What are the Pros & Cons of having a PowerVR GPU in a Next Gen Console?

If claims here in the handheld forum were correct the SGX543MP4+(@200MHz) in the PS Vita should be around 35mm2 at Samsung 45nm.



No idea yet about die area and/or power consumption considering Rogue. However a 500MHz frequency is quite modest for a hypothetical high end console design under 28nm. 500MHz for GPU blocks will be common place for 28nm/2012 smartphone/tablet designs. Tegra3 albeit being manufactured at 40nm TSMC should have in its tablet variant the GPU clocked already at 500MHz and I'd be very surprised if upcoming SoCs with single or MP SGX544's won't be clocked at least at 500MHz.

Reverse speculative math based on the ST Ericsson Novathor A9600: they're claiming >210 GFLOPs, >5 GTexels fill-rate (without overdraw) and >350M Tris. Assuming it's a quad core design, always per core a possible scenario would be 8 Vec5 ALUs, 2 TMUs per core clocked at 667MHz.

8 * 10 FLOPs * 0.667GHz = 53.36 GFLOPs
2 * 667MHz = 1.33 GTexels

53.36 * 4 cores = 213.44 GFLOPs
1.33 * 4 cores = 5.32 GTexels

...and that's still high end smart-phone/tablet ballpark for the A9600. If true you'd need roughly 37 cores to reach the 2 TFLOPs mark at 667MHz, or as an alternative clock at 1GHz and get away with 25 cores.

***edit: if you now would want to change VecX ALUs into the typical SP (stream processor) parlance of the desktop GPU world, you'd have for 25 cores = 1000 SPs and for 37 cores = 1480 SPs. I'm fairly sure that I'll be wrong at the end, but that's what speculative (reverse) math is good for.

Your numbers sounds right to me, but I was speculating based on information from John that If i'm not mistake works at IMG. And if we have 2 times power of SGX554 (72 flop / cycle per core) with perhaps the A9600 with 210Gflop are generated by two core "SGX600" 733MHz due to the size most appropriate for a cell phone or in extreme hypothesis 4 core at 366MHz.

If the rogue has the same size as a SGX543 about 8mm^2 per core at 65nm (or SGX543 speculating less than 3.5mm^2 at 28nm) with four times its power (better efficiency in the density of transistors etc.) maybe 32 cores series 6 are possible in 260mm^2 at 28nm processing each 144 flops per cycle at 500MHz or 2.3 Tflop reach the most optimistic of scenarios.
 
Albeit just a relatively minor update, I'd very surprised if it wouldn't be also up to DX11.1. Isn't DX11.1 checking whether there's a TBDR at work in order to offload the CPU from geometry sorting with early Z?

DX11.1 being an update meant for mobile devices (targeted by Windows 8), I would expect Rogue to support it.
 
I would hope console makers have the option of building prototypes before signing for a chip.
Best way to see if it works well is the Vita (which might be too early in its life cycle at this stage) and building a prototype to test.

wouldn't that be as simple as using a modified PowerVR series 5 with 32 cores or 2 - 16 core PowerVR series 5 running together to make a Prototype? or they might have the early series 6 prototypes.
 
Your numbers sounds right to me, but I was speculating based on information from John that If i'm not mistake works at IMG. And if we have 2 times power of SGX554 (72 flop / cycle per core) with perhaps the A9600 with 210Gflop are generated by two core "SGX600" 733MHz due to the size most appropriate for a cell phone or in extreme hypothesis 4 core at 366MHz.

Look again at the fill-rate ST Erricson states for the A9600. I'd like to stand corrected but I take it as granted that both SGX554 as Rogue cores have a 4:1 ALU/TMU ratio. Besides whether my frequency guess is right or wrong is completely irrelevant since in a smartphone your GPU might be restricted to burn only a couple of hundred milliwatts, while in a console obviously not.

If the rogue has the same size as a SGX543 about 8mm^2 per core at 65nm (or SGX543 speculating less than 3.5mm^2 at 28nm) with four times its power (better efficiency in the density of transistors etc.) maybe 32 cores series 6 are possible in 260mm^2 at 28nm processing each 144 flops per cycle at 500MHz or 2.3 Tflop reach the most optimistic of scenarios.

543 should be bigger than just 8mm2 unless you're not willing to count complimentary cores that come with it for multi-core configurations. Besides 543 isn't even DX9 L3 compliant and the die area for either SGX544 and 554 which are, are still unknown. The Rogue area is even darker beyond that.

However 260mm2 is whole damn lot under 28nm and I'd be very surprised if mainstream GPU chips from AMD end up larger than that, with a comparable TFLOP rate where then it'll come down that any theoretical figure like FLOPs aren't equal between architectures. Else speculative math leads only into dead ends. Point in case being that yes Rogue should theoretically be able to fit the next generation console GPU IP demands. Apart from what I can tell IMG hasn't landed neither at Microsoft nor at SONY any such deal. In the contrary case I'd be more pleasantly surprised than most of you guys here.
 
DX11.1 being an update meant for mobile devices (targeted by Windows 8), I would expect Rogue to support it.

OT: since I haven't talked to you in a long time, congratulations monsieur Ingenu ;)

Apple already use PowerVR

RudeCurve obviously knows that. He's probably hinting at Apple possibly thinking of some sort of console or better home entertainment device. If yes (since Apple's plans are blurrier than anyone else's) I'm not so sure that what they have in mind is really you could call a gaming console. If Apple is interested to invest more into 3D/gaming in the longrun their application store would leave a few very interesting opportunities open for future possibilities.
 
There can only be three things happening in this thread:

a) people are genuinely having a conversation about cakes, pies, bacon and brown sauce (ew)

b) JohnH and Rhys of the venerable IMG are rather cleverly and wryly leaking some epic, mega-tonne worthy, NDA-shafting, termination of contract-type stuff in here... and the majority of us are not seeing it :mad:

or

c) an extra terrrestrial self-aware computer virus has broken into B3D, hyjacked accounts and is now using food-code to predict the end of the world (wherein it will link every Windows, Android and iOS based device together to form skynet and commence judgement day).

I'm hoping for option c...
 
There can only be three things happening in this thread:

a) people are genuinely having a conversation about cakes, pies, bacon and brown sauce (ew)

b) JohnH and Rhys of the venerable IMG are rather cleverly and wryly leaking some epic, mega-tonne worthy, NDA-shafting, termination of contract-type stuff in here... and the majority of us are not seeing it :mad:

or

c) an extra terrrestrial self-aware computer virus has broken into B3D, hyjacked accounts and is now using food-code to predict the end of the world (wherein it will link every Windows, Android and iOS based device together to form skynet and commence judgement day).

I'm hoping for option c...

It unfortunately just sounds like option (a). Just for the record's sake they're not just two IMG folks throwing around with food in this thread :p
 
It unfortunately just sounds like option (a). Just for the record's sake they're not just two IMG folks throwing around with food in this thread :p

Given this has gone on for longer than a few days now & there are no signs of mods coming in to clean up the thread (given the sub-forum where it lives I highly doubt talk of cakes & bacon make particularly relevant subject matter...) I'd wager it's closer to (b) myself...
 
in my mind there is 2 different Companies planing to make Consoles or 'Console like boxes' with the PowerVR chips in them.


prototypes being , Vita & iPads
 
Given this has gone on for longer than a few days now & there are no signs of mods coming in to clean up the thread (given the sub-forum where it lives I highly doubt talk of cakes & bacon make particularly relevant subject matter...) I'd wager it's closer to (b) myself...

Well on a more serious note chances are high that PowerVR most likely will find itself in further future Apple devices. I just wouldn't call the result a gaming console per se. Other than that I don't see IMG having landed any real console deal. Unless of course you trust sites like BSN which suggested a while ago that the PS4 will contain a octacore PowerVR.
 
Given this has gone on for longer than a few days now & there are no signs of mods coming in to clean up the thread (given the sub-forum where it lives I highly doubt talk of cakes & bacon make particularly relevant subject matter...) I'd wager it's closer to (b) myself...
That is all.
Just to be clear, AlStrong has no knowledge of PowerVR bacon or brown sauce (which he thinks is made out of soy). And this thread wouldn't be complete without Fast Food Pizza Cake.
 
It unfortunately just sounds like option (a). Just for the record's sake they're not just two IMG folks throwing around with food in this thread :p
I read that as "IMG folks throwing around with food in this thread" and was about to call "Food Fight!" :mrgreen:

edit: Someone had the sense to spin off the cake talk, so let's get back to normal here...
 
Ok, I'll give it a go.

Bacon = Strips = IMG PowerVR processes in strips of pixels. Obvious.

Cake = Layers or stacks(pancakes) = Stack of PowerVR chips.

Pie = Filled with fruit = Apple Pie = Apple

Brown Sauce = Tough = Google = HP Sauce = High Performance (HP) = console = Aha!

So there you go, A high performanced, stacked, PowerVR chip for an Apple console!!

Now I'm hungry.
 
Back
Top