PowerVR Series 6 now official

[anexanhume]

What does under config core the 2nd digit stand for? TMUs?

Just looked at this again. On series 3, it broke it down as pixel shader: vertex shader: TMU : ROP. It condensed to two numbers on series 5 with a missing footnote. So, last number is TMU/ROPs? The first term is changing definition across generations, for sure.http://en.wikipedia.org/wiki/Render_Output_unit

 

[Ailuros]

There were no shader units at all on KYROs obviously.

 

[onQ]

 

[Ailuros]

http://semiaccurate.com/2013/03/12/imagination-shows-off-rogue-and-caustic-silicon/#.UT9F7TdWL8c

Charlie/SA on Rogue and Caustic.

 

[Exophase]

http://semiaccurate.com/2013/03/12/imagination-shows-off-rogue-and-caustic-silicon/#.UT9F7TdWL8c

Charlie/SA on Rogue and Caustic.

Good to see Charlie is still throwing silly insults at big.LITTLE rather than gaining an even menial understanding of what it actually is..

No real news on Rogue here. Maybe it'd be news if any of us knew that there were rumors of OpenCL problems to begin with. I wonder if it's really going to be only platform doing OpenCL "right", Mali-T604 seems like it can manage..

The claim that Caustic chose DDR2 to optimize latency is laughable. I suppose they also chose a 90nm process to optimize something? Even his claims that it's cheap and low power are wrong, since it's ostensibly more expensive than DDR3 (at least outside of China, and even there things probably started changing) and more power hungry for the same performance (1.8V vs 1.5V for starters). It uses DDR2 because it's old and desperately needs to be updated, not because it's preferable.

I don't see Caustic at this point being much more than a proof of concept, I doubt it'll be that useful in its current incarnation, but maybe it'll give people a platform to start developing for in the hopes that it'll get better.

 

[Rys]

Memory technology doesn't even really matter in the RT2 generation of the hardware, IMHO. As for the Rogue bits in the article -- while it's nice to see Charlie positive on the technology so maybe I shouldn't say this -- there are some pretty glaring mistakes. Rogue has never been demonstrated in FPGA form.

I think he's just completely confusing the Caustic RT2 FPGA and Rogue, but still.

And what does it even mean that before it existed in silicon we were supposedly the only company to get it right, but we'd also made severe architectural missteps?

I love people talking about our technology but it'd be nice if it was some way towards being correct. The dumb.Dumber jibe is just laughable.

 

[ams]

Rys, are you trying to say that he is a rogue? :smile: But seriously, information is good, but misinformation not so good

 

[tangey]

Rogue has never been demonstrated in FPGA form.

Well it seems to be a common misconception.

Brian Klug from anandtech seems to think he saw it demoed at ces:-
http://www.anandtech.com/show/6644/imagination-demonstrates-powervr-series-6-rogue-on-lgs-h13-soc

"Alongside that live demo on real production silicon was a four-cluster Series6 GPU being simulated on an FPGA inside a PC running much slower (obviously) thanks to very constrained memory bandwidth and clocks."


And your own alex says its demoing on an fpga in the Engadget video
"We've included our rogue 2 cluster chip inside an fpga...."
http://www.engadget.com/2013/01/03/powervr-series-6-test-chip/
(Around 1:30)

Off topic, why is it about half the time I try to go to semiaccurate, the web server doesn't respond, is his dns server semiaccurate too?

 

[Rys]

Easy to misunderstand I guess. On the test chip the FPGA chip is what connects it to the outside world (memory controller, PCIe, etc). Notice in the pics the board has two planes/PCBs? One has the FPGA with the connectivity, the other the 'real silicon' GPU test chip using that infrastructure.

 

[tangey]

Well, they were being told rogue was in an FPGA !
"We've included our rogue 2 cluster chip inside an fpga...."

 

[Ailuros]

Rys, are you trying to say that he is a rogue? :smile: But seriously, information is good, but misinformation not so good

Rys isn't affected due to the nature of his job, but ask yourself seriously how often and how many efforts IMG as a company really makes to inform websites properly for any underlying technology. When it then comes to misinformation who's then to blame really? Will someone contact SA and try to set the record straight for those few aspects? I'd like to be pleasantly surprised but I don't think so.

 

[tangey]

incidently, Brian Klug talked about seeing a 4-cluster rogue demo-ed, Alex in the video talked about a 2-cluster demo.

Unless you have two rogue demo platforms, one of them was wrong.

 

[Rys]

The system Alex is demoing is the same 4 cluster chip. An unfortunate mistake. In my personal opinion as an ex hardware graphics journalist, IMG should say more. As someone who works there I also respect the reasons why we don't.

 

[Ailuros]

The system Alex is demoing is the same 4 cluster chip. An unfortunate mistake. In my personal opinion as an ex hardware graphics journalist, IMG should say more. As someone who works there I also respect the reasons why we don't.

Selling IP is one of those awkward things where you have to be extremely careful what you say, where and why; that's perfectly understandable.

On the other hand I can also understand any journalist's side that probably wants to write something overall nice about N company; if then along the way there are mistakes or misconceptions there's no one to help him out either. And no it's not just Charlie in the given case, there are other examples too.

 

[3dcgi]

Selling IP is one of those awkward things where you have to be extremely careful what you say, where and why; that's perfectly understandable.

Why would you need to be any more careful when selling IP over say selling chips? I'd think if anything it would be the opposite, that IP vendors would talk more about their hardware than those that sell chips.

 

[Wellington]

http://gigaom.com/2013/03/18/ericss...son-divorce-terms-with-1600-jobs-on-the-line/

However, a spokewoman for Ericsson told me that these new chipsets — namely the the L8540 and L8580 lines — are being immediately discontinued. The ageing U-series, such as the U8500 that was announced in 2011 and that still powers midrange devices such as the Samsung Galaxy SIII Mini, will continue under STMicro’s auspices, she added

With STE dipping out of shipping Rougue, TI "refocussing" the OMAP business - who does it leave in the market to ship Rogue silicon for mobile/tablet ? (Other than Apple, who don't sell chips) ?

Cheers,
Rog.

 

[Nebuchadnezzar]

Reneseas announced a big.LITTLE / Rogue chip. Possibly Samsung?

 

[Wellington]

Reneseas announced a big.LITTLE / Rogue chip. Possibly Samsung?

Hmm. Renesas always seem to be in some sort of restructure or turmoil. Woudn't be suprised if they follwed the same path as STE.
Samsung maybe, but I thought they only have licenced 5XT (or atleast that's all that has been disclosed) ?

 

[Nebuchadnezzar]

Samsung maybe, but I thought they only have licenced 5XT (or atleast that's all that has been disclosed) ?

During CES they've stated they would have some "big" things in line in terms of graphics when asked about performance versus Apple, but only towards the end of 2013 / beginning 2014.

 

[Wellington]

In which case I would be surprised if it was IMG at all, they are not going to be able to differentiate against Apple with the same GPU.