PowerVR Series 6 now official

[tangey]

Mediatek MT8135 (2xA15 + 2xa7) confirmed as being Rogue.

http://www.mediatek.com/_en/03_news/01-2_newsDetail.php?sn=1115

http://withimagination.imgtec.com/i...ervr-series6-gpus-to-a-mobile-device-near-you

G6200


Ties in rather nicely with this article in digitimes today
http://www.digitimes.com/news/a20130729PD201.html

"MediaTek aims to push sales of its next-generation ARM big.LITTLE-based tablet processor, the MT8135, to Google, Amazon and Sony in addition to branded notebook and tablet vendors"

don't know how accurate the engadget tests are, but they show 15% down in 2.5 egypt offscreen compared to the ipad4 SGX554MP4@280Mhz.
http://www.engadget.com/2013/07/29/mediatek-mt8135-biglittle-mp-powervr-series6-g6200/#continued

I'd really like to see some 2.7 benchmarks

 

[mboeller]

the first(?) Rogue-Benchmarks:

http://www.engadget.com/2013/07/29/mediatek-mt8135-biglittle-mp-powervr-series6-g6200/

...verd...nur zweiter.

 

[french toast]

Now that's what I call a perfect smartphone big-LITTLE configuration, I didnt manage to get clock speeds or process but im guessing that is the optimium setup for efficiency.

That rogue looks awesome if you consider its the low end one on immature drivers, who knows how fast a high clocked 6400 with frame buffer compression is going to be

 

[mboeller]

Now that's what I call a perfect smartphone big-LITTLE configuration, I didnt manage to get clock speeds or process but im guessing that is the optimium setup for efficiency.

That rogue looks awesome if you consider its the low end one on immature drivers, who knows how fast a high clocked 6400 with frame buffer compression is going to be

you can find the clockspeeds for the A7 / A15 in the video. 1.2GHz for the A7 and 1,7GHz for the A15 CPUs

the G6200 seems to have >40GFlops based on the comparison with the MT8125 (SGX544 @ 300MHz)

 

[french toast]

you can find the clockspeeds for the A7 / A15 in the video. 1.2GHz for the A7 and 1,7GHz for the A15 CPUs

the G6200 seems to have >40GFlops based on the comparison with the MT8125 (SGX544 @ 300MHz)

Thanks, I went back and watched the video after writing my post.

If the a15s are clocked so conservatively, how come performance is so much better than the exynos 5250?.
Better memory controller? A7s?

 

[Ailuros]

you can find the clockspeeds for the A7 / A15 in the video. 1.2GHz for the A7 and 1,7GHz for the A15 CPUs

the G6200 seems to have >40GFlops based on the comparison with the MT8125 (SGX544 @ 300MHz)

If the latter is true then a wee bit over 300MHz for the G6200. The Kishonti database seems to have a problem so I can't have a look at 544@300MHz results in GLB2.5.

 

[Laurent06]

If the a15s are clocked so conservatively, how come performance is so much better than the exynos 5250?.
Better memory controller? A7s?

Vellamo and AnTuTu are a mix of both CPU and GPU so one can't say anything using these benchmarks.

This leaves us with Geekbench which is about 10% higher. We'd need individual scores to draw a conclusion.

I guess this helps: http://browser.primatelabs.com/geekbench2/compare/2205730/2092237

If this really is mt8135, then it seems the memory controller and/or RAM are indeed better.

 

[tangey]

If the latter is true then a wee bit over 300MHz for the G6200. The Kishonti database seems to have a problem so I can't have a look at 544@300MHz results in GLB2.5.

Plamensitos piece (http://forum.beyond3d.com/showpost.php?p=1768467&postcount=10)

Suggested his guess for G6200 is 105Gflops @ 533Mhz. which would lead 40 Gflops @ 200Mhz.

Are you thinking the GPU is more than 40Gflops, or that those estimates in general are wrong ?

 

[Ailuros]

Plamensitos piece (http://forum.beyond3d.com/showpost.php?p=1768467&postcount=10)

Suggested his guess for G6200 is 105Gflops @ 533Mhz. which would lead 40 Gflops @ 200Mhz.

I don't know how anyone is counting but there are 16 lanes per cluster * 4 FLOPs * 0.32GHz = 40+ GFLOPs

Are you thinking the GPU is more than 40Gflops, or that those estimates in general are wrong ?

Kishonti database quirks are gone so I could re-check; unless I've made a mistake a OMAP4470 (SGX544MP1@384MHz = 12 GFLOPs) gets 12,4 fps offscreen. Now we have for a G6200 46,0 fps offscreen in GLB2.5 which sounds fine to me.

If you want to put that into perspective that G6200 is yielding similar performance to a HTC One Google Edition Adreno320 with about 45% more FLOPs.

iPhone5 SGX543MP3@325MHz = 31,2 GFLOPs for 29,80 fps in GLB2.5.

If all that makes sense you can already estimate the iPhone5S GLB2.5 score

 

[gribble]

Eetimes article suggests G6200 in the MT8135 is ~80gflops

http://www.eetimes.com/document.asp?page_number=2&doc_id=1319064&image_number=1

 

[tangey]

I don't know how anyone is counting but there are 16 lanes per cluster * 4 FLOPs * 0.32GHz = 40+ GFLOPs

Those figures, would imply ST were going with something like a G6600@>500Mhz to get their stated >210 Gflops with the A9600 (RIP), which seems an extreme rogue implementation to be the first one planned.

 

[french toast]

Vellamo and AnTuTu are a mix of both CPU and GPU so one can't say anything using these benchmarks.

This leaves us with Geekbench which is about 10% higher. We'd need individual scores to draw a conclusion.

I guess this helps: http://browser.primatelabs.com/geekbench2/compare/2205730/2092237

If this really is mt8135, then it seems the memory controller and/or RAM are indeed better.

Thanks.

Wow, the exynos a15s are higher performant... due to the 4th revision? (R4).
Got to be memory/latency related.

 

[Ailuros]

Those figures, would imply ST were going with something like a G6600@>500Mhz to get their stated >210 Gflops with the A9600 (RIP), which seems an extreme rogue implementation to be the first one planned.

A G66x0 has inevitably 12 TMUs; doing some backwards speculative math would have told you that the A9600 might have had 8 TMUs * 667MHz = 5,34 GTexels/s; with 12 TMUs you'd have a by 50% higher texel fillrate

***edit: by the way in terms of GPU performance (concentrated on 2.5 obviously) the MT8135 delivers 3.6x times the GPU performance of a MT6589T and 4.5x times of a MT6589 with only small differences in frequency.

 

[mboeller]

Thanks.

Wow, the exynos a15s are higher performant... due to the 4th revision? (R4).
Got to be memory/latency related.

Don't you mix up the A9 r4 with a AFAIK nonexistent A15 r4?

 

[ltcommander.data]

Don't you mix up the A9 r4 with a AFAIK nonexistent A15 r4?

http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.ddi0438i/index.html

They are apparently up to r4 on the A15 too.

 

[ltcommander.data]

Rogue/Series6 is supposed to range from DX10 to DX11.1 (ok now with the low end G6100 it's actually DX9L3 to DX11.1). Let's see if the soon to be announced Series6XT brings >DX10.0 to the table; would make sense since for any windows platform you'd also need a specific level of performance.

http://withimagination.imgtec.com/i...ervr-series6-gpus-to-a-mobile-device-near-you

Thanks to the PowerVR G6200 GPU inside the MT8135 application processor, MediaTek brings high-quality, low-power graphics to unprecedented levels by delivering up to four times more ALU horsepower compared to MT8125, its PowerVR Series5XT-based predecessor. PowerVR G6200 fully supports a wide range of graphics APIs including OpenGL ES 1.1, 2.0 and 3.0, OpenGL 3.x, 4.x and DirectX 10_1, along with compute programming interfaces such as OpenCL 1.x, Renderscript and Filterscript.

If the Imagination blog is to be believed, the G6200 or at least the G6200 as implemented in the MT8135 now supports OpenGL 4.x and DX10.1?

 

[Ailuros]

http://withimagination.imgtec.com/i...ervr-series6-gpus-to-a-mobile-device-near-you


If the Imagination blog is to be believed, the G6200 or at least the G6200 as implemented in the MT8135 now supports OpenGL 4.x and DX10.1?

G6100 is DX9L3 and all the other variants from G6200 up to 6630 should be DX10.0 (yes that's a 0) and OGL3.2. Later variants that haven't been announced yet will support DX11.x/OGL4.x.

http://www.imgtec.com/factsheets/powervr/PowerVR_Series6_IP_Core_Family_-1.0-.pdf

 

[french toast]

Don't you mix up the A9 r4 with a AFAIK nonexistent A15 r4?

No im not, it lists the revision number on the geekbench link laurent put up.

 

[ltcommander.data]

G6100 is DX9L3 and all the other variants from G6200 up to 6630 should be DX10.0 (yes that's a 0) and OGL3.2. Later variants that haven't been announced yet will support DX11.x/OGL4.x.

http://www.imgtec.com/factsheets/powervr/PowerVR_Series6_IP_Core_Family_-1.0-.pdf

http://www.imgtec.com/news/Release/index.asp?NewsID=666

All members of the Series6 family support all features of the latest graphics APIs including OpenGL ES ‘Halti’*, OpenGL 3.x/4.x, OpenCL 1.x and DirectX10 with certain family members extending their capabilities to full WHQL-compliant DirectX11.1 functionality.

Imagination did claim OpenGL 4.x support when the G6200 was first announced. Perhaps the hardware support is there, but current drivers only enable OpenGL 3.2? The DX10.1 claim seems to be new (or incorrect) though.

 

[Ailuros]

http://www.imgtec.com/news/Release/index.asp?NewsID=666


Imagination did claim OpenGL 4.x support when the G6200 was first announced. Perhaps the hardware support is there, but current drivers only enable OpenGL 3.2? The DX10.1 claim seems to be new though.

No it's not new; it's always worded somewhat confusing if you don't watch it. It's either DX10.0 and/or OGL3.2 or DX11.x and/or OGL4.x. It doesn't make sense for a GPU to be say OGL4.3 or even 4.4 and be just DX10.0 and vice versa. You have to watch the "up to" notes wherever present; in the contrary case think which OGL version "matches" which DX number.

If I'd rewrite that paragraph I'd say:

All members of the Series6 family support all features of the latest graphics APIs including OpenGL ES ‘Halti’*, OpenCL 1.x and DirectX10 and/or OpenGL3x with certain family members extending their capabilities to full WHQL-compliant DirectX11.1 and/or OpenGL4.x functionality.

Unless they've changed something last minute (what for exactly?) all so far variants are DX10.0. You can actually ask that PR gentleman in the blog link in the notes underneath if it's 10.0 or 10.1 after all.