quote:
Originally Posted by winterspan
"I don't doubt that Apple will keep their options open. Realistically, ATI is a bigger supporter of OpenCL, since they've said they will support standards like OpenCL and DX11 Compute Shaders over proprietary GPGPU frameworks. nVidia however, is going to continue to promote their proprietary C for CUDA language alongside OpenCL and DX11 CS. "
Do you know how nVidia is going to be pursuing this? Are they writing a CUDA->OpenCL compiler, using an abstraction layer, etc? I haven't fiddled with programming any of the GPGPU stuff yet....
Thanks
You'll need to parse nVidia's marketing. Before there were standardized OpenCL or DX11 Compute Shaders, I believe nVidia just referred to all GPGPU operation on their GPUs as CUDA. They now basically break things down into a language and an interface to convert that language into the machine code that the GPU hardware understands. OpenCL and DX11 CS are two languages, and nVidia now specifically refers to C for CUDA as their proprietary language. CUDA itself is going to be the interface layer to convert GPGPU languages, whichever they are, to something the GPU will understand.
ATI is similar. Their proprietary language is Brook+, and Stream is their broader label for interfacing with the GPU.
Anytime the IGP reads or writes something it'll have to go through the DMI link and 10Gb/s or 1.25GB/s is not enough ...
So I guess the good news is that the only likely integrated GPUs of Nehalem-series chips will be in the budget dual-core models with the on-package GPUs (with direct access to the dual-channel DDR3). So someone like nVidia or ATI can't bolt on a crappy integrated GPU onto the quad-core clarksfield and lynnfield platforms.. Hopefully Apple expands their use of discrete (and more powerful) GPUs in the future with the release of SnowL..
When it comes to quick path it really does matter if Nvidia wants to go with an integrated GPU on that bus or not they will likely need some sort of license from intel. Now the question is do they already have one.
With Quick Path the distinction as to what may be considered a integrated GPU will likely become blurred. The thing here is that both Apple and Nvidia would likely want the GPUs to plug directly into Quick Path. That to keep the GPU as close to system memory as possible. Even then I'd expect a Quick Pathe connected GPU to have local memory so they may look more like a descrete GPU no matter the performance. I just have a hardtime seeing a GPU using Quick Path for all its memory accesses, it is not like it would be the only user of the bus.
I can actually see two classes of Quick Path using GPUs. One would be the high integration device that is thought of as an integrated chip today. That is a device with the GPU and all your I/O ports onboard. For the higher performance market I could see a Quick Path connected GPU that is much higher performance and passes the I/O thru to a south bridge. In essence if you are a GPU maker the performance of Quick Path is hard to ignore. If you are Apple you will want that mass of vector processors close to the memory they will work on.
As to what intel thinks it will ship between the Quick Path and DMI ports that has yet to be seen. Quick Pqth does have huge advantages and I can see intel getting all worked up over the DMI interface and then having it's various vendors dismiss it outright. For some apps the DMI based systems will be simple but OK for the intended uses. As a mother board maker though will you have enough options to produce outstanding product. I suspect that if Intel where to allow it to grow Quick Path could have a huge market with respect to support chipsets. It looks like Intel is working overtime to manipulate the market here.
Which brings us to AMD. Yeah their processors are a little slow but maybe not as bad as some think. If intel leaves them a big hole with respect to high speed interfacing to a GPU AMD will fill it. AND already has plenty of experience with on board memory controllers and high speed interfaces, with ATI they also have good GPU tech. It won't take much effort on their part to do an integrated, on the processor die, GPU. One should have much trouble imagining that this would be faster than the Intel solution and supportive of OpenCL. What I'm saying is that even though AMD lags a bit they could still force Intels hand via a much better processor with onboard video. Better in the sense that the GPU is often more important than the i86 unit. So what I'm saying is that it is not a sure thing that intels processor + GPU option will have the market share intel hopes for.
Part of this article is BS in my mind as it speculates as to what is holding up Apples new hardware. None of this speculation about i7 has anything to do with the Mini. At least I've not heard of anyone saying the Mini will get an i7 class processor. The iMac is a different story though as there is more than a good chance i7 could go in there. I just don't see Apple holding up one machine, a low cost one at that, due to issues on another.
As to people thinking Apple has painted themselves into a corner here with Nvidia, I see nothing that implies that. If they went to ATI they likely would have had the same issues. The other problem is that there will be years ahead of us before the GPUs in processor + GPU combos catch up with GPUs outside the main chip. If they ever catch up, though they will be certainly good enough for many apps, Apple won't enhance it's reputation as a producer of great graphics hardware by shipping huge quanities of them.
....The GMA X4500 is still about twice faster than the GMA X3100...The IGP on Clarksdale and Arrandale will be built on a 45nm process. I think the most likely configuration will be 16 stream processors clocked at 1066MHz (8 times Nehalem's CPU bus of 133MHz). That's probably conservative and achievable, which is good since this is the first time an IGP has been put on a CPU even if it is on package as a separate die from the CPU. This configuration won't beat the 9400M but will come close, especially if Intel is serious about their drivers which you'd think they will be in the lead-up to Larrabee. Being close to the CPU, and the raw power of Nehalem will only help performance. Intel was also on the working group for OpenCL and made a whole bunch of flowery talk about it in press releases so it's very likely future IGP's will have OpenCL support... I'm sure if Apple put up enough fuss, Intel may even make versions with overclocked IGPs, like how current iMacs use overclocked Santa Rosa platform chipsets and CPUs specifically made for Apple rather than the newer Montevina platform chipsets and CPUs...
Quote:
Originally Posted by winterspan
....as Itcommander said, it should be close to the 9400, I don't think anyone really knows the details. The good part is that the chips not only have the iGPU on the same CPU package so the connection is much faster than the old FSB, but likewise the iGPU has direct access from the CPU to two channels of fast DDR3 memory...
OK, thanks for the info.
To me though, my intepretation of this is that towards the 2nd half of 2009 Intel will have an integrated graphics solution that maybe comes close to what Nvidia had in October 2008.
So we still don't know if Intel can lift its GPU game.
Maybe if Intel holds out against Nvidia maybe this will force Apple to bite the bullet and go full discrete graphics across the Mac line by this 2009's holiday season. Nehalem duals and quads all with discrete GPUs, Snow Leopard, OpenCL, GrandCentral... would be very, very sweet.
*HOWEVER* with the current economy, that dream for 2nd half of 2009 may not materialise. Apple needs to shift its strategy for 2009 and 2010 now, so... who knows what we will see out of them.
...None of this speculation about i7 has anything to do with the Mini. At least I've not heard of anyone saying the Mini will get an i7 class processor. The iMac is a different story though as there is more than a good chance i7 could go in there. I just don't see Apple holding up one machine, a low cost one at that, due to issues on another...Dave
That's what I've been saying. Even with the iMac, think about the 20" and can you imagine what it would take to get a Core i7 in there? Nothing less than some stunning, margin-gobbling engineering.
I mean, the HP Firebird is pretty impressive and it's only a Core 2 Quad and still requires liquid cooling.
If Apple goes liquid cooling on a Core i7 iMac 22" and 26" or something like that, that would be stunning, but the economic conditions in 2009 and 2010, just don't make that feasible no matter what the price. Not without Steve actively in the picture.
.... With Quick Path the distinction as to what may be considered a integrated GPU will likely become blurred. The thing here is that both Apple and Nvidia would likely want the GPUs to plug directly into Quick Path. That to keep the GPU as close to system memory as possible...
Now I am not going to say it's impossible, but at this time it's surely a stretch to envision a GPU that connects through the quickpath interface. There are many reasons for this:
1) Perhaps most relevant, The PCIexpress bus is plenty capable of continuing to handle the job. PCIe 2.0 with a standard x16 connector moves over 8.0 GB/sec, and even the latest ridiculously high-end dual-chip GPUs didn't saturate the old PCIe 1.1 standard with 4.0 GB/sec. The PCIe 3.0 spec which will be finalized this year will bump that up to 16.0GB/sec. The high memory bandwidth requirements of modern GPUs is predominately between the GPU and the dedicated GDRAM on the card itself.
2) As i mentioned in my other post, the vast majority of future Nehalem systems will *NOT* even have quickpath. Only the high-end Bloomfield CPU/X58 platform and the Xeon/server platforms even have Quickpath. All the future consumer/enthusiast focused Nehalems will have the northbridge/PCIexpress bus built right into the CPU die itself and will connect straight to the southbridge chip via DMI --- thus no quickpath.
Quote:
Originally Posted by wizard69
Part of this article is BS in my mind as it speculates as to what is holding up Apples new hardware. None of this speculation about i7 has anything to do with the Mini. At least I've not heard of anyone saying the Mini will get an i7 class processor. The iMac is a different story though as there is more than a good chance i7 could go in there. I just don't see Apple holding up one machine, a low cost one at that, due to issues on another.
Not only the Mini, it is *incredibly* unlikely the iMac would see an i7. The TDP and power usage is far higher than the currently used Core 2 duos, not to mention it hasn't even seen a low-power Core 2 Quad CPU yet. Also, as I mentioned above, the future Nehalems, namely the quad-core "Clarksfield" and "Lynnfield" are much better suited to the iMac as they use a lot less power and have a smaller 1-chip chipset/motherboard (no northbridge). For this reason I think we'll soon see a Core 2 Quad iMac update, and we won't see a Nehalem variant in the iMac until Q4 2009 or Q1 2010 (or whenever Intel has the new chips ready).
I mean, the HP Firebird is pretty impressive and it's only a Core 2 Quad and still requires liquid cooling....
While I agree that a 100W Core i7 won't see its way into the iMac, a Core 2 Quad would not require liquid cooling. The HP is almost certainly not using Intel's newest low-power variant of the Core 2 Quad, and probably using them at high-frequency too. It also is potentially more for cooling the GPU and reducing fan noise than anything else.
I'd bet we'll see 55W/65W TDP Core 2 Quad iMacs pretty soon, probably running at around 2.5Ghz.
ltcommander.data you make some really good points but discrete gfx on mobile Westmere have nothing to do with DMI. The IGP part of the MCM holds the Intel GPU, the integrated DDR3 controller, and a PCIe 2.0 controller, almost surely 16x. Any discrete gpu on the Westmere platform will be connected directly to the cpu through a PCE 2.0 16x bus. Not surprising since Lynnfield will also have a PCIe controller on the cpu, as will all future intel cpu's except the high-end socket 1336 x58 platform, it needs more than 16x. After the IMC, it was the next logical move on the way to total SoC.
page 9, as you can see the DMI connection is to the southbridge only, while another bus comes out of the iGFX die to PCIe Gfx. The next slide (pg 10) clearly says switchable gfx supported, intel isn't dumb enough to force everyone who wants these to use their IGP but then require discrete gpus to go through DMI making said discrete gpu useless, the platform would fail.
Discrete graphics will be just fine on any new westmere-based Macs, hopefully I can resist the urge til then. Never liked apple but after using my gf's new aluminum wonder with the amazing multitouch gestures I'm converted. I found a way to get rid of OS X's wretched mouse acceleration so that's one less obstacle If they only offered the illuminated keyboard on the 2.0ghz model I would probably buy it immediately.
Now the question is will the IGP be as fast as the 9400m because I don't see apple including a discrete gpu on the 13''...
"The northern Californian companies have been arguing for a year about whether a deal they inked in 2004 allows NVIDIA to produce chipsets that work with Intel microprocessors that have integrated-memory controller features."
This is reported in most of the news sources.
Therefore, it looks like this applies across the Nehalem line because they will all have integrated memory controllers.
Nvidia is definitely somewhat concerned because Bloomfield, Lynnfield and Clarksfield (in other words Nehalem 45nm) all have integrated memory controllers.
Firstly, Nvidia loses out on the "FSB" parts of the chipset like the memory controller that are now built into the CPU. This is 100% regardless of the squabble. Secondly, Nvidia may lose out on integrated graphics. Even worse. Thirdly is the southbridge (QPI or DMI...??not sure about this part) stuff at risk -possibly bad too for Nvidia.
The MAIN thing is...
For Bloomfield, Lynnfield and Clarksfield (in other words Nehalem 45nm), if the courts rule in favour of Intel and Intel-Nvidia don't hammer out another deal, it means that *any* "integrated graphics" on any Intel-CPU-based motherboard is an Intel. Regardless of whether this Intel integrated is in the CPU (which it isn't, until Westmere) or on the motherboard, if Nvidia is locked out, it is a big chunk of Nvidia's strategy at large risk.
Forget about Westmere 32nm towards the end of the year. We're talking Intel's latest and greatest these three quarters, January to September 2009. What are the best graphics Intel can offer right now? X4500HD with no real roadmap and nowhere near ATI and Nvidia advances.
So the real risk to Apple is not the iMac which wouldn't have integrated graphics anyway or core i7, it is *the very next refresh of the Aluminium MacBook line... in otherwords, between now and September 2009, if Apple goes Nehalem in mobile, they might have to kiss the 9400M/Nvidia chipset goodbye, or, stick with Core 2 for the next 9 months.
And while this legal stuff goes on for a few months, again, Apple in the same position, forget the iMac, talkin' about the Aluminium MacBooks here... No NEHALEM. NO NEHALEM IN MACBOOK PRO until/ unless Apple goes back and throws out the 9400M/Nvidia-whatever chipset.
ltcommander.data you make some really good points but discrete gfx on mobile Westmere have nothing to do with DMI. The IGP part of the MCM holds the Intel GPU, the integrated DDR3 controller, and a PCIe 2.0 controller, almost surely 16x. Any discrete gpu on the Westmere platform will be connected directly to the cpu through a PCE 2.0 16x bus. Not surprising since Lynnfield will also have a PCIe controller on the cpu, as will all future intel cpu's except the high-end socket 1336 x58 platform, it needs more than 16x. After the IMC, it was the next logical move on the way to total SoC.
Yep, the on-CPU northbridge will have at least 16 PCIe 2.0 lanes.
ltcommander.data you make some really good points but discrete gfx on mobile Westmere have nothing to do with DMI. The IGP part of the MCM holds the Intel GPU, the integrated DDR3 controller, and a PCIe 2.0 controller, almost surely 16x. Any discrete gpu on the Westmere platform will be connected directly to the cpu through a PCE 2.0 16x bus. Not surprising since Lynnfield will also have a PCIe controller on the cpu, as will all future intel cpu's except the high-end socket 1336 x58 platform, it needs more than 16x. After the IMC, it was the next logical move on the way to total SoC.
page 9, as you can see the DMI connection is to the southbridge only, while another bus comes out of the iGFX die to PCIe Gfx. The next slide (pg 10) clearly says switchable gfx supported, intel isn't dumb enough to force everyone who wants these to use their IGP but then require discrete gpus to go through DMI making said discrete gpu useless, the platform would fail.
Discrete graphics will be just fine on any new westmere-based Macs, hopefully I can resist the urge til then. Never liked apple but after using my gf's new aluminum wonder with the amazing multitouch gestures I'm converted. I found a way to get rid of OS X's wretched mouse acceleration so that's one less obstacle If they only offered the illuminated keyboard on the 2.0ghz model I would probably buy it immediately.
Now the question is will the IGP be as fast as the 9400m because I don't see apple including a discrete gpu on the 13''...
My point was never the inability to add a discrete GPU to any Nehalem or Westmere parts because as you noted they have integrated PCIe controllers. The AppleInsider article was concerned about nVidia IGPs on nVidia chipsets, and my point is that even if nVIdia got a license to produce chipsets for Nehalem, those chipsets aren't likely to have viable IGPs like the current 9400M IGP/chipset because the memory controller is no longer on the chipset and an external IGP will need to connect through DMI which as you note is designed for southbridges and doesn't have the bandwidth. My suggestion was that nVidia focus on fast discrete low-cost GPUs, since those will also be faster than an IGP, not necessarily that much more expensive, and will benefit from the integrated PCIe controller allowing the GPU to be closer to the CPU.
Comments
What Apple needs to do is refresh the price of the iMac...
Bring the prices down by about $200 on each iMac model and more on the Mac Pro...
This will help Apple sell more in the meantime
I agree Apple needs to increase value to customers, but remember, all apple needs to do is make the 20" and 24" have an LED backlight, use the Nvidia 9400M chipset with 9600GT discrete GPU, and Core2Duo Penryns at 2.5ghz and above. Boom! You have a great iMac you can sell at current prices, which have great value and are attractive, and much less re-engineering to be done, can use all the existing case designs! And yes, Core2Duo Quad mobile processor in the highest-end 24" iMac...
It's not technology. The lack of any iMac update so far is indecision on Apple's part. Core i7 can wait. In the current economic climate a Core i7 iMac just does not make sense unless it is that upgradeable "xMac". A super-sexy Core i7 iMac with liquid cooling is just far too "premium" even for Apple to push as their core desktop line.
Then Apple would just buy Nvidia's Chip and renamed them to Apple's Chip.
Thanks for taking the trouble to write such an informative post (and winterspan likewise.) Much appreciated.
Does that mean Apple will be forced to offer discrete GPUs on all Nehalem-based models to promote the adoption of OpenCL given that GMA is too slow and the fact that third party IGPs are not viable? That's good news then.
Well that is the hope since DMI is to low-bandwidth to hang an IGP off of and QPI equipped CPUs aren't the right market for IGPs.
The other option is to do like Sideport in the ATI 780G, 790G, and 790GX. Basically have an IGP with a small amount of dedicated memory, say 128MB, with the rest shared with system memory. Of course, by the time you start giving IGPs dedicated memory, you probably should consider whether you might as well go with a discrete GPU to avoid thermal and transistor budget limitations of sticking the IGP with the rest of the chipset.
As someone pointed out above, Apple may be Intel's bitch... Well, just like Apple had a long running exit strategy with compiling for x86 (codename Marklar or something)... Apple had better keep in touch with AMD, ATI and PASemi etc. to avoid being painting themselves into a corner with Intel and/or Nvidia.
Well I don't view the decision to integrate the northbridge on the CPU as particularly malicious since it makes sense from a performance perspective. Using only low-bandwidth DMI-links for most chips also makes sense since high-bandwidth QPI is mainly needed for multi-socket setups. In terms of licensing, Intel is probably just holding out hoping to get concessions from nVidia. Afterall, nVidia has agreed to license SLI now which is important since PCIe controllers are going to be integrated on the CPU. If the bus license agreement said FSB, then nVidia will be in the wrong and will need to negotiate a new agreement for QPI and/or DMI.
I don't doubt that Apple will keep their options open. Realistically, ATI is a bigger supporter of OpenCL, since they've said they will support standards like OpenCL and DX11 Compute Shaders over proprietary GPGPU frameworks. nVidia however, is going to continue to promote their proprietary C for CUDA language alongside OpenCL and DX11 CS.
It should be noted too that while Apple may have a lot riding on Intel, Intel has a lot riding on Apple too. With Dell now selling AMD CPUs, Apple is actually the lone major OEM that is Intel exclusive for computer CPUs. With Apple's marketing power, Intel losing Apple as a customer would be a critical blow to Intel's image even if it doesn't directly effect their bottom-line. It's in Intel's best interest to keep Apple if not happy, then at least satisfied.
Do we know if those Clarksdale and Arrandale on-chip GPUs come even close to a 9400M level of performance?.
The GMA X3100 was produced on a 90nm process and the GMA X4500 was produced on a 65nm process. The desktop GMA X3100 was clocked at 667MHz with 8 stream processors and the GMA X4500 was clocked at 800MHz with 10 stream processors with some internal redesign and optimization. Usually, a shrink from 90nm to 65nm would allow the doubling of stream processors, so of that was given up for higher clock speed, but Intel was definitely conservative with the GMA X4500's design. The GMA X4500 is still about twice faster than the GMA X3100.
The IGP on Clarksdale and Arrandale will be built on a 45nm process. I think the most likely configuration will be 16 stream processors clocked at 1066MHz (8 times Nehalem's CPU bus of 133MHz). That's probably conservative and achievable, which is good since this is the first time an IGP has been put on a CPU even if it is on package as a separate die from the CPU. This configuration won't beat the 9400M but will come close, especially if Intel is serious about their drivers which you'd think they will be in the lead-up to Larrabee. Being close to the CPU, and the raw power of Nehalem will only help performance. Intel was also on the working group for OpenCL and made a whole bunch of flowery talk about it in press releases so it's very likely future IGP's will have OpenCL support.
I'm sure if Apple put up enough fuss, Intel may even make versions with overclocked IGPs, like how current iMacs use overclocked Santa Rosa platform chipsets and CPUs specifically made for Apple rather than the newer Montevina platform chipsets and CPUs.
@ltcommander.data
Thanks for taking the trouble to write such an informative post (and winterspan likewise.) Much appreciated.
Thanks
However this is more troublesome for all PC users who like to build their own systems. The idea of not being able to use an nVidia motherboard for a core i7 system is disturbing as this rules out the use of core i7 for serious gamers and the like who want standard nVidia features like SLI. This would be a bad blow for nVidia at one of its core markets
That's exactly what I was thinking and maybe that's where Apple is sh*ting its pants right now. Because now that Apple is fairly committed to Snow Leopard, GPGPU/OpenCL, etc... Apple has to pretty much ensure decent GPUs across its *entire* product line. Which may mean even the lower end Macs having *discrete* GPUs (Nvidia or ATI) which may be more expensive and cut into margins compared to having a cost-effective, powerful integrated GPU like the 9400M.
As someone pointed out above, Apple may be Intel's bitch... Well, just like Apple had a long running exit strategy with compiling for x86 (codename Marklar or something)... Apple had better keep in touch with AMD, ATI and PASemi etc. to avoid being painting themselves into a corner with Intel and/or Nvidia.
PASemi is Apple.
@ltcommander.data Thanks for taking the trouble to write such an informative post (and winterspan likewise.) Much appreciated.
Thanks.. You should know that even short little comments like that really do make it worthwhile...
Do we know if those Clarksdale and Arrandale on-chip GPUs come even close to a 9400M level of performance?
as Itcommander said, it should be close to the 9400, I don't think anyone really knows the details. The good part is that the chips not only have the iGPU on the same CPU package so the connection is much faster than the old FSB, but likewise the iGPU has direct access from the CPU to two channels of fast DDR3 memory.
Apologies if I sound non-techie and stuff, brain not functioning so optimally today. We need more info from the 1337 forumers following this thread.
No not at all... I'm no expert either, just a geek.
..
Thanks for all your input. I do have some questions.
"Well that is the hope since DMI is to low-bandwidth to hang an IGP off of and QPI equipped CPUs aren't the right market for IGPs."
Wikipedia says the newest DMI is basically a modified PCIe x4 link running at 10Gb/s (1.25GB/sec). I'm wondering how much bandwidth an X4500-class integrated GPU would need to perform well? Is it really not feasible at all for them to connect one through the southbridge over DMI?
"In terms of licensing, Intel is probably just holding out hoping to get concessions from nVidia. Afterall, nVidia has agreed to license SLI now which is important since PCIe controllers are going to be integrated on the CPU. If the bus license agreement said FSB, then nVidia will be in the wrong and will need to negotiate a new agreement for QPI and/or DMI."
Thats what it looks like to me as well..
"I don't doubt that Apple will keep their options open. Realistically, ATI is a bigger supporter of OpenCL, since they've said they will support standards like OpenCL and DX11 Compute Shaders over proprietary GPGPU frameworks. nVidia however, is going to continue to promote their proprietary C for CUDA language alongside OpenCL and DX11 CS. "
Do you know how nVidia is going to be pursuing this? Are they writing a CUDA->OpenCL compiler, using an abstraction layer, etc? I haven't fiddled with programming any of the GPGPU stuff yet....
Thanks[/QUOTE]
I agree that this argument isnt much about equipment that Apple will use.
However this is more troublesome for all PC users who like to build their own systems. The idea of not being able to use an nVidia motherboard for a core i7 system is disturbing as this rules out the use of core i7 for serious gamers and the like who want standard nVidia features like SLI. This would be a bad blow for nVidia at one of its core markets
I think you guys will be ok.. Nvidia will no doubt eventually give in and just pay Intel whatever they want.. They can't afford to lose all the integrated GPU platforms..
Dan,
Your writing is thorough and thoughtful, as usual.
,dave
I completely agree. I always come to Appleinsider for some excellent in-depth analysis of all things Mac. Thank you Dan and A.I.!
I agree that this argument isnt much about equipment that Apple will use.
However this is more troublesome for all PC users who like to build their own systems. The idea of not being able to use an nVidia motherboard for a core i7 system is disturbing as this rules out the use of core i7 for serious gamers and the like who want standard nVidia features like SLI. This would be a bad blow for nVidia at one of its core markets
nVidia has now licensed SLI to work on Intel chipsets just like ATI has long done with Crossfire. So now Intel's chipsets have the advantage of being compatible with both SLI and Crossfire, while ATI and nVidia's chipsets only support their own technology.
"Well that is the hope since DMI is to low-bandwidth to hang an IGP off of and QPI equipped CPUs aren't the right market for IGPs."
Wikipedia says the newest DMI is basically a modified PCIe x4 link running at 10Gb/s (1.25GB/sec). I'm wondering how much bandwidth an X4500-class integrated GPU would need to perform well? Is it really not feasible at all for them to connect one through the southbridge over DMI?
An IGP is not the same as a discrete GPU connecting through PCIe x4. As I understand things, as GPU process data they need a place to hold information to serve as inputs or to temporarily store their outputs before they are reprocessed. Discrete GPUs have dedicated VRAM to store information and the PCIe link only serves to an input source from the CPU and HDD. In an IGP, it relies on system memory as a cache and it needs to be constantly reading and writing to it, which is why IGPs are slower since they are competing with the CPU for memory bandwidth.
Normally, the IGP and memory controller are both on the northbridge of the chipset so the IGP technically has direct access to the memory, while the CPU has to jump through the FSB to access memory. This has been the common complaint with the FSB, because it can often be the bottleneck for CPU performance. In current MacBook and MacBook Pros, the FSB is 1066MHz which equates to 1 channel of DDR3-1066 memory with dual channel largely being wasted since the FSB basically can't allow anymore information through. The whole point of integrating the memory controller on the CPU like in AMD chips and in Nehalem is to remove the FSB bottleneck and allow the CPU to take advantage of the full dual or triple channel bandwidth. (It's even more important in multiprocessor systems like the Mac Pro so that memory bandwidth scales linearly with CPU count since each CPU has their own memory pool.)
The disadvantage of integrated memory controllers is that an external IGP would now have to be 1 hop away from system memory and be bottlenecked by the bus interface, in this case DMI. Anytime the IGP reads or writes something it'll have to go through the DMI link and 10Gb/s or 1.25GB/s is not enough considering the current 9400M has direct access to the northbridge memory controller and has the full 17GB/s of dual channel DDR3-1066 available. What's more, the DMI link is used for all peripherals too like the HDD, USB, ethernet, etc to reach the CPU for processing. DMI just doesn't have the bandwidth if the IGP relies on it exclusively for memory access.
"I don't doubt that Apple will keep their options open. Realistically, ATI is a bigger supporter of OpenCL, since they've said they will support standards like OpenCL and DX11 Compute Shaders over proprietary GPGPU frameworks. nVidia however, is going to continue to promote their proprietary C for CUDA language alongside OpenCL and DX11 CS. "
Do you know how nVidia is going to be pursuing this? Are they writing a CUDA->OpenCL compiler, using an abstraction layer, etc? I haven't fiddled with programming any of the GPGPU stuff yet....
Thanks
You'll need to parse nVidia's marketing. Before there were standardized OpenCL or DX11 Compute Shaders, I believe nVidia just referred to all GPGPU operation on their GPUs as CUDA. They now basically break things down into a language and an interface to convert that language into the machine code that the GPU hardware understands. OpenCL and DX11 CS are two languages, and nVidia now specifically refers to C for CUDA as their proprietary language. CUDA itself is going to be the interface layer to convert GPGPU languages, whichever they are, to something the GPU will understand.
ATI is similar. Their proprietary language is Brook+, and Stream is their broader label for interfacing with the GPU.
Anytime the IGP reads or writes something it'll have to go through the DMI link and 10Gb/s or 1.25GB/s is not enough ...
So I guess the good news is that the only likely integrated GPUs of Nehalem-series chips will be in the budget dual-core models with the on-package GPUs (with direct access to the dual-channel DDR3). So someone like nVidia or ATI can't bolt on a crappy integrated GPU onto the quad-core clarksfield and lynnfield platforms.. Hopefully Apple expands their use of discrete (and more powerful) GPUs in the future with the release of SnowL..
With Quick Path the distinction as to what may be considered a integrated GPU will likely become blurred. The thing here is that both Apple and Nvidia would likely want the GPUs to plug directly into Quick Path. That to keep the GPU as close to system memory as possible. Even then I'd expect a Quick Pathe connected GPU to have local memory so they may look more like a descrete GPU no matter the performance. I just have a hardtime seeing a GPU using Quick Path for all its memory accesses, it is not like it would be the only user of the bus.
I can actually see two classes of Quick Path using GPUs. One would be the high integration device that is thought of as an integrated chip today. That is a device with the GPU and all your I/O ports onboard. For the higher performance market I could see a Quick Path connected GPU that is much higher performance and passes the I/O thru to a south bridge. In essence if you are a GPU maker the performance of Quick Path is hard to ignore. If you are Apple you will want that mass of vector processors close to the memory they will work on.
As to what intel thinks it will ship between the Quick Path and DMI ports that has yet to be seen. Quick Pqth does have huge advantages and I can see intel getting all worked up over the DMI interface and then having it's various vendors dismiss it outright. For some apps the DMI based systems will be simple but OK for the intended uses. As a mother board maker though will you have enough options to produce outstanding product. I suspect that if Intel where to allow it to grow Quick Path could have a huge market with respect to support chipsets. It looks like Intel is working overtime to manipulate the market here.
Which brings us to AMD. Yeah their processors are a little slow but maybe not as bad as some think. If intel leaves them a big hole with respect to high speed interfacing to a GPU AMD will fill it. AND already has plenty of experience with on board memory controllers and high speed interfaces, with ATI they also have good GPU tech. It won't take much effort on their part to do an integrated, on the processor die, GPU. One should have much trouble imagining that this would be faster than the Intel solution and supportive of OpenCL. What I'm saying is that even though AMD lags a bit they could still force Intels hand via a much better processor with onboard video. Better in the sense that the GPU is often more important than the i86 unit. So what I'm saying is that it is not a sure thing that intels processor + GPU option will have the market share intel hopes for.
Part of this article is BS in my mind as it speculates as to what is holding up Apples new hardware. None of this speculation about i7 has anything to do with the Mini. At least I've not heard of anyone saying the Mini will get an i7 class processor. The iMac is a different story though as there is more than a good chance i7 could go in there. I just don't see Apple holding up one machine, a low cost one at that, due to issues on another.
As to people thinking Apple has painted themselves into a corner here with Nvidia, I see nothing that implies that. If they went to ATI they likely would have had the same issues. The other problem is that there will be years ahead of us before the GPUs in processor + GPU combos catch up with GPUs outside the main chip. If they ever catch up, though they will be certainly good enough for many apps, Apple won't enhance it's reputation as a producer of great graphics hardware by shipping huge quanities of them.
Dave
....The GMA X4500 is still about twice faster than the GMA X3100...The IGP on Clarksdale and Arrandale will be built on a 45nm process. I think the most likely configuration will be 16 stream processors clocked at 1066MHz (8 times Nehalem's CPU bus of 133MHz). That's probably conservative and achievable, which is good since this is the first time an IGP has been put on a CPU even if it is on package as a separate die from the CPU. This configuration won't beat the 9400M but will come close, especially if Intel is serious about their drivers which you'd think they will be in the lead-up to Larrabee. Being close to the CPU, and the raw power of Nehalem will only help performance. Intel was also on the working group for OpenCL and made a whole bunch of flowery talk about it in press releases so it's very likely future IGP's will have OpenCL support... I'm sure if Apple put up enough fuss, Intel may even make versions with overclocked IGPs, like how current iMacs use overclocked Santa Rosa platform chipsets and CPUs specifically made for Apple rather than the newer Montevina platform chipsets and CPUs...
....as Itcommander said, it should be close to the 9400, I don't think anyone really knows the details. The good part is that the chips not only have the iGPU on the same CPU package so the connection is much faster than the old FSB, but likewise the iGPU has direct access from the CPU to two channels of fast DDR3 memory...
OK, thanks for the info.
To me though, my intepretation of this is that towards the 2nd half of 2009 Intel will have an integrated graphics solution that maybe comes close to what Nvidia had in October 2008.
So we still don't know if Intel can lift its GPU game.
Maybe if Intel holds out against Nvidia maybe this will force Apple to bite the bullet and go full discrete graphics across the Mac line by this 2009's holiday season. Nehalem duals and quads all with discrete GPUs, Snow Leopard, OpenCL, GrandCentral... would be very, very sweet.
*HOWEVER* with the current economy, that dream for 2nd half of 2009 may not materialise. Apple needs to shift its strategy for 2009 and 2010 now, so... who knows what we will see out of them.
...None of this speculation about i7 has anything to do with the Mini. At least I've not heard of anyone saying the Mini will get an i7 class processor. The iMac is a different story though as there is more than a good chance i7 could go in there. I just don't see Apple holding up one machine, a low cost one at that, due to issues on another...Dave
That's what I've been saying. Even with the iMac, think about the 20" and can you imagine what it would take to get a Core i7 in there? Nothing less than some stunning, margin-gobbling engineering.
I mean, the HP Firebird is pretty impressive and it's only a Core 2 Quad and still requires liquid cooling.
If Apple goes liquid cooling on a Core i7 iMac 22" and 26" or something like that, that would be stunning, but the economic conditions in 2009 and 2010, just don't make that feasible no matter what the price. Not without Steve actively in the picture.
.... With Quick Path the distinction as to what may be considered a integrated GPU will likely become blurred. The thing here is that both Apple and Nvidia would likely want the GPUs to plug directly into Quick Path. That to keep the GPU as close to system memory as possible...
Now I am not going to say it's impossible, but at this time it's surely a stretch to envision a GPU that connects through the quickpath interface. There are many reasons for this:
1) Perhaps most relevant, The PCIexpress bus is plenty capable of continuing to handle the job. PCIe 2.0 with a standard x16 connector moves over 8.0 GB/sec, and even the latest ridiculously high-end dual-chip GPUs didn't saturate the old PCIe 1.1 standard with 4.0 GB/sec. The PCIe 3.0 spec which will be finalized this year will bump that up to 16.0GB/sec. The high memory bandwidth requirements of modern GPUs is predominately between the GPU and the dedicated GDRAM on the card itself.
2) As i mentioned in my other post, the vast majority of future Nehalem systems will *NOT* even have quickpath. Only the high-end Bloomfield CPU/X58 platform and the Xeon/server platforms even have Quickpath. All the future consumer/enthusiast focused Nehalems will have the northbridge/PCIexpress bus built right into the CPU die itself and will connect straight to the southbridge chip via DMI --- thus no quickpath.
Part of this article is BS in my mind as it speculates as to what is holding up Apples new hardware. None of this speculation about i7 has anything to do with the Mini. At least I've not heard of anyone saying the Mini will get an i7 class processor. The iMac is a different story though as there is more than a good chance i7 could go in there. I just don't see Apple holding up one machine, a low cost one at that, due to issues on another.
Not only the Mini, it is *incredibly* unlikely the iMac would see an i7. The TDP and power usage is far higher than the currently used Core 2 duos, not to mention it hasn't even seen a low-power Core 2 Quad CPU yet. Also, as I mentioned above, the future Nehalems, namely the quad-core "Clarksfield" and "Lynnfield" are much better suited to the iMac as they use a lot less power and have a smaller 1-chip chipset/motherboard (no northbridge). For this reason I think we'll soon see a Core 2 Quad iMac update, and we won't see a Nehalem variant in the iMac until Q4 2009 or Q1 2010 (or whenever Intel has the new chips ready).
I mean, the HP Firebird is pretty impressive and it's only a Core 2 Quad and still requires liquid cooling....
While I agree that a 100W Core i7 won't see its way into the iMac, a Core 2 Quad would not require liquid cooling. The HP is almost certainly not using Intel's newest low-power variant of the Core 2 Quad, and probably using them at high-frequency too. It also is potentially more for cooling the GPU and reducing fan noise than anything else.
I'd bet we'll see 55W/65W TDP Core 2 Quad iMacs pretty soon, probably running at around 2.5Ghz.
http://download.intel.com/pressroom/..._WSM_Press.pdf
page 9, as you can see the DMI connection is to the southbridge only, while another bus comes out of the iGFX die to PCIe Gfx. The next slide (pg 10) clearly says switchable gfx supported, intel isn't dumb enough to force everyone who wants these to use their IGP but then require discrete gpus to go through DMI making said discrete gpu useless, the platform would fail.
Discrete graphics will be just fine on any new westmere-based Macs, hopefully I can resist the urge til then. Never liked apple but after using my gf's new aluminum wonder with the amazing multitouch gestures I'm converted. I found a way to get rid of OS X's wretched mouse acceleration so that's one less obstacle
Now the question is will the IGP be as fast as the 9400m because I don't see apple including a discrete gpu on the 13''...
This is reported in most of the news sources.
Therefore, it looks like this applies across the Nehalem line because they will all have integrated memory controllers.
Nvidia is definitely somewhat concerned because Bloomfield, Lynnfield and Clarksfield (in other words Nehalem 45nm) all have integrated memory controllers.
Firstly, Nvidia loses out on the "FSB" parts of the chipset like the memory controller that are now built into the CPU. This is 100% regardless of the squabble. Secondly, Nvidia may lose out on integrated graphics. Even worse. Thirdly is the southbridge (QPI or DMI...??not sure about this part) stuff at risk -possibly bad too for Nvidia.
The MAIN thing is...
For Bloomfield, Lynnfield and Clarksfield (in other words Nehalem 45nm), if the courts rule in favour of Intel and Intel-Nvidia don't hammer out another deal, it means that *any* "integrated graphics" on any Intel-CPU-based motherboard is an Intel. Regardless of whether this Intel integrated is in the CPU (which it isn't, until Westmere) or on the motherboard, if Nvidia is locked out, it is a big chunk of Nvidia's strategy at large risk.
Forget about Westmere 32nm towards the end of the year. We're talking Intel's latest and greatest these three quarters, January to September 2009. What are the best graphics Intel can offer right now? X4500HD with no real roadmap and nowhere near ATI and Nvidia advances.
So the real risk to Apple is not the iMac which wouldn't have integrated graphics anyway or core i7, it is *the very next refresh of the Aluminium MacBook line... in otherwords, between now and September 2009, if Apple goes Nehalem in mobile, they might have to kiss the 9400M/Nvidia chipset goodbye, or, stick with Core 2 for the next 9 months.
And while this legal stuff goes on for a few months, again, Apple in the same position, forget the iMac, talkin' about the Aluminium MacBooks here... No NEHALEM. NO NEHALEM IN MACBOOK PRO until/ unless Apple goes back and throws out the 9400M/Nvidia-whatever chipset.
ltcommander.data you make some really good points but discrete gfx on mobile Westmere have nothing to do with DMI. The IGP part of the MCM holds the Intel GPU, the integrated DDR3 controller, and a PCIe 2.0 controller, almost surely 16x. Any discrete gpu on the Westmere platform will be connected directly to the cpu through a PCE 2.0 16x bus. Not surprising since Lynnfield will also have a PCIe controller on the cpu, as will all future intel cpu's except the high-end socket 1336 x58 platform, it needs more than 16x. After the IMC, it was the next logical move on the way to total SoC.
Yep, the on-CPU northbridge will have at least 16 PCIe 2.0 lanes.
ltcommander.data you make some really good points but discrete gfx on mobile Westmere have nothing to do with DMI. The IGP part of the MCM holds the Intel GPU, the integrated DDR3 controller, and a PCIe 2.0 controller, almost surely 16x. Any discrete gpu on the Westmere platform will be connected directly to the cpu through a PCE 2.0 16x bus. Not surprising since Lynnfield will also have a PCIe controller on the cpu, as will all future intel cpu's except the high-end socket 1336 x58 platform, it needs more than 16x. After the IMC, it was the next logical move on the way to total SoC.
http://download.intel.com/pressroom/..._WSM_Press.pdf
page 9, as you can see the DMI connection is to the southbridge only, while another bus comes out of the iGFX die to PCIe Gfx. The next slide (pg 10) clearly says switchable gfx supported, intel isn't dumb enough to force everyone who wants these to use their IGP but then require discrete gpus to go through DMI making said discrete gpu useless, the platform would fail.
Discrete graphics will be just fine on any new westmere-based Macs, hopefully I can resist the urge til then. Never liked apple but after using my gf's new aluminum wonder with the amazing multitouch gestures I'm converted. I found a way to get rid of OS X's wretched mouse acceleration so that's one less obstacle
Now the question is will the IGP be as fast as the 9400m because I don't see apple including a discrete gpu on the 13''...
My point was never the inability to add a discrete GPU to any Nehalem or Westmere parts because as you noted they have integrated PCIe controllers. The AppleInsider article was concerned about nVidia IGPs on nVidia chipsets, and my point is that even if nVIdia got a license to produce chipsets for Nehalem, those chipsets aren't likely to have viable IGPs like the current 9400M IGP/chipset because the memory controller is no longer on the chipset and an external IGP will need to connect through DMI which as you note is designed for southbridges and doesn't have the bandwidth. My suggestion was that nVidia focus on fast discrete low-cost GPUs, since those will also be faster than an IGP, not necessarily that much more expensive, and will benefit from the integrated PCIe controller allowing the GPU to be closer to the CPU.