The whole idea of OpenCL is that the developer just gives it work packets to execute and it distributes them transparently across all CPUs and GPUs in the system.
If Apple use Intel or ATI graphics and the graphics driver does not support OpenCL, then all work packets simply go to the CPU and the user knows no different. So the idea that OpenCL could be delayed by available GPUs is not quite right. It would just be slower is all.
Since OpenCL is a framework for executing programs (see wikipedia) over a CPU/GPU environment , how would Intel incorporate it? Its not a hardware standard.
It's an API. The framework/implementation compliant with that open standard for Intel is their problem.
How Apple leverages it in Cocoa Frameworks is done and their responsibility. The same goes for AMD and Nvidia.
If this continues on Apple should buy a chunk of AMD and invest in their ATi chipsets.
They can shoot a shot over both Intel and Nvidia's bows to grow the hell up.
AMD already has GPUs compliant with OpenCL in the pipeline, plus currently existing ones that has OpenCL and Streams.
Again, i have doubt on the hardware side of Info. But this time since i do not have concrete information i will not say Appleinsider is wrong.
Personally i do not believe Corei7 will be in next iMac update.
1. TDP too high, iMac are based mostly on Notebook parts, and Nehalem Notebook are not coming till Q4 09 or Q1 10.
2. If they are solve the TDP problem with the rumors of Water Cooling solution. Corei7 is too expensive and it is a High End product rather then mainstream product.
3. Even if Nvidia HAD the right to produce chip for Corei7 now, it is still not yet available.
Intriguing article, I agree it is a good read... I must point out, however, as for the "current delay" of the iMacs, we are assuming several big things.
1. They were planning as early as mid 2008 on placing a core i7 into the iMac or "xMac"
2. If finalising core i7 in iMac by late 2008, they were able to do a lot of redesign
3. Following no.2, they sorted out a lot of thermal engineering
4. They decided to skip the very simple and obvious update to Core 2 Duo/ Core 2 Quad (desktop or mobile) with 9400M+9600
Again, i have doubt on the hardware side of Info. But this time since i do not have concrete information i will not say Appleinsider is wrong.
Personally i do not believe Corei7 will be in next iMac update.
1. TDP too high, iMac are based mostly on Notebook parts, and Nehalem Notebook are not coming till Q4 09 or Q1 10.
2. If they are solve the TDP problem with the rumors of Water Cooling solution. Corei7 is too expensive and it is a High End product rather then mainstream product.
3. Even if Nvidia HAD the right to produce chip for Corei7 now, it is still not yet available.
So a MacPro using i7? Yes. iMac? Highly Unlikely.
Agreed. See my post just after yours. Mac Pro will use the "Xeon" version of i7, that is, the dual-CPU version of Core i7 (there's another thread about this) to give 16 cores and supports DDR3 with ECC, and so on.
Unless the Mac mini suddenly will be able to handle the thermal output of the i5, there are no immediate problems there either, since it'll be using Core 2 processors.
All in all, this will have absolutely no effect on Apple until Intel releases a Nehalem derived processor with low enough TDP to fit into the 13.3" Macbooks and Mac Minis, which is at the end of this year at the very earliest.
Also, since when is QuickPath the memory interface connection in Nehalem?
QuickPath (QPI) is only applicable to Mac Pro and XServe, both of which should see update once Intel releases Nehalem-based Xeons in March. Neither Core i5 (Clacksfield) nor Core i3 (Arrandale) utilizes QPI.
Intel is making this into a mess, SHAME on you! Its your fault that Apple used Nvidia GPU in their MacBooks anyway...
Quote:
Originally Posted by nvidia2008
Exactly. The GMA950 and even X3100, were just so bad Apple gave up on Intel GPUs. Intel is really missing some "secret recipe" of GPUs.
This is really the crux of the matter. It was sheer arrogance on the part of Intel to think that Nvidia or someone like them wouldn't step ino the vacuum created from their continued neglect for on-board graphics. Worse, it was an insult to Apple to think the status quo was going to continue to be good enough. And now, instead of innovation, we get their legal team. Way to jump the shark, Intel.
I should point out that the fundamental assumption of this article, that the Intel's Nehalem chips that Apple will use will have QPI to connect to nVidia IGPs, is most likely flawed.
The current Nehalem processors out are Bloomfield cores (4 cores, 1 QPI, 3 channel memory) and are marketed under the Core i7 branding. It is intended as a high-end desktop chip and the chances of Apple using are extremely low, since Mac Pros use Xeon chips while iMacs use either mobile chips or perhaps will move to mainstream desktop chips.
Now the mainstream Nehalem quad core chip is going to be Lynnfield in desktop and Clarksfield in mobile. They are both quad core, 0 QPI, and 2 channel memory. Notice the key detail: 0 QPI. Mainstream desktop and all mobile chips won't have QPI and will only have DMI, because the entire northbridge, in this case memory controller and PCIe controller is integrated into Lynnfield and Clarksfield. DMI is designed as a low-bandwidth link to connect a northbridge to a southbridge. Lynnfield and Clarksfield won't have an Intel IGP on package, but you won't want to attach an IGP through the DMI link since it's low bandwidth, only 10Gb/s, and with the memory controller on the CPU, an IGP running through DMI would basically have 133MHz SDRAM like bandwidth.
Similarly, the only dual cores in the Nehalem architecture will be part of the Westmere 32nm shrink (Clarkdale for desktop and Arrandale for mobile) and will all include an Intel IGP. These will again have no QPI links and only a DMI link to a southbridge. Again, disabling the onboard IGP and using an external nVidia one will cripple it because DMI doesn't have enough bandwidth to the memory controller.
High-end Nehalems may have QPI which is suitable for an IGP, but it's pretty much pointless to give them an IGP. Hybrid Power or hybrid SLI may make sense for a notebook, but QPI won't appear in mobile chips. On a desktop, a decent dedicated GPU makes far more sense and would overpower an IGP so much that Hybrid SLI would add little performance. Afterall, a desktop 9800GTX can be found for $150 now and is a mainstream card, and trying to SLI a 9400 IGP to it isn't going to gain you much fps. So even if Apple used high-end desktop Nehalems, it doesn't make sense to be combining it with an IGP regardless or whether nVidia has one available or not.
For the Mac Pro, nVidia has never made chipsets for dual processor Xeons, so it really isn't unexpected that they don't make them for Nehalem Xeons with QPI either. Even if they did, like high-end Core i7, there is little benefit in giving a Mac Pro and nVidia IGP.
These plans have long been known. That virtually all Nehalem will not have high-bandwidth QPI links, regardless of whether they have Intel IGP, the included DMI link is unsuitable for an external IGP. It doesn't matter how hard nVidia argues to produce chipsets for Nehalem, it won't make QPI exist on chips that just don't have them, and remove the DMI bandwidth constraint for IGPs. Apple choosing to switch to nVidia chipsets was a dead-end that they should have realized and it was something that I mentioned in forum responses to the initial nVidia/Apple partnership.
If nVidia wants to make chipsets for Nehalem, they will essentially be making southbridges for DMI links. With P.A. Semi, Apple has enough experience to make southbridges themselves, and it makes more sense for them to anyways so that they can combine things like Firewire and touchpad controllers which are currently separate chips into the southbridge to save costs. Custom Apple designed southbridges would definitely go with Apple's statements of introducing features that competitors can't match. This is something that going with nVidia chipsets still doesn't allow since nVidia still sells to others.
Realistically, instead of trying to fight to make southbridges for DMI, which is no doubt a low-margin market, or making an IGP for QPI, which is pointless for high-end desktop processors, nVidia should be focused on making the fastest low-cost discrete GPU they can. Discrete GPUs will always be faster than an IGP that shares system memory. And with the PCIe controller now integrated into the CPU, Intel is actually doing a favour for GPU makers since it'll only increase graphics performance.
And as a note on anti-competition arguments, Intel not including QPI on all Nehalem processors and only including a low-bandwidth DMI link is not necessarily deliberately malicious. It just doesn't make sense to raise cost and waste transistors on a more complex high-bandwidth QPI link for all Nehalems, when they already have northbridges built-in and only need to attach to a southbridge which only requires low-bandwidth. nVidia argues for a 2 chips design with CPU + combined northbridge/southbridge, and Intel is adopting a 2 chip design, just that they are doing combined CPU/northbridge + a separate southbridge which is a higher performance approach that benefits customers.
Personally, I'd like to see the terms of the Intel-nVidia bus license.
Intel's pre-Nehalem processors used an AGTL+ FSB while Nehalem uses either QPI or DMI bus links. I would think nVidia's previous bus license would have been for the FSB, which would mean Intel is right that the license wouldn't extend to QPI or DMI since those are obviously quite different. If nVidia's bus license was some blanket bus license for general x86 processors made by Intel, then nVidia would be right and Intel's lawyers really messed up if they wrote such a wide-ranging contract.
Apple still has leverage with Intel. If they see that they cannot stay with NVIDIA for GPUs then they'll convince Intel to incorporate OpenCL. I'm sure Intel is already in the works in doing that if they haven't already done so. OpenCL is an open standard so why wouldn't they?
Even if they can do this, how is this "Apple having leverage with intel"???
I should point out that the fundamental assumption of this article, that the Intel's Nehalem chips that Apple will use will have QPI to connect to nVidia IGPs, is most likely flawed.
I am not debating your post, it is absolutely correct. I happened to write out a whole response before I read all the way through the thread, so I figured I'd post it anyways as perhaps it can add something to your already very detailed post.
------
Ok as usual, there are some important issues being missed with this article.
1) "with QuickPath integrated memory controllers"
The Quickpath point-to-point connection is used exclusively for CPU->Northbridge(one chip in the traditional "chipset") and CPU->CPU connections, it has nothing to do with the Integrated memory controllers on the new Intel CPUs. (The only time memory traffic travels over Quickpath is on a dual-CPU computer like the Mac Pro where each CPU has its own bank of memory and one of them has to access data stored in the other's memory (known as a NUMA architecture).
This is a VERY IMPORTANT distinction to be made. While the high-end desktop Nehalem platform codenamed "Bloomfield" and all the Nehalem-based server Xeon platforms use two-chip northbridge+southbridge chipsets that connect via the Quickpath interface, all of the future mainstream (lower-end/low-power) Nehalem parts and all laptop models DO NOT USE QUICKPATH AT ALL.
These parts, AKA the quad-core "lynnfield" and "clarksfield" CPUs, and the lower-end dual-cores known as "clarksdale" and "arrandale" (these have an on-chip GPU) use a new socket and platform in which the northbridge functionality has been integrated into the CPU itself and so the CPU connects directly to the southbridge (via a DMI connection), NO Quickpath involved. Meanwhile, The (expensive) Quickpath-using Bloomfield desktop platform is projected to barely surpass 1% of nehalem sales, and the Xeons will be exclusively used on servers and workstations, so the vast majority of Nehalem cpus (including those in the iMac, MB, MB Pro, Mini, etc) Apple uses will NOT even have quickpath hardware.
Now I am not positive if the Intel->Nvidia fight is purely over the Quickpath containing chipsets. It would seem to me that in the platforms where Quickpath is not even used (mainstream/lower-end chips which would be used in iMac, MB, Mini, etc) Nvidia would be in the clear to continue producing chipsets.
2) "Apple has a lot riding on the dispute, as its next iMac and Mac Pro are both expected to use Nehalem CPUs.
#The Mac Pro situation is easy... It is a dual-CPU Xeon workstation and Nvidia doesn't make these types of chipset anyways. Secondly, even if they did there is not much of an advantage as the Mac Pro doesn't use an integrated GPU.
The iMac situation is similar.. Even if Apple decided to use the current "Bloomfield" Nehalem platform which does use Quickpath (and a two-chip northbridge/southbridge chipset) and so was forced to use an Intel chipset, the iMac doesn't use an integrated GPU and so would not benefit from an nVidia chipset as a Macbook, MB Pro or Mini would.
3) "furthermore, a mobile version of Nehalem is expected next year"
Again, I am NOT positive about this, but based on the info presented it appears the lawsuit is solely about the Quickpath interface. ALL future Nehalem laptop chipsets are based on the new (yet-to-be-released) CPUs that integrate the northbridge (PCIexpress, DMI, etc) onto the processor die and connect directly to the southbridge --- NO quickpath involved. If that's the case, then there is nothing to worry about.
On the other hand, if Intel is fighting Nvidia or ALL Nehalem platform licensing -- including the sockets/platforms that connect solely via the southbridge (DMI) without Quickpath... well, that is another thing entirely and a much greater problem. Either way, I don't see this continuing for very long. Nvidia will just cough up more dough and the thing will be settled..
If nVidia wants to make chipsets for Nehalem, they will essentially be making southbridges for DMI links. With P.A. Semi, Apple has enough experience to make southbridges themselves, and it makes more sense for them to anyways so that they can combine things like Firewire and touchpad controllers which are currently separate chips into the southbridge to save costs. Custom Apple designed southbridges would definitely go with Apple's statements of introducing features that competitors can't match. This is something that going with nVidia chipsets still doesn't allow since nVidia still sells to others.
Realistically, instead of trying to fight to make southbridges for DMI, which is no doubt a low-margin market, or making an IGP for QPI, which is pointless for high-end desktop processors, nVidia should be focused on making the fastest low-cost discrete GPU they can. Discrete GPUs will always be faster than an IGP that shares system memory. And with the PCIe controller now integrated into the CPU, Intel is actually doing a favour for GPU makers since it'll only increase graphics performance.
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.
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.
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.
....These parts, AKA the quad-core "lynnfield" and "clarksfield" CPUs, and the lower-end dual-cores known as "clarksdale" and "arrandale" (these have an on-chip GPU)...
Do we know if those Clarksdale and Arrandale on-chip GPUs come even close to a 9400M level of performance?
And if Lynnfield and Clarksfield don't have on-chip GPUs/ integrated graphics maybe that's what Nvidia is trying to get at.
Even if Nvidia only gets the soutbridge DMI thingy, marketing the motherboard as "Nvidia chipset" is important to then push their on-board graphics and/or discrete GPUs ("this Nvidia motherboard works best with an Nvidia discrete GPU or "this Nvidia motherboard is perfect for slim profile HTPC with powerful onboard graphics")
I think this Intel-Nvidia thing has to do with the whole Nehalem line, not just QuickPath.
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.
Comments
If Apple use Intel or ATI graphics and the graphics driver does not support OpenCL, then all work packets simply go to the CPU and the user knows no different. So the idea that OpenCL could be delayed by available GPUs is not quite right. It would just be slower is all.
Since OpenCL is a framework for executing programs (see wikipedia) over a CPU/GPU environment , how would Intel incorporate it? Its not a hardware standard.
It's an API. The framework/implementation compliant with that open standard for Intel is their problem.
How Apple leverages it in Cocoa Frameworks is done and their responsibility. The same goes for AMD and Nvidia.
If this continues on Apple should buy a chunk of AMD and invest in their ATi chipsets.
They can shoot a shot over both Intel and Nvidia's bows to grow the hell up.
AMD already has GPUs compliant with OpenCL in the pipeline, plus currently existing ones that has OpenCL and Streams.
Also, I do hope that this is something that Intel and NVIDIA can work out soon. I've been holding my breath for this iMac refresh for way too long!
Personally i do not believe Corei7 will be in next iMac update.
1. TDP too high, iMac are based mostly on Notebook parts, and Nehalem Notebook are not coming till Q4 09 or Q1 10.
2. If they are solve the TDP problem with the rumors of Water Cooling solution. Corei7 is too expensive and it is a High End product rather then mainstream product.
3. Even if Nvidia HAD the right to produce chip for Corei7 now, it is still not yet available.
So a MacPro using i7? Yes. iMac? Highly Unlikely.
1. They were planning as early as mid 2008 on placing a core i7 into the iMac or "xMac"
2. If finalising core i7 in iMac by late 2008, they were able to do a lot of redesign
3. Following no.2, they sorted out a lot of thermal engineering
4. They decided to skip the very simple and obvious update to Core 2 Duo/ Core 2 Quad (desktop or mobile) with 9400M+9600
Intel is making this into a mess, SHAME on you! Its your fault that Apple used Nvidia GPU in their MacBooks anyway...
Exactly. The GMA950 and even X3100, were just so bad Apple gave up on Intel GPUs. Intel is really missing some "secret recipe" of GPUs.
Again, i have doubt on the hardware side of Info. But this time since i do not have concrete information i will not say Appleinsider is wrong.
Personally i do not believe Corei7 will be in next iMac update.
1. TDP too high, iMac are based mostly on Notebook parts, and Nehalem Notebook are not coming till Q4 09 or Q1 10.
2. If they are solve the TDP problem with the rumors of Water Cooling solution. Corei7 is too expensive and it is a High End product rather then mainstream product.
3. Even if Nvidia HAD the right to produce chip for Corei7 now, it is still not yet available.
So a MacPro using i7? Yes. iMac? Highly Unlikely.
Agreed. See my post just after yours. Mac Pro will use the "Xeon" version of i7, that is, the dual-CPU version of Core i7 (there's another thread about this) to give 16 cores and supports DDR3 with ECC, and so on.
Unless the Mac mini suddenly will be able to handle the thermal output of the i5, there are no immediate problems there either, since it'll be using Core 2 processors.
All in all, this will have absolutely no effect on Apple until Intel releases a Nehalem derived processor with low enough TDP to fit into the 13.3" Macbooks and Mac Minis, which is at the end of this year at the very earliest.
Also, since when is QuickPath the memory interface connection in Nehalem?
QuickPath (QPI) is only applicable to Mac Pro and XServe, both of which should see update once Intel releases Nehalem-based Xeons in March. Neither Core i5 (Clacksfield) nor Core i3 (Arrandale) utilizes QPI.
Intel is making this into a mess, SHAME on you! Its your fault that Apple used Nvidia GPU in their MacBooks anyway...
Exactly. The GMA950 and even X3100, were just so bad Apple gave up on Intel GPUs. Intel is really missing some "secret recipe" of GPUs.
This is really the crux of the matter. It was sheer arrogance on the part of Intel to think that Nvidia or someone like them wouldn't step ino the vacuum created from their continued neglect for on-board graphics. Worse, it was an insult to Apple to think the status quo was going to continue to be good enough. And now, instead of innovation, we get their legal team. Way to jump the shark, Intel.
The current Nehalem processors out are Bloomfield cores (4 cores, 1 QPI, 3 channel memory) and are marketed under the Core i7 branding. It is intended as a high-end desktop chip and the chances of Apple using are extremely low, since Mac Pros use Xeon chips while iMacs use either mobile chips or perhaps will move to mainstream desktop chips.
Now the mainstream Nehalem quad core chip is going to be Lynnfield in desktop and Clarksfield in mobile. They are both quad core, 0 QPI, and 2 channel memory. Notice the key detail: 0 QPI. Mainstream desktop and all mobile chips won't have QPI and will only have DMI, because the entire northbridge, in this case memory controller and PCIe controller is integrated into Lynnfield and Clarksfield. DMI is designed as a low-bandwidth link to connect a northbridge to a southbridge. Lynnfield and Clarksfield won't have an Intel IGP on package, but you won't want to attach an IGP through the DMI link since it's low bandwidth, only 10Gb/s, and with the memory controller on the CPU, an IGP running through DMI would basically have 133MHz SDRAM like bandwidth.
Similarly, the only dual cores in the Nehalem architecture will be part of the Westmere 32nm shrink (Clarkdale for desktop and Arrandale for mobile) and will all include an Intel IGP. These will again have no QPI links and only a DMI link to a southbridge. Again, disabling the onboard IGP and using an external nVidia one will cripple it because DMI doesn't have enough bandwidth to the memory controller.
High-end Nehalems may have QPI which is suitable for an IGP, but it's pretty much pointless to give them an IGP. Hybrid Power or hybrid SLI may make sense for a notebook, but QPI won't appear in mobile chips. On a desktop, a decent dedicated GPU makes far more sense and would overpower an IGP so much that Hybrid SLI would add little performance. Afterall, a desktop 9800GTX can be found for $150 now and is a mainstream card, and trying to SLI a 9400 IGP to it isn't going to gain you much fps. So even if Apple used high-end desktop Nehalems, it doesn't make sense to be combining it with an IGP regardless or whether nVidia has one available or not.
For the Mac Pro, nVidia has never made chipsets for dual processor Xeons, so it really isn't unexpected that they don't make them for Nehalem Xeons with QPI either. Even if they did, like high-end Core i7, there is little benefit in giving a Mac Pro and nVidia IGP.
These plans have long been known. That virtually all Nehalem will not have high-bandwidth QPI links, regardless of whether they have Intel IGP, the included DMI link is unsuitable for an external IGP. It doesn't matter how hard nVidia argues to produce chipsets for Nehalem, it won't make QPI exist on chips that just don't have them, and remove the DMI bandwidth constraint for IGPs. Apple choosing to switch to nVidia chipsets was a dead-end that they should have realized and it was something that I mentioned in forum responses to the initial nVidia/Apple partnership.
If nVidia wants to make chipsets for Nehalem, they will essentially be making southbridges for DMI links. With P.A. Semi, Apple has enough experience to make southbridges themselves, and it makes more sense for them to anyways so that they can combine things like Firewire and touchpad controllers which are currently separate chips into the southbridge to save costs. Custom Apple designed southbridges would definitely go with Apple's statements of introducing features that competitors can't match. This is something that going with nVidia chipsets still doesn't allow since nVidia still sells to others.
Realistically, instead of trying to fight to make southbridges for DMI, which is no doubt a low-margin market, or making an IGP for QPI, which is pointless for high-end desktop processors, nVidia should be focused on making the fastest low-cost discrete GPU they can. Discrete GPUs will always be faster than an IGP that shares system memory. And with the PCIe controller now integrated into the CPU, Intel is actually doing a favour for GPU makers since it'll only increase graphics performance.
And as a note on anti-competition arguments, Intel not including QPI on all Nehalem processors and only including a low-bandwidth DMI link is not necessarily deliberately malicious. It just doesn't make sense to raise cost and waste transistors on a more complex high-bandwidth QPI link for all Nehalems, when they already have northbridges built-in and only need to attach to a southbridge which only requires low-bandwidth. nVidia argues for a 2 chips design with CPU + combined northbridge/southbridge, and Intel is adopting a 2 chip design, just that they are doing combined CPU/northbridge + a separate southbridge which is a higher performance approach that benefits customers.
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
If this continues on Apple should buy a chunk of AMD and invest in their ATi chipsets.
They can shoot a shot over both Intel and Nvidia's bows to grow the hell up.
AMD already has GPUs compliant with OpenCL in the pipeline, plus currently existing ones that has OpenCL and Streams.
I like this suggestion. Sooner or later Apple is going to regret relying on only one
processor supplier. Also, if AMD does not survive, Intel will become even more
intransigent.
Intel's pre-Nehalem processors used an AGTL+ FSB while Nehalem uses either QPI or DMI bus links. I would think nVidia's previous bus license would have been for the FSB, which would mean Intel is right that the license wouldn't extend to QPI or DMI since those are obviously quite different. If nVidia's bus license was some blanket bus license for general x86 processors made by Intel, then nVidia would be right and Intel's lawyers really messed up if they wrote such a wide-ranging contract.
Apple still has leverage with Intel. If they see that they cannot stay with NVIDIA for GPUs then they'll convince Intel to incorporate OpenCL. I'm sure Intel is already in the works in doing that if they haven't already done so. OpenCL is an open standard so why wouldn't they?
Even if they can do this, how is this "Apple having leverage with intel"???
This is more like Apple being intel's bitch.
I should point out that the fundamental assumption of this article, that the Intel's Nehalem chips that Apple will use will have QPI to connect to nVidia IGPs, is most likely flawed.
I am not debating your post, it is absolutely correct. I happened to write out a whole response before I read all the way through the thread, so I figured I'd post it anyways as perhaps it can add something to your already very detailed post.
------
Ok as usual, there are some important issues being missed with this article.
1) "with QuickPath integrated memory controllers"
The Quickpath point-to-point connection is used exclusively for CPU->Northbridge(one chip in the traditional "chipset") and CPU->CPU connections, it has nothing to do with the Integrated memory controllers on the new Intel CPUs. (The only time memory traffic travels over Quickpath is on a dual-CPU computer like the Mac Pro where each CPU has its own bank of memory and one of them has to access data stored in the other's memory (known as a NUMA architecture).
This is a VERY IMPORTANT distinction to be made. While the high-end desktop Nehalem platform codenamed "Bloomfield" and all the Nehalem-based server Xeon platforms use two-chip northbridge+southbridge chipsets that connect via the Quickpath interface, all of the future mainstream (lower-end/low-power) Nehalem parts and all laptop models DO NOT USE QUICKPATH AT ALL.
These parts, AKA the quad-core "lynnfield" and "clarksfield" CPUs, and the lower-end dual-cores known as "clarksdale" and "arrandale" (these have an on-chip GPU) use a new socket and platform in which the northbridge functionality has been integrated into the CPU itself and so the CPU connects directly to the southbridge (via a DMI connection), NO Quickpath involved. Meanwhile, The (expensive) Quickpath-using Bloomfield desktop platform is projected to barely surpass 1% of nehalem sales, and the Xeons will be exclusively used on servers and workstations, so the vast majority of Nehalem cpus (including those in the iMac, MB, MB Pro, Mini, etc) Apple uses will NOT even have quickpath hardware.
Now I am not positive if the Intel->Nvidia fight is purely over the Quickpath containing chipsets. It would seem to me that in the platforms where Quickpath is not even used (mainstream/lower-end chips which would be used in iMac, MB, Mini, etc) Nvidia would be in the clear to continue producing chipsets.
2) "Apple has a lot riding on the dispute, as its next iMac and Mac Pro are both expected to use Nehalem CPUs.
#The Mac Pro situation is easy... It is a dual-CPU Xeon workstation and Nvidia doesn't make these types of chipset anyways. Secondly, even if they did there is not much of an advantage as the Mac Pro doesn't use an integrated GPU.
The iMac situation is similar.. Even if Apple decided to use the current "Bloomfield" Nehalem platform which does use Quickpath (and a two-chip northbridge/southbridge chipset) and so was forced to use an Intel chipset, the iMac doesn't use an integrated GPU and so would not benefit from an nVidia chipset as a Macbook, MB Pro or Mini would.
3) "furthermore, a mobile version of Nehalem is expected next year"
Again, I am NOT positive about this, but based on the info presented it appears the lawsuit is solely about the Quickpath interface. ALL future Nehalem laptop chipsets are based on the new (yet-to-be-released) CPUs that integrate the northbridge (PCIexpress, DMI, etc) onto the processor die and connect directly to the southbridge --- NO quickpath involved. If that's the case, then there is nothing to worry about.
On the other hand, if Intel is fighting Nvidia or ALL Nehalem platform licensing -- including the sockets/platforms that connect solely via the southbridge (DMI) without Quickpath... well, that is another thing entirely and a much greater problem. Either way, I don't see this continuing for very long. Nvidia will just cough up more dough and the thing will be settled..
If nVidia wants to make chipsets for Nehalem, they will essentially be making southbridges for DMI links. With P.A. Semi, Apple has enough experience to make southbridges themselves, and it makes more sense for them to anyways so that they can combine things like Firewire and touchpad controllers which are currently separate chips into the southbridge to save costs. Custom Apple designed southbridges would definitely go with Apple's statements of introducing features that competitors can't match. This is something that going with nVidia chipsets still doesn't allow since nVidia still sells to others.
Realistically, instead of trying to fight to make southbridges for DMI, which is no doubt a low-margin market, or making an IGP for QPI, which is pointless for high-end desktop processors, nVidia should be focused on making the fastest low-cost discrete GPU they can. Discrete GPUs will always be faster than an IGP that shares system memory. And with the PCIe controller now integrated into the CPU, Intel is actually doing a favour for GPU makers since it'll only increase graphics performance.
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.
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.
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.
....These parts, AKA the quad-core "lynnfield" and "clarksfield" CPUs, and the lower-end dual-cores known as "clarksdale" and "arrandale" (these have an on-chip GPU)...
Do we know if those Clarksdale and Arrandale on-chip GPUs come even close to a 9400M level of performance?
And if Lynnfield and Clarksfield don't have on-chip GPUs/ integrated graphics maybe that's what Nvidia is trying to get at.
Even if Nvidia only gets the soutbridge DMI thingy, marketing the motherboard as "Nvidia chipset" is important to then push their on-board graphics and/or discrete GPUs ("this Nvidia motherboard works best with an Nvidia discrete GPU or "this Nvidia motherboard is perfect for slim profile HTPC with powerful onboard graphics")
I think this Intel-Nvidia thing has to do with the whole Nehalem line, not just QuickPath.
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.