Nebuchadnezzar
Legend
With the tesselation talk going on in the other threads, I find this statement from ARM interesting (kudos to ltcommander.data for bringing it up in the Q&A):
ARM Midgard Architecture
- Thread starter arjan de lumens
- Start date
So they're doing it in software or how should I understand that?
Some more interesting comments from Jem Davies at Anandtech:
http://www.anandtech.com/show/8226/ask-the-experts-arm-fellow-jem-davies-answers-your-gpu-questions
In fact the whole Q & A with users and Jem Davies is a highly interesting read.
So T760 is feature_level DX11.1 after all, but doesn't have a ff tessellation unit; so they're probably not using a sw solution but the Midgaard programmable parts of the hw ie the ALUs?
http://www.anandtech.com/show/8226/ask-the-experts-arm-fellow-jem-davies-answers-your-gpu-questions
In fact the whole Q & A with users and Jem Davies is a highly interesting read.
So T760 is feature_level DX11.1 after all, but doesn't have a ff tessellation unit; so they're probably not using a sw solution but the Midgaard programmable parts of the hw ie the ALUs?
jedibeeftrix
Newcomer
they were my questions:
Q - Mali 7xx series - full DX11 compliance or not?
Q - Did Khronos drop the ball on progressing OpenGL ES and force Google to release the "extension pack"?
Last edited by a moderator:
Good job on those; the good thing is that he's doing an excellent job despite that users have bombarded him with questions
jedibeeftrix
Newcomer
yes, he's answered a lot of questions, many of which are overlapping and we'd be tempted to skip.
D
Deleted member 13524
Guest
Extensive article about the Midgard architecture at Anandtech:
http://www.anandtech.com/show/8234/arms-mali-midgard-architecture-explored
http://www.anandtech.com/show/8234/arms-mali-midgard-architecture-explored
I looove it when IHVs come out in the open - at least a bit - about their architectures.
So T7xx is using after all its ALUs for generation geometry for tessellation. I understand their point why they skipped the ff tessellation unit, but I don't really know how efficienct the architecture is with geometry either.
Both the T628MP6 as the T760MP4 in the GLB2.7 (rather simple) geometry tests seem to fair quite well but that's all I know.
Nebuchadnezzar
Legend
https://translate.googleusercontent...v.html&usg=ALkJrhiU79zY77Y5GJM06FERGlP9teEraQ
15fps Manhattan offscreen
22.9fps TRex offscreen
T760 scores from the 5433. Still don't know what core arrangement it is.
Looking at the MP4 scores from the Rockchip SoC you could deduce it to be an MP6 going by the Manhattan scores alone: https://gfxbench.com/device.jsp?benchmark=gfx30&D=Rockchip+rk3288&testgroup=overall. The fillrate also sort of points out to MP6.
The TRex scores are very low for some reason.
Edit: Korean sources report 17.2fps and 37.6fps. Very comparable to the Adreno 420.
http://i.imgur.com/icXZVlj.png
15fps Manhattan offscreen
22.9fps TRex offscreen
T760 scores from the 5433. Still don't know what core arrangement it is.
Looking at the MP4 scores from the Rockchip SoC you could deduce it to be an MP6 going by the Manhattan scores alone: https://gfxbench.com/device.jsp?benchmark=gfx30&D=Rockchip+rk3288&testgroup=overall. The fillrate also sort of points out to MP6.
The TRex scores are very low for some reason.
Edit: Korean sources report 17.2fps and 37.6fps. Very comparable to the Adreno 420.
http://i.imgur.com/icXZVlj.png
Last edited by a moderator:
I don't think Midgaard cores have as low die area as the Mali 4xx configs had. Other than that if that's truly a MP6 at 650 or even 700MHz yielding a bit over 17 fps in Manhattan, Apple gets roughly 10% more performance with a tad over 500MHz (estimated) from a 4 cluster config.
Unfortunately we have neither any die area or power indications for any of the 760 variants, but if there's nothing wrong with results like that: http://gfxbench.com/device.jsp?benc...i=gl&D=Samsung SM-G900x Galaxy S5 (Mali-T628)
it's predecessor throttled a helluva lot.
Shamelessly plugged from mboeller from another forum:
http://www.gizmochina.com/2014/10/07/hisilicon-kirin-920-tear-down/
http://www.gizmochina.com/2014/10/07/hisilicon-kirin-920-tear-down/
Nebuchadnezzar
Legend
That's the same breakdown I posted in June, you even commented on it...
Really? Sorry then; they posted it as "new" at gizmochina.
Nebuchadnezzar
Legend
I actually went through that SoC quite thoroughly if you happen to have missed it: http://www.anandtech.com/show/8425/huawei-honor-6-review/4
anandtech has an article online:
http://www.anandtech.com/show/8649/arm-announces-mali-800-series-gpus-t860-t830-t820
Unfortunately, nothing technical. Frame buffer compression everywhere. looks like a small step after 700 series
http://www.anandtech.com/show/8649/arm-announces-mali-800-series-gpus-t860-t830-t820
Unfortunately, nothing technical. Frame buffer compression everywhere. looks like a small step after 700 series