PowerVR Series 7

I would bet on a new sub-family of Series 6 with hardware support for the features we saw in the Android Extension Pack and are lacking in 6XT. That would be Tesselation and Geometry Shaders.

IMG was listed as a partner for developing AEP with Google, so this was a bit expected.
ARM should also be expected to announce an iGPU with support for these features in the near future, since they're in the same situation.

 

Pretty safe bet IMHO; why they have to call that though "Series7" or a "new generation" is obviously subject their marketing department's lack of imagination

Somewhat OT: http://www.imgtec.com/investors/presentations/24062014/Imagination_Full_Year_Results_2014.pdf Page 30 is relatively obvious, while Page 31 isn't. There's also a slide about their technology GPU roadmap in there which shows how "short" generations seem to be from now on.

 

We haven't developed new Series6 (6/6XT/6XE) cores just for AEP support.

We'll say more about Series7, AEP compatibility, and the new roadmap cadence, in the future.

 

AFAIK, Apple and Intel are the only 2 known implementations of G6430. If so then unless that die shot on page 30 is completed fabricated, which I severely doubt, it must be one of those two. Its not entirely clear which of the two die shots is supposed to be the one with G6430 (I'm guessing it's the top one), unless someone can recognise the competing die shot, which must be either a qualcomm chip, or a chip implementing Mali.

 

Yes, the top one definitely looks like Apple A7; cf this die shot.

 

Indeed it does, I'd forgotten that chipworks had done a die-shot.

Now assuming IMG didn't just go along with Chipworks analysis, then it confirms that chipworks correctly identified the graphics block.

 

Well.. Series 7 it is..

I guess this makes Series 6's lifespan a lot shorter than Series 5's.

 

http://www.imgtec.com/investors/presentations/24062014/Imagination_Full_Year_Results_2014.pdf

Page 32:

Both Series7 and Series8 are marked relatively short compared to previous generations.

 

A point not made in the presentation, but made in the follow-up conference audio, is that going forward, future generations should not be seen as replacements of the prior generation. Expectation is that series 6 and series 7 will actively be getting design wins concurrently.

 

I wonder if Wizard ray tracing will become a standard feature in Series7XT and/or Series8? At the very least it should feature more prominently than 1 model.

 

Is there more to AEP than upgrading the Opengl version and profile?

 

I am pessimistic that WIzard will ever be licensed by somebody with large volumes. Large enough to drive widespread hw accelerated ray casting.

It might get licensed here or there by some small ihv, but I am doubtful. Hope I am proven wrong.

If they had any licensees so far, I think they would have announced it.

 

I suspect that RT will continue as it is, one or two RT variants will be available in series7. No way is it going to be built in as standard. Series 7 extends both the upper and lower reaches of the performance envelope.

IMG have signalled that RT is still at early levels of engagement/discussion with customers.

 

Do any of the current PowerVR models support adaptive tesselation, or could that be one of the features of series 7?

 

Current Rogue cores range from DX9L3 to DX10.0 AFAIK.

 

to OpenGL ES3.1, AEP brings:
Tesselation
Geometry shaders
Compute shaders
ASTC texture compression

its basically a slap in the face to the morons of Khronos group who are too slow for GPU vendors and made a too much crippled ES profile

 

That's completely wrong and ignores the pace and complexity of hardware and software development, and the realities of when features were introduced.

ASTC is the prime example. It didn't firm up into something IHVs could build until well after ES 3.0 ratification had happened. Hardware shipping today still doesn't support the full ASTC spec, partially because of that.

The pace of core API development must best fit to the hardware, especially in the embedded world where you can't throw every feature that might be nice into the silicon. This is exactly why we have an extension mechanism that lets something like AEP exist.

Calling Khronos morons, when they're the ones who developed and agreed on the very API you're lauding, is just completely tone deaf to the whole process.

 

Isn't Neil Trevett still president for the Khronos group?

 

http://www.khronos.org/registry/gles/extensions/ANDROID/ANDROID_extension_pack_es31a.txt

For reference AEP includes 20 extensions of which 9 are official Khronos KHR or OES extensions. And AEP only includes partial support for ASTC by including the LDR variant but not the HDR variant, which I believe is what Rys is referring to in terms of limitations even in hardware that claim ASTC support.

Neil Trevett is also chair of the OpenCL working group and releases flowery PR messages when new OpenCL versions are released yet nVidia drivers still only support OpenCL 1.1.