Well that's not really the whole truth. The thing is if nvidia doesn't get a QPI licence they are out of buisness.
It ist true that the next generation mobile nehalems only connect everything on the die with the rest via DMI, but the rest isn't very much.
Intel integrates on die a memory controller, a GPU (G55) and PCIe x16 links in one Chip and as a second chip on the same die sits the chip with the two westmere cores. Both of these chips are connected via QPI. There ist absolutely no reason to use a nvidia southbridge over an intel one, thus there needs to be a special edition with only the Intel cpu on die and a nvidia chipset that connects with the CPU via QPI and provides all the remaining functionality.
If this edition doesn't come and/or nvidia mustn't make something with QPI, we end up with an Intel plattform where nvidia can only provide the switchable dedicated grafics. The onboard CPU will be a Intel G55.
It doesn't matter if nvidia can build a chipset using the dmi link, because that would be almost worthless unless they build a super low power version.
PS: I think we'd have already heard anything if there was some nvidia nehalem chipset in the making. The 32nm Nehalems will launch around Christmas and there is just not enough time to do anything but provide the dedicated GPUs for nvidia with all the legal things still not sorted out.
Maybe this means there will be dedicated GPUs in the 13" Model since Apple's marketing propably doesn't want to explain that the newest Models have less GPU Power. The G55 will be about 2 times as fast as the G45 which is closer but still not at the level of a 9400m.
I'm well aware that "Core i5" (or whatever its final name is) processors use QPI internally to connect cores and the on-package graphics (where applicable). But they use DMI to connect to an I/O chipset which is what Intel calls P55/P57/G55 (codename Ibex Peak). That is what Nvidia could produce an alternative for, and the only real improvement they could make would be an integrated GPU that is better than the one in the processor.
The replacement for MCP79 (the IGP component of which is 9400M), which Nvidia is calling ION2, will double the number of stream processors in the IGP to 32. If Nvidia makes an Ibex Peak competitor, its GPU will be at least as powerful. G55 will be fighting a losing battle on the performance front.
They are still playing catchup as high end ATI GPUs can do this quality already but I think they want to go it alone.
Manufacturers are optimizing algorithms and hardware to deliver the ideal of real-time photorealism. When any of them reach this goal in notebook hardware, the competition is over and not just for desktop computing but games consoles too.
Maybe 5 years from now this will happen, getting the equivalent of an R770 in integrated form.
That is an idea. But I doubt it will come till Christmas. There'd be some gossip.
I'm not entirly sure though if it is a smart idea. The silizium they use for southbridges is afaik more the low leakage optimized type, while the cpu uses the high clock speed optimized silizium. And the Board layout might be very different. Probably not a very simple way to go.
What you don't seem to be aware of is that a onboard GPU also needs fast access to the CPU and the memory. And the CPU needs fast access to the memory. DMI is sufficient for I/O stuff but not fast enough to handle memory access and CPU to GPU communication.
You can't just add a chipset with better IGP and connect it to the IMC+CPU via DMI. Maybe one could heavily utilise the energy sucking PCIe Links which isn't very efficient. And still there is the fact that almost the whole second intel chip is still needed while its Intel IGP sits around for nothing.
The only way an nvidia integrated GPU makes sense the 2nd one, which requires a QPI license.
I think the 3rd find its way into apples MBP >15,4" around Christmas.
What they will do with the 13" I don't know maybe ignore that an intel IGP will be a step back or they try to put in a little dedicated GPU or they just wait to update them later.
The G55 (probably X5500) will, according to intel, be around twice as fast as the X4500, which means it will still not be as fast as the current 9400m and definitely won't be a match for the upcoming ion2. Maybe they have a newer better architecture and they achieve the level of a 9400m this would solve Apple problems. For most people the current X4500 is a more than fast enough IGP.
As for Larrabee. This chip will first be desktop only I believe. It is way to flexible to be as efficient as the current nvidida architecture. I think we can wait for a 22nm shrink before we see this chip in a notebook.
QPI has been confused with DMI in one of my previous posts. Some how I thought DMI was the next generation interconnect.
QPI is what I believe Intel and Nvidia are arguing about. If Nvidia could license QPI that would certainly help with Graphics performance. Further Apple has to be very concerned that the out come of the arguement is positive for them.
On the otherhand I don't believe the mobile platforms are going to implement QPI initially. It is a bit of a mystery what is happening on the mobile front anyways as we have heard repeatedly that they are going directly to 32nm on the next generation mobile chips. So I wonder if any public info on the mobile line up is even valid.
The thing is if Apple / nvidia don't have mobile QPI chips to work with they are kinda screwed with respect to the current design formula for iMacs and Mini. So if there is no mobile with QPI then I expect a major overhaul of Apples products.
In the mean time the world plus dog is silent on ATI and their access to QPI. This idea that Nvidia is the only choice for low end "integrated" GPUs belies the fact that ATI is capable in this regard.
In any event if Intel can deliver 32nm mobile chips in the fourth quarter 2009 I will be impressed. It will basically be a greater than generation jump if Apple implements the chips.
Intel and nvidia are fighting about what the current license agreement includes. Intel thinks if there is no FSB but an integrated memory controller nvidia can't make a chipset under with the current license. Nvidia thinks different. Afaik are they still fighting about what the current license means, I don't know when either side gives up and they discuss a new agreement.
ATI/AMD is very much capable but I think they aren't interested in making Intel Chipsets anymore.
The new 32nm CPU will be awesome. About the energy consumption of the current chips but 4 threads and cores that work much faster on important software while not slower in less important stuff like games and such. And for the everyday guy the turbe mode is very different. If one core idles the other can go up in clock speed a significant step.
Clarksdale 2.0 Ghz Standard will overclock one core zu more than 3 Ghz.
Charlie says that nvidia have been kicked out from Apple, meaning that they have been excluded from new designs as punishment for their poor quality on the 8600 chips and for denying there were any problems when they knew very well there were.
He does not say that Apple have decided to redesign their entire product range around ATI chips as of today for no performance gain and at huge cost.
Fudo, who is a cretin of the highest order, claims that it can't be true that nvidia have been kicked out because Apple are still buying their chips.
Well, guess what, they're designed in. Apple have to keep buying the chips. That has no relation to whether nvidia will or will not be designed into any future products.
Think about this from nvidia's perspective... their reputation and share price is damaged by these claims and yet they haven't sued Charlie for defamation or forced him to take down the article. Much like they didn't over Charlie's articles at the Inq about solder bump quality because they couldn't, because they were true. Why not? Do they not like their shareholders or something?
I'd hazard a strong guess that Charlie is, for now at least, completely correct. There may well have been a fiery meeting as he suggests, especially now with the Steve back in town. But things may not be final and I'm sure that nvidia will be grovelling and trying to make amends. Well maybe not Jen-Hsung, but probably everybody else.
At the end of the day, until they develop internal high end graphics competence or bring Imagination up to speed, Apple can't really afford to kick out nvidia completely and effectively limit their choice to one company - if they do this then AMD will rub their hands and keep their prices high. They need nvidia there as leverage if nothing else.
nvidia know that and will use it.
But for nvidia, with Intel trying to take both their discrete GPU business through Larabee and their integrated business through on-chip GPUs and DMI, they just really can't afford to lose a key strategic customer like Apple.
And Apple know that too - plus they are seriously angry over the 8600 debacle.
I'd hazard a strong guess that Charlie is, for now at least, completely correct. There may well have been a fiery meeting as he suggests, especially now with the Steve back in town. But things may not be final and I'm sure that nvidia will be grovelling and trying to make amends. Well maybe not Jen-Hsung, but probably everybody else.
Except that where Steve Jobs' ire is concerned there's no such thing as a fiery meeting. There is simply Steve telling the other party that they screwed up and that's it. His willingness to take losses in order to send a message is legendary.
If Apple needs to put some sort of discrete graphics into every machine so they can dump nVidia they will. If they need to cripple some low end machines so they can dump nVidia they will. It's not like Apple really cares about performance anyway. If they did their mainstream desktop computer (iMac) wouldn't still be slower than cheap hackintosh machines from 2007.
Having said that I don't think Apple is dumping nVidia any time soon. As usual the rumors probably contain half truths and exaggerations. I believe the real battle, and the only one with long term implications, is the one between Intel and nVidia for the right to make QPI compatible parts.
Comments
Well that's not really the whole truth. The thing is if nvidia doesn't get a QPI licence they are out of buisness.
It ist true that the next generation mobile nehalems only connect everything on the die with the rest via DMI, but the rest isn't very much.
Intel integrates on die a memory controller, a GPU (G55) and PCIe x16 links in one Chip and as a second chip on the same die sits the chip with the two westmere cores. Both of these chips are connected via QPI. There ist absolutely no reason to use a nvidia southbridge over an intel one, thus there needs to be a special edition with only the Intel cpu on die and a nvidia chipset that connects with the CPU via QPI and provides all the remaining functionality.
If this edition doesn't come and/or nvidia mustn't make something with QPI, we end up with an Intel plattform where nvidia can only provide the switchable dedicated grafics. The onboard CPU will be a Intel G55.
It doesn't matter if nvidia can build a chipset using the dmi link, because that would be almost worthless unless they build a super low power version.
PS: I think we'd have already heard anything if there was some nvidia nehalem chipset in the making. The 32nm Nehalems will launch around Christmas and there is just not enough time to do anything but provide the dedicated GPUs for nvidia with all the legal things still not sorted out.
Maybe this means there will be dedicated GPUs in the 13" Model since Apple's marketing propably doesn't want to explain that the newest Models have less GPU Power. The G55 will be about 2 times as fast as the G45 which is closer but still not at the level of a 9400m.
I'm well aware that "Core i5" (or whatever its final name is) processors use QPI internally to connect cores and the on-package graphics (where applicable). But they use DMI to connect to an I/O chipset which is what Intel calls P55/P57/G55 (codename Ibex Peak). That is what Nvidia could produce an alternative for, and the only real improvement they could make would be an integrated GPU that is better than the one in the processor.
The replacement for MCP79 (the IGP component of which is 9400M), which Nvidia is calling ION2, will double the number of stream processors in the IGP to 32. If Nvidia makes an Ibex Peak competitor, its GPU will be at least as powerful. G55 will be fighting a losing battle on the performance front.
I'm waiting for the news that Intel is buying out NVIDIA.
Intel look like they're not just stepping into the graphics stuff lightly. They've raytraced Quake Wars in real-time:
http://software.intel.com/en-us/arti...ts-ray-traced/
That wasn't on Larrabee but using 4 x Dunnington CPUs (possibly 6-core each).
They are working on optimizing graphics algorithms for Larrabee:
http://software.intel.com/en-us/articles/larrabee/
They are still playing catchup as high end ATI GPUs can do this quality already but I think they want to go it alone.
Manufacturers are optimizing algorithms and hardware to deliver the ideal of real-time photorealism. When any of them reach this goal in notebook hardware, the competition is over and not just for desktop computing but games consoles too.
Maybe 5 years from now this will happen, getting the equivalent of an R770 in integrated form.
(this is dashboard failing to draw properly) and later this:
A software update later and it's never happened again
Be proud. Having such a creative videocard
@Hatting
That is an idea. But I doubt it will come till Christmas. There'd be some gossip.
I'm not entirly sure though if it is a smart idea. The silizium they use for southbridges is afaik more the low leakage optimized type, while the cpu uses the high clock speed optimized silizium. And the Board layout might be very different. Probably not a very simple way to go.
@FuturePastNow
What you don't seem to be aware of is that a onboard GPU also needs fast access to the CPU and the memory. And the CPU needs fast access to the memory. DMI is sufficient for I/O stuff but not fast enough to handle memory access and CPU to GPU communication.
You can't just add a chipset with better IGP and connect it to the IMC+CPU via DMI. Maybe one could heavily utilise the energy sucking PCIe Links which isn't very efficient. And still there is the fact that almost the whole second intel chip is still needed while its Intel IGP sits around for nothing.
The only possible ways as I see it are:
Cores <-> QPI <-> Intel Controller hub/IGP/IMC <-> DMI,PCIe <-> nvidia GPU(with dedicated memory) & southbridge hybrid
Cores <-> QPI <-> nvidia allin one Chip like the current 9400
and the most likely
Cores <-> QPI <-> Intel Controller hub/IGP/IMC <-> DMI <-> Ibex Peak M
----------------------------------------------------------- <-> PCIe <-> dedicated nvdia/AMD GPU.
The only way an nvidia integrated GPU makes sense the 2nd one, which requires a QPI license.
I think the 3rd find its way into apples MBP >15,4" around Christmas.
What they will do with the 13" I don't know maybe ignore that an intel IGP will be a step back or they try to put in a little dedicated GPU or they just wait to update them later.
The G55 (probably X5500) will, according to intel, be around twice as fast as the X4500, which means it will still not be as fast as the current 9400m and definitely won't be a match for the upcoming ion2. Maybe they have a newer better architecture and they achieve the level of a 9400m this would solve Apple problems. For most people the current X4500 is a more than fast enough IGP.
As for Larrabee. This chip will first be desktop only I believe. It is way to flexible to be as efficient as the current nvidida architecture. I think we can wait for a 22nm shrink before we see this chip in a notebook.
QPI is what I believe Intel and Nvidia are arguing about. If Nvidia could license QPI that would certainly help with Graphics performance. Further Apple has to be very concerned that the out come of the arguement is positive for them.
On the otherhand I don't believe the mobile platforms are going to implement QPI initially. It is a bit of a mystery what is happening on the mobile front anyways as we have heard repeatedly that they are going directly to 32nm on the next generation mobile chips. So I wonder if any public info on the mobile line up is even valid.
The thing is if Apple / nvidia don't have mobile QPI chips to work with they are kinda screwed with respect to the current design formula for iMacs and Mini. So if there is no mobile with QPI then I expect a major overhaul of Apples products.
In the mean time the world plus dog is silent on ATI and their access to QPI. This idea that Nvidia is the only choice for low end "integrated" GPUs belies the fact that ATI is capable in this regard.
In any event if Intel can deliver 32nm mobile chips in the fourth quarter 2009 I will be impressed. It will basically be a greater than generation jump if Apple implements the chips.
Dave
Well it very much looks like they will launch 32nm parts in Q4 2009 but they won't be available in any significant number till Q1 2010.
Here is what the new chips will look like.
http://www.computerbase.de/bildstrecke/26106/8/
Intel and nvidia are fighting about what the current license agreement includes. Intel thinks if there is no FSB but an integrated memory controller nvidia can't make a chipset under with the current license. Nvidia thinks different. Afaik are they still fighting about what the current license means, I don't know when either side gives up and they discuss a new agreement.
ATI/AMD is very much capable but I think they aren't interested in making Intel Chipsets anymore.
The new 32nm CPU will be awesome. About the energy consumption of the current chips but 4 threads and cores that work much faster on important software while not slower in less important stuff like games and such. And for the everyday guy the turbe mode is very different. If one core idles the other can go up in clock speed a significant step.
Clarksdale 2.0 Ghz Standard will overclock one core zu more than 3 Ghz.
He does not say that Apple have decided to redesign their entire product range around ATI chips as of today for no performance gain and at huge cost.
Fudo, who is a cretin of the highest order, claims that it can't be true that nvidia have been kicked out because Apple are still buying their chips.
Well, guess what, they're designed in. Apple have to keep buying the chips. That has no relation to whether nvidia will or will not be designed into any future products.
Think about this from nvidia's perspective... their reputation and share price is damaged by these claims and yet they haven't sued Charlie for defamation or forced him to take down the article. Much like they didn't over Charlie's articles at the Inq about solder bump quality because they couldn't, because they were true. Why not? Do they not like their shareholders or something?
I'd hazard a strong guess that Charlie is, for now at least, completely correct. There may well have been a fiery meeting as he suggests, especially now with the Steve back in town. But things may not be final and I'm sure that nvidia will be grovelling and trying to make amends. Well maybe not Jen-Hsung, but probably everybody else.
At the end of the day, until they develop internal high end graphics competence or bring Imagination up to speed, Apple can't really afford to kick out nvidia completely and effectively limit their choice to one company - if they do this then AMD will rub their hands and keep their prices high. They need nvidia there as leverage if nothing else.
nvidia know that and will use it.
But for nvidia, with Intel trying to take both their discrete GPU business through Larabee and their integrated business through on-chip GPUs and DMI, they just really can't afford to lose a key strategic customer like Apple.
And Apple know that too - plus they are seriously angry over the 8600 debacle.
All very interesting stuff to watch from afar!
On my very limited sample of one (me)... I think this is an Apple graphics problem... rather than nvidia.... One hot day my ATI HD2600 did this
*snip*
A software update later and it's never happened again
As you say, a software update cured it... the nvidia problems were badly-built chips that failed after a few months.
Big difference.
I'd hazard a strong guess that Charlie is, for now at least, completely correct. There may well have been a fiery meeting as he suggests, especially now with the Steve back in town. But things may not be final and I'm sure that nvidia will be grovelling and trying to make amends. Well maybe not Jen-Hsung, but probably everybody else.
Except that where Steve Jobs' ire is concerned there's no such thing as a fiery meeting. There is simply Steve telling the other party that they screwed up and that's it. His willingness to take losses in order to send a message is legendary.
If Apple needs to put some sort of discrete graphics into every machine so they can dump nVidia they will. If they need to cripple some low end machines so they can dump nVidia they will. It's not like Apple really cares about performance anyway. If they did their mainstream desktop computer (iMac) wouldn't still be slower than cheap hackintosh machines from 2007.
Having said that I don't think Apple is dumping nVidia any time soon. As usual the rumors probably contain half truths and exaggerations. I believe the real battle, and the only one with long term implications, is the one between Intel and nVidia for the right to make QPI compatible parts.