PowerVR Series 6 now official
- Thread starter roninja
- Start date
Sgx5->sgx5xt was about an improvement in the core, but much more importantly for the road map, revealed a theoretical max 16 core implementation, (only x4 was implemented), and thus pushing the upper limit of the 5 series designs to x8 and beyond what sgx5 had allowed, on a clock per clock basic.
Sgx6--> sgx6xt as announced describes a 0.5x improvement.
Now on examination of the timeline, 5xt was announced, and it was a couple of months after that, that the MP came to be publicised. So perhaps there is a bigger reveal around the corner.
But it seems to me that with the announcements of the last couple of days, there are now more that a dozen 6/xe/6xt variants. Is it reasonable to expect to see multi-whatever variants of 6XT to show up in the short term ?
Last edited by a moderator:
Whereby all marketing told you back then is what the maximum design latency for 5XT was; it would had been nonsense to scale over MP16 as it wouldn't scale anymore as expected. With anything Series6/6XT in theory again even 32 clusters aren't necessarily an upper threshold from what I understand.
To be honest with as many different core variants I'm losing track; they're too much from my POV. Other than that I can't say I disagree with their marketing this time that they actually limit any PR/announcement more to what realistically will get licensed within a given timeframe. We know already that the design itself can scale in more than one way beyond 1 TFLOP so what?
At this stage I'd consider even an 8-cluster variant announcement overkill.
The number of core variants publicly announced is a reasonable barometer for market demands, and a function of how nicely modular the core is. The G6x00/30, GX6x40/50 and G6x50/60 variants obviously have some key differences to make the alternative variant show up in public, but they all come from the same codebase.
So a noisy roadmap (in your opinion, I think it shows nicely that we can do smaller cores if the customer wants to sacrifice some features, which isn't what you get elsewhere when shopping for graphics IP) is mostly serene and calm behind the scenes.
It quite obviously distills down to just 6, 6XT and 6XE, each with area-conscious variants, which scale by number of clusters. That's pretty easy to follow.
So a noisy roadmap (in your opinion, I think it shows nicely that we can do smaller cores if the customer wants to sacrifice some features, which isn't what you get elsewhere when shopping for graphics IP) is mostly serene and calm behind the scenes.
It quite obviously distills down to just 6, 6XT and 6XE, each with area-conscious variants, which scale by number of clusters. That's pretty easy to follow.
I don't know who wrote this article for daily tech,
http://www.dailytech.com/CES+2014+P...icial+Pack+up+to+384+Shaders/article34078.htm
But I'm pretty sure I could have done a better job, and I ain't no expert. Its full of rogue "multi-core" references etc. it even says that series5 wasn't a unified shader design.
So I'm not holding my breath when they put in the throw-away line of:
"It's expected that we'll see a Snapdragon chips with the PowerVR 6 series (the Snapdragon 410, 610, and 810) as well later in H1."
http://www.dailytech.com/CES+2014+P...icial+Pack+up+to+384+Shaders/article34078.htm
But I'm pretty sure I could have done a better job, and I ain't no expert. Its full of rogue "multi-core" references etc. it even says that series5 wasn't a unified shader design.
So I'm not holding my breath when they put in the throw-away line of:
"It's expected that we'll see a Snapdragon chips with the PowerVR 6 series (the Snapdragon 410, 610, and 810) as well later in H1."
Good effort, but there are some big errors in that article. The author refers to "Adreno" as Samsung's mobile GPU unit. He also says that the Tegra SoC line packs "CUDA cores", which is not true and will only come to fruition with TK1. And as you mentioned, he also mistakenly stated that 5 series cores do not have unified shaders.
I'm taking it for granted that that last sentence is utter bullshit. Other than that the author covered quite a large timespan.
With the number of comments around the web noting the absence of a multi-core announcement in the Series 6XT reveal and the corresponding With Imagination blog entry, no one seems to be taking note of the description of the new "Hierarchical Scheduling Technology" in the blog article which is said to include the logic for demand-driven workload and tile distribution to, when available, multiple cores.
So, while scaling clusters has been enough at this time for ImgTec to best cover the range of performance wanted by their potential partners, they obviously will have future-proofed the design with multi-core flexibility if the R&D savings from not having to develop so many cluster scalings outweighs the loss of efficiency in processing performance from the MP overhead like at the end of last generation.
So, while scaling clusters has been enough at this time for ImgTec to best cover the range of performance wanted by their potential partners, they obviously will have future-proofed the design with multi-core flexibility if the R&D savings from not having to develop so many cluster scalings outweighs the loss of efficiency in processing performance from the MP overhead like at the end of last generation.
Last edited by a moderator:
So K1 has set the cat amongst the GPU pigeons, although we have no idea about its perf/w, compared to Series 6. Now how will Apple respond, will Series 6XT be ready for inclusion in Apple's A8?, combined with a move to 20nm and thus high clocks it should rival the K1's performance.
Lead licencees (read Apple) have had it since late last year. How close that makes it to be productised, is another matter I guess.
SA starts to deal increasingly with ULP which is a good thing: http://semiaccurate.com/2014/02/07/allwinner-shows-8-core-a80-soc-mid-range
Wasn't that A80 also rumored for a Microsoft winRT (yuck) device?
Wasn't that A80 also rumored for a Microsoft winRT (yuck) device?
I note that some of the thread has been culled and moved elsewhere. fair enough if it is deemed to be off-topic. However, it would be nice if it was noted on the thread, and link made to the new location. Took me a while to figure out if I was imagining where a conversation was, and then another while to find out that I wasn't imagining it, and that it had ben moved to a different forum.
For the moved discussion, 3D and Semiconductor Industry Forum:
http://forum.beyond3d.com/showthread.php?t=64890
http://forum.beyond3d.com/showthread.php?t=64890
Not sure where to stick this, but from my personal interests, I'm going with the rogue thread.
MediaTek Announces MT6595, World’s First 4G LTE Octa-Core Smartphone SOC with ARM Cortex-A17 and Ultra HD H.265 Codec Support
4xa17 + 4xa7
Integrated LTE MODEM
powervr rogue (variation unknown)
I know mediatek has good video IP inhouse, but I wonder if the h265 hardware video encode/decode is provided by IMG.
Good design win for IMG, first confirmation of rogue in mediatek smartphone roadmap. Considering their recent trend of smartphone socs switching to Mali, this must considered a design loss for Mali t-series. (Mali continue to struggle to get a significant smartphone design win with T series)
Mediatek continues with the trend of odd numbered socs using powerVR.
http://www.mediatek.com/_en/03_news/01-2_newsDetail.php?sn=1137
MediaTek Announces MT6595, World’s First 4G LTE Octa-Core Smartphone SOC with ARM Cortex-A17 and Ultra HD H.265 Codec Support
4xa17 + 4xa7
Integrated LTE MODEM
powervr rogue (variation unknown)
I know mediatek has good video IP inhouse, but I wonder if the h265 hardware video encode/decode is provided by IMG.
Good design win for IMG, first confirmation of rogue in mediatek smartphone roadmap. Considering their recent trend of smartphone socs switching to Mali, this must considered a design loss for Mali t-series. (Mali continue to struggle to get a significant smartphone design win with T series)
Mediatek continues with the trend of odd numbered socs using powerVR.
http://www.mediatek.com/_en/03_news/01-2_newsDetail.php?sn=1137
It's a little peculiar that Mediatek claim this will be available in devices in the second half of the year when ARM's announcement about the Cortex-A17 said devices wouldn't be available until 2015!
I suppose if we see the 6595 arriving in December, this wouldn't be too far off schedule, but it would be nice to see it sooner than that.
I still have to wonder if a 2+2 big.LITTLE chip of this type would be better for most midrange phones, but I suppose the 4 Cortex-A7s might provide enough performance at lower power to make the use of a 2+2 design unnecessary.
Sorry for the relative OT about the A17 but, as we know from experience, we're unlikely to hear much about the Rogue part of this chipset until a while after it has been released!
Wonder what this means for the Cortex-A12? Stillborn?
I suppose if we see the 6595 arriving in December, this wouldn't be too far off schedule, but it would be nice to see it sooner than that.
I still have to wonder if a 2+2 big.LITTLE chip of this type would be better for most midrange phones, but I suppose the 4 Cortex-A7s might provide enough performance at lower power to make the use of a 2+2 design unnecessary.
Sorry for the relative OT about the A17 but, as we know from experience, we're unlikely to hear much about the Rogue part of this chipset until a while after it has been released!
Wonder what this means for the Cortex-A12? Stillborn?
