By: Brett (ggtgp.delete@this.yahoo.com), January 28, 2017 3:33 pm
Room: Moderated Discussions
NickK (redhavoc.delete@this.yahoo.co.uk) on January 23, 2017 2:36 pm wrote:
> > It's all very well analysing the docs you reference but in this
> > case, there is a much easier way to get the information.
> >
> > http://www.anandtech.com/show/8706/imagination-announces-powervr-series7-gpus-series7xt-series7xe
>
> From what I understand Apple has most, if not all, of the software development in house. A GPU being a
> difficult beast on its own I would not be surprised if ImgTech supplies some of the very low level interfacing
> software/driver of the GPU and Apple creates all the clever stuff on top (driver/compiler).
>
> Concerning the GPU hirings that Apple has done until recently, they were always high level architects/managers
> rather than people doing the dirty work of implementing a GPU. I would not be surprised if they were giving
> a direction in the software teams, and deciding the special 'ingredients' they wanted in their hardware.
> Similar to what Sony/Microsoft have done to their latest generation consoles with AMD.
OMG, you must be a PC guy and think Apple would CLONE an existing NVidea/AMD decade old now obsolete design. Thus Apple would need a bunch of mid-level and low level employees from NVidea/AMD.
Was the Macintosh a CP/M clone just like DOS? Was the Apple disk controller a copy of a typical disk controller? Was the iPad just like Windows Mobile? Was the iPhone just a clone of a flip phone?
Wavefront technology is not efficient, it was just least worst when dealing with memory that is ~80 cycles away. The iPad Pro is the only iDevice that renders from ram, all the other iDevices render from L3 cache. One huge massively ported cache that is far away is not great either, something crazy like distributed ram could work really well.
When/if Apple produces a GPU it will hit you like Apple's 64 bit ARM, you will look at the performance and die size and thermals and you will go WHAT THE F*CK IS THAT!?!
> Also any IP going into an apple product will need to be released about a year before it will make to the product
> itself. It needs to be implemented into a chip (it would take 2-3 of months once all IP is collected), manufactured
> in the Fab (another couple of months), and then build at least in a development platform for further software
> development, optimisation and testing. I would not be surprised if this testing development process takes about
> 5-6 months from the rumors I hear. Finally there needs to be a production ramp up of the final product before
> release so that they have adequate numbers, but that could overlap with the previous stage.
>
> Also Apple has the muscle and influence to dictate to ImgTech what becomes public and what
> not. They may also pay a premium for it, and apple is knows as paying well for services
> and silicon. I would not be surprised that the 'free' precision conversion is an exaggeration
> from apple, as some power would be spend, and possibly a cautious approach from ImgTech
> for the same exact reason. Both may see this from completely different angles.
>
> I would be brave and rule out 5s, 6, 6s and 7 as an apple designed GPU. given the recent hirings,
> I would not be surprised if they may have bought a license to get the a current design and start
> making their own derivatives, and potentially we could see their in-house GPU in the 2018 line.
> If they start from scratch then I would expect the in-house chip to be delivered 2019-2020.
>
> > It's all very well analysing the docs you reference but in this
> > case, there is a much easier way to get the information.
> >
> > http://www.anandtech.com/show/8706/imagination-announces-powervr-series7-gpus-series7xt-series7xe
>
> From what I understand Apple has most, if not all, of the software development in house. A GPU being a
> difficult beast on its own I would not be surprised if ImgTech supplies some of the very low level interfacing
> software/driver of the GPU and Apple creates all the clever stuff on top (driver/compiler).
>
> Concerning the GPU hirings that Apple has done until recently, they were always high level architects/managers
> rather than people doing the dirty work of implementing a GPU. I would not be surprised if they were giving
> a direction in the software teams, and deciding the special 'ingredients' they wanted in their hardware.
> Similar to what Sony/Microsoft have done to their latest generation consoles with AMD.
OMG, you must be a PC guy and think Apple would CLONE an existing NVidea/AMD decade old now obsolete design. Thus Apple would need a bunch of mid-level and low level employees from NVidea/AMD.
Was the Macintosh a CP/M clone just like DOS? Was the Apple disk controller a copy of a typical disk controller? Was the iPad just like Windows Mobile? Was the iPhone just a clone of a flip phone?
Wavefront technology is not efficient, it was just least worst when dealing with memory that is ~80 cycles away. The iPad Pro is the only iDevice that renders from ram, all the other iDevices render from L3 cache. One huge massively ported cache that is far away is not great either, something crazy like distributed ram could work really well.
When/if Apple produces a GPU it will hit you like Apple's 64 bit ARM, you will look at the performance and die size and thermals and you will go WHAT THE F*CK IS THAT!?!
> Also any IP going into an apple product will need to be released about a year before it will make to the product
> itself. It needs to be implemented into a chip (it would take 2-3 of months once all IP is collected), manufactured
> in the Fab (another couple of months), and then build at least in a development platform for further software
> development, optimisation and testing. I would not be surprised if this testing development process takes about
> 5-6 months from the rumors I hear. Finally there needs to be a production ramp up of the final product before
> release so that they have adequate numbers, but that could overlap with the previous stage.
>
> Also Apple has the muscle and influence to dictate to ImgTech what becomes public and what
> not. They may also pay a premium for it, and apple is knows as paying well for services
> and silicon. I would not be surprised that the 'free' precision conversion is an exaggeration
> from apple, as some power would be spend, and possibly a cautious approach from ImgTech
> for the same exact reason. Both may see this from completely different angles.
>
> I would be brave and rule out 5s, 6, 6s and 7 as an apple designed GPU. given the recent hirings,
> I would not be surprised if they may have bought a license to get the a current design and start
> making their own derivatives, and potentially we could see their in-house GPU in the 2018 line.
> If they start from scratch then I would expect the in-house chip to be delivered 2019-2020.
>
