PowerVR going DX11 ?

[tangey]

A couple of months ago, IMG put a DX11 driver job vacancy on their website. I thought at the time it was strange, 'cause first of all they don't have any public DX11 hardware, and second, what's the relevance of DX11 with the mobile segment they are in.

Well, there are now 3 DX11 driver design careers on on the website, so something is moving afoot.

Also, they have recently started advertising vacancies for positions based in Poland (there are now 6-7 positions, including DX10 and DX11 driver engineers). My understanding is that they don't have a place in Poland. However one of their "technology partners", Mobica does have R&D facilities in Poland, and does include IMG as one of their customers, and Mobica does include driver development in their portfolio.

Does either windows CE or the window phone OS's have DX11 in their road map ?

 

[Panajev2001a]

A couple of months ago, IMG put a DX11 driver job vacancy on their website. I thought at the time it was strange, 'cause first of all they don't have any public DX11 hardware, and second, what's the relevance of DX11 with the mobile segment they are in.

Well, there are now 3 DX11 driver design careers on on the website, so something is moving afoot.

Also, they have recently started advertising vacancies for positions based in Poland (there are now 6-7 positions, including DX10 and DX11 driver engineers). My understanding is that they don't have a place in Poland. However one of their "technology partners", Mobica does have R&D facilities in Poland, and does include IMG as one of their customers, and Mobica does include driver development in their portfolio.

Does either windows CE or the window phone OS's have DX11 in their road map ?

Intel IGP's do count .

 

[Ailuros]

Aside from their possible plans for future generations, isn't the DX9.0 Level 3 compliance in SGX544 a DX11 thingy?

 

[Deleted member 13524]

Well, Intel will have to do DX11 eventually, and tesselation is a feature that kinda makes sense in mobile graphics.

 

[tangey]

Am I right in thinking that Intel has no DX11 hardware, even the sandybridge integrated graphics ?

 

[rpg.314]

None that has been publicly disclosed so far.

 

[Ailuros]

Well, Intel will have to do DX11 eventually, and tesselation is a feature that kinda makes sense in mobile graphics.

Eventually it's a direction the embedded space will take with or without Intel. DX11 isn't just about tessellation, as the requirements are quite high compared to what today's mobile GPUs really incorporate. SGX has already a tessellation pipeline, for which I figure to reach DX11 compliance there shouldn't be all too difficult.

Today the SGX variants that have been integrated have all TMUs with a max texture size of 2k*2k. Imagine how it'll look like if we'd be talking about 8k*8k and that's just one example out of many.

Am I right in thinking that Intel has no DX11 hardware, even the sandybridge integrated graphics ?

Which doesn't mean that Intel's GenX team hasn't been working on DX11 hw in the meantime. Does it have to be great?

 

[tangey]

Which doesn't mean that Intel's GenX team hasn't been working on DX11 hw in the meantime. Does it have to be great?

Of course they probably are, but its been suggested that IMG are often competing with Intel's own in-house graphics dept for various graphics seats. It would be a coup it IMG could offer DX11 compliance before Intel's own dept, it might win them a seat in a place they may not otherwise have gotten..., as is often the case, its all about having the right feature set/ power envelope at the appropriate time in the design cycle.

 

[Grall]

Today the SGX variants that have been integrated have all TMUs with a max texture size of 2k*2k. Imagine how it'll look like if we'd be talking about 8k*8k and that's just one example out of many.

What's the point of having hardware capable of addressing textures larger than the amount of available RAM or covering ~100x larger area than the screen resolution?

It may be in DX11 specs, but does it make sense in a mobile application? No.

 

[arjan de lumens]

What's the point of having hardware capable of addressing textures larger than the amount of available RAM or covering ~100x larger area than the screen resolution?

It may be in DX11 specs, but does it make sense in a mobile application? No.

There are use-cases for >2K textures, even in a mobile device:

• If the mobile device in question has a camera that can take pictures larger than 2Kx2K (~4MPixels), supporting >2K textures means that you can map a full picture onto a single texture, making any processing involving the GPU much easier to manage.
• Large texture atlases, potentially reducing the number of distinct texture objects and associated render state needed for game content.

A texture "larger than the amount of available RAM" would actually be pretty damn big on a modern phone; e.g. assuming an Iphone 4 with 512 MB RAM, and say, PVRTC at 2bpp, such a texture would need to have a size of about 47Kx47K.

 

[Entropy]

To me it seems straightforward - DX11 is a desktop Windows OS feature. Since they supply Intel with graphics IP for their Windows OS systems, hiring DX11 driver guys seems like a given.

As for relevance of desktop WindowsOS DX11 for mobile devices? I can't really see it. Ballmer may though.

 

[tangey]

To me it seems straightforward - DX11 is a desktop Windows OS feature. Since they supply Intel with graphics IP for their Windows OS systems, hiring DX11 driver guys seems like a given.

As for relevance of desktop WindowsOS DX11 for mobile devices? I can't really see it. Ballmer may though.

Yes, but IMG have no current DX11 hardware, and given intel is now integrating its own graphics IP on the chip with sandybridge, one might assume they had plans to move that down a few places to notebook/netbook/tablet over the next couple of years and go up featureset-wise to DX11.

However if IMG are working on DX11 hardware, then it leads to the conclusion that IMG will still feature in Intel chips running full DX11 windows, in 2-3 years time, whereas many were thinking by that time IMG would be in Intel's embedded/ce/utra-mobile segements only.

Unless DX11 is destined for mobile variations of windows (windows phone etc)

 

[Entropy]

Yes, but IMG have no current DX11 hardware, and given intel is now integrating its own graphics IP on the chip with sandybridge, one might assume they had plans to move that down a few places to notebook/netbook/tablet over the next couple of years and go up featureset-wise to DX11.

However if IMG are working on DX11 hardware, then it leads to the conclusion that IMG will still feature in Intel chips running full DX11 windows, in 2-3 years time, whereas many were thinking by that time IMG would be in Intel's embedded/ce/utra-mobile segements only.

Unless DX11 is destined for mobile variations of windows (windows phone etc)

The conclusion is that IMG will still, at least, compete for Intel integrated graphics.

 

[Ailuros]

Of course they probably are, but its been suggested that IMG are often competing with Intel's own in-house graphics dept for various graphics seats. It would be a coup it IMG could offer DX11 compliance before Intel's own dept, it might win them a seat in a place they may not otherwise have gotten..., as is often the case, its all about having the right feature set/ power envelope at the appropriate time in the design cycle.

Assume IMG announces its next generation IP in 2011 (purely speculative); when do you exactly think we'll see parts of that family integrated in real devices let alone any full DX11 versions of that?

Series5 started out with 530/535 (DX9.0) and despite there's a DX10.1 SGX545 variant the latter has been integrated where exactly until today?

Let me oversimplify things and say that SGX has a "base level" of SM3.0 GPU IP which can scale for the moment up to SM4.1 if a partner requests it. The fact that they actually went through the trouble for a 545 means that they have at least one licensee that wanted it. Irrelevant to that, they're scaling feature-sets exactly in order to save die area and in extension power consumption.

I don't see how their next generation would be any different in that regard. According to their own white-papers under 65nm@200MHz a SGX545/SM4.1 captures 12.5mm2 and a SGX543/SM3.0 2MP 16mm2. I'd rather have the latter and it's not too hard to guess that it has a better perf/mm2 and perf/W than the first.

As for Intel as I said since they managed to get out of GenX DX10 hw I don't see what would be so impossible or complicated for that team to also have designed DX11 hw for the future. If they'd use IMG IP it would of course be a very pleasant surprise, but I don't see on the other hand why Intel would be interested to pay as high royalty fees while they're still sustaining a graphics department.

What's the point of having hardware capable of addressing textures larger than the amount of available RAM or covering ~100x larger area than the screen resolution?

It may be in DX11 specs, but does it make sense in a mobile application? No.

arjan covered those questions quite adequately. Note that their ARM Mali GPUs (at least Mali200 and above) have TMUs with a max texture size of 4096*4096 afaik.

 

[Rys]

And it's also worth noting that 2Kx2K is not the maximum texture size on all SGX variants.

 

[tangey]

Assume IMG announces its next generation IP in 2011 (purely speculative); when do you exactly think we'll see parts of that family integrated in real devices let alone any full DX11 versions of that?

around 2-3 years, as per my previous post (#12)

Series5 started out with 530/535 (DX9.0) and despite there's a DX10.1 SGX545 variant the latter has been integrated where exactly until today?

One imagines that as the entire segment accelerates, the move to DX11 will not take anywhere as long as the move to DX10. IMG in my opinion are unlikely to be hiring a number of DX11 driver designers today, if they don't see DX11 on IMG hardware for 4-5 years.

I don't see how their next generation would be any different in that regard. According to their own white-papers under 65nm@200MHz a SGX545/SM4.1 captures 12.5mm2 and a SGX543/SM3.0 2MP 16mm2. I'd rather have the latter and it's not too hard to guess that it has a better perf/mm2 and perf/W than the first.

Indeed, however 545 also has full OpenCL profile (as opposed to embedded for 543) and has OpenGL3.1 compliance (as opposed to 2.1 for 543). So the enlarged die space is not solely for DX10.1 compliance. How much is for it, we don't know.

As for Intel as I said since they managed to get out of GenX DX10 hw I don't see what would be so impossible or complicated for that team to also have designed DX11 hw for the future. If they'd use IMG IP it would of course be a very pleasant surprise, but I don't see on the other hand why Intel would be interested to pay as high royalty fees while they're still sustaining a graphics department.

Which raises the question, why are IMG actively seeking 3 personal solely for DX11 driver development, unless DX11 is heading towards the mobile/embedded OSes.

 

[Ailuros]

And it's also worth noting that 2Kx2K is not the maximum texture size on all SGX variants.

Obviously. Otherwise SGX544 wouldn't be DX9 L3 or SGX545 DX10.1

around 2-3 years, as per my previous post (#12)

Time enough for the GenX team to have a DX11 solution on shelves. Note that I'm being merely negative to the option because Intel has been so far and I haven't seen a single indication yet from their side that points otherwise.

One imagines that as the entire segment accelerates, the move to DX11 will not take anywhere as long as the move to DX10. IMG in my opinion are unlikely to be hiring a number of DX11 driver designers today, if they don't see DX11 on IMG hardware for 4-5 years.

Of course will IMG need them anyway; as I said they'll even need them for SGX544 because of DX9 L3.

Indeed, however 545 also has full OpenCL profile (as opposed to embedded for 543) and has OpenGL3.1 compliance (as opposed to 2.1 for 543). So the enlarged die space is not solely for DX10.1 compliance. How much is for it, we don't know.

OGL3.2 but that's besides the point. My point is for the embedded space I'd rather have transistors invested in performance than too high feature-sets.

Which raises the question, why are IMG actively seeking 3 personal solely for DX11 driver development, unless DX11 is heading towards the mobile/embedded OSes.

While I don't know how Series6 really will look like I wouldn't be surprised that their base level is this time DX10 and they will be able to scale up to DX11. It goes without saying that they'll also need driver support in such a case. I still don't understand why any of it should be tied to Intel and more precisely anything above netbooks.

Hell we don't even know who in the end might have licensed SGX545. It looks likelier that Intel's next generation embedded SoC integrates SGX54x MP. Wouldn't you think that 545 should also come with DX10.1 drivers for whoever integrates it?

 

[Entropy]

Which raises the question, why are IMG actively seeking 3 personal solely for DX11 driver development, unless DX11 is heading towards the mobile/embedded OSes.

I quote from an old tweakers.net article on Windows Phone 7:

For 3D graphics, Windows Phone 7 by default uses Windows Direct3D 11, which is based on DirectX10 (previous versions of Windows Mobile used a mobile version of Direct3D). Device manufacturers will have to write their own 2D and 3D drivers.

The diagrams indicate that although the API will be Direct3D 11, the hardware itself will be only Direct3D 9-capable, and that Windows Phone 7 will use the same kind of Direct3D 10-on-9 capability that made its debut with Windows 7, and which similarly allows the Direct3D 10 and 11 APIs to be used on Direct3D 9 hardware.

When it comes to drawing conclusions from these job postings, it bears remembering that DX11 is Microsoft proprietary technology. They are big on, well, Windows PCs. They are miniscule in the mobile market, and have very capable competition.

I doubt that Windows Phone 7 will ever be other than a marginal player at best. And I find it hard to believe that Microsofts phone efforts would be the prime driver for DX11 capable graphics IP from IMG since their IP is already powering the platform. I'm sure they would like to be competing for inclusion in some or all of Intels Windows targeted CPUs, and to compete they've got to have something to offer. If you're hoping for DX11 driving IP development, I'd say the motivation lies there, rather than Windows Phone 7. Remember, Intel is a large shareholder of IMG. If Intel targeted investments were pointless, I think IMG might have known.

Somehow I fail to see the significance about DX11 specifically. IMG designs IP for their market, and so if you want to know where they'll go, well what would their largest customers want to see? And that has very little to do with DX11 in mobile space even technical reasons cited above aside. I'd rather ask where Khronos might be heading with OpenGL ES (and why).

 

[MfA]

Intel would probably be pretty happy if Microsoft developed a mobile phone version of Windows 7 proper.

 

[Entropy]

Intel would probably be pretty happy if Microsoft developed a mobile phone version of Windows 7 proper.

In order to compete with their own Windows Phone 7? Or should they ditch that effort too? I'm sure the shareholders would be thrilled and celebrate Ballmers genius. Besides, I think Intel would rather see Nokia push MeeGo as hard as possible.

Incidentally, a month ago Gartner released their predictions for the future marketshares of various mobile device OSs. Their long term prediction stretches to 2014, at which point they think the installed base of Windows Phone 7 will have grown to 3.9%. I think that's being overly optimistic, but regardless whether you give or take a factor of two, Windows Phone 7 is still not in a position to drive much of anything.