Mediatek 8xA7 @ 2Ghz ?
- Thread starter tangey
- Start date
Imagination put out PR recognizing MediaTek and the MT8135 for its sophistication in heterogeneous multi-processing.
The article is extremely vague on details. It might simply be referring to the expanded GPGPU assistance that Rogue can give within a SoC and the software/drivers MediaTek is including to facilitate that and may not necessarily be referring to any kind of shared memory addressing model between the CPU and GPU. Maybe that'll be a development down the line for expanded HMP in future SoCs.
I found the mention of the MT8135 being the first "publicly announced mobile SoC" to enable HMP utilizing asymmetric processor configurations to be an amusing statement (for a number of reasons), leaving room for earlier announced Rogue SoCs like the smartTV focused LG H13 and also the as-yet-unannounced Apple A7 (which is closer to release than even the MT8135).
http://imgtec.com/News/Release/index.asp?NewsID=797
The article is extremely vague on details. It might simply be referring to the expanded GPGPU assistance that Rogue can give within a SoC and the software/drivers MediaTek is including to facilitate that and may not necessarily be referring to any kind of shared memory addressing model between the CPU and GPU. Maybe that'll be a development down the line for expanded HMP in future SoCs.
I found the mention of the MT8135 being the first "publicly announced mobile SoC" to enable HMP utilizing asymmetric processor configurations to be an amusing statement (for a number of reasons), leaving room for earlier announced Rogue SoCs like the smartTV focused LG H13 and also the as-yet-unannounced Apple A7 (which is closer to release than even the MT8135).
http://imgtec.com/News/Release/index.asp?NewsID=797
Mediatek just signed a major extended agreement with IMG.
Mediatek already has a rogue license. This announcement talks about multiple cores from PowerVR series6 "Rogue" family and beyond.
Given all Mediatek has use to date is a (rumoured low clocked) G6200, then they have numerous performance steps within the existing announced rogue cores.
Its the "and beyond" that looks especially interesting to me, suggests strongly that this is the first public license for RogueXT.
http://www.imgtec.com/News/Release/index.asp?NewsID=801
Mediatek already has a rogue license. This announcement talks about multiple cores from PowerVR series6 "Rogue" family and beyond.
Given all Mediatek has use to date is a (rumoured low clocked) G6200, then they have numerous performance steps within the existing announced rogue cores.
Its the "and beyond" that looks especially interesting to me, suggests strongly that this is the first public license for RogueXT.
http://www.imgtec.com/News/Release/index.asp?NewsID=801
Well, MTK have announced their big.LITTLE chip with 2 x Cortex-A15 + 2 x Cortex-A7 + G6200 which ought to provide reasonable enough performance.
I can only assume that we'll see faster Rogue cores in their future, more powerful big.LITTLE chips, which I'd assume would probably be 4xA15 + 4xA7 or possibly even 4xA12 + 4xA7?
Since the MT8135 isn't shipping yet, I'd guess we're looking at the middle of 2014 until any faster Rogue GPUs come out of MTK.
I can only assume that we'll see faster Rogue cores in their future, more powerful big.LITTLE chips, which I'd assume would probably be 4xA15 + 4xA7 or possibly even 4xA12 + 4xA7?
Since the MT8135 isn't shipping yet, I'd guess we're looking at the middle of 2014 until any faster Rogue GPUs come out of MTK.
First benchmarks up of the MT8136 (mediatek with rogue)
http://gfxbench.com/device.jsp?benc...+MT8135+(PowerVR+Rogue+Han)&testgroup=overall
"rogue Han"
It is about 17% quicker than the apple A6 in GL2.5, but in comparison a startling x2.1 quicker in GL2.7. In fact it gets 3/4 of the ipad4's A6X score in that test, which is running a 70+Gflop GPU in an soc from a 1st tier designer. I don't think for a second the mediatek chip is running anywhere near this level in terms of Gflops.
Seems that Rogue is a LOT more efficient in handling shader heavy stuff than SGX.
Given that the iphone5s got x1.9 in GL2.5 of the iphone5 score, it will be very interesting to see the iphone5s GL 2.7 scores. Could be VERY surprising.
http://gfxbench.com/device.jsp?benc...+MT8135+(PowerVR+Rogue+Han)&testgroup=overall
"rogue Han"
It is about 17% quicker than the apple A6 in GL2.5, but in comparison a startling x2.1 quicker in GL2.7. In fact it gets 3/4 of the ipad4's A6X score in that test, which is running a 70+Gflop GPU in an soc from a 1st tier designer. I don't think for a second the mediatek chip is running anywhere near this level in terms of Gflops.
Seems that Rogue is a LOT more efficient in handling shader heavy stuff than SGX.
Given that the iphone5s got x1.9 in GL2.5 of the iphone5 score, it will be very interesting to see the iphone5s GL 2.7 scores. Could be VERY surprising.
Last edited by a moderator:
D
Deleted member 13524
Guest
The MT8136 is looking really good. 3D performance seems to be close to the underclocked Snapdragon 600 in Nexus 7, which is really good news for cheaper tablet makers.
Now I'd like to see how this fares against the Rockchip RK32xx coming in Q2'14 with quad-core Cortex A12 and a Mali T624.
Rockchips seem to be more popular with tablets, which is what this MT8136 is targeting.
I thought things would slow down with the loss of Texas Instruments and ST-Ericsson, but it turns out the pace is accelerating.
Now I'd like to see how this fares against the Rockchip RK32xx coming in Q2'14 with quad-core Cortex A12 and a Mali T624.
Rockchips seem to be more popular with tablets, which is what this MT8136 is targeting.
I thought things would slow down with the loss of Texas Instruments and ST-Ericsson, but it turns out the pace is accelerating.
MediaTek already demonstrated the MT8135 scoring quite a bit higher -- 46 fps versus 35 -- in the Egypt HD benchmark.
http://youtu.be/Kf_ZGDoEPMo
Unless this latest run of the test only picked up on the speed of the LITTLE CPU cores, it's potentially underclocked here as the reported top frequency is 1.2 GHz.
http://youtu.be/Kf_ZGDoEPMo
Unless this latest run of the test only picked up on the speed of the LITTLE CPU cores, it's potentially underclocked here as the reported top frequency is 1.2 GHz.
Yes, I did spot that. Also wondered was GLbench being confused by the big.little clocks when reporting the cpu clock.
D
Deleted member 13524
Guest
This just in from the MT6592:
This 8x Cortex A7 SoC from Mediatek has a Mali-450 GPU.
Everyone was wondering where in hell was ARM going to put that GPU, here it is.
As for 3D performance comparison, the Exynos 4412 with a Mali 400MP4 @ 533MHz in the Galaxy Note 2 does 4477 points, whereas this Mali 450MPx does 5181.
They also seem to believe that these are lower clocks than the ones coming in the final product.
So it's either a much lower clocked Mali 450MP8 (~300MHz?) or a similarly clocked Mali 450MP4 with the increased performance coming from that much needed second vertex shader.
Nonetheless, it seems to be a nice upgrade from the 3D performance of the SGX544MP1 of its MT6589 predecessor.
This 8x Cortex A7 SoC from Mediatek has a Mali-450 GPU.
Everyone was wondering where in hell was ARM going to put that GPU, here it is.
As for 3D performance comparison, the Exynos 4412 with a Mali 400MP4 @ 533MHz in the Galaxy Note 2 does 4477 points, whereas this Mali 450MPx does 5181.
They also seem to believe that these are lower clocks than the ones coming in the final product.
So it's either a much lower clocked Mali 450MP8 (~300MHz?) or a similarly clocked Mali 450MP4 with the increased performance coming from that much needed second vertex shader.
Nonetheless, it seems to be a nice upgrade from the 3D performance of the SGX544MP1 of its MT6589 predecessor.
It would be nice if Mediatek released a chip with a 4-core Cortex-A7 running around 2GHz along with a decent GPU. That would make a very good chip for the budget/mid-range market. Probably won't see this, however, as it will no doubt be easier to continue to pump out the 6589 chips.
Can't really see how useful an 8-core device will be much of the time except in a few corner cases.
The MT8135 is intended for tablets and the like, I believe? A pity we won't be seeing a dual A15 + dual A7 in phones any time soon as this might prove very capable in mid-range devices.
Can't really see how useful an 8-core device will be much of the time except in a few corner cases.
The MT8135 is intended for tablets and the like, I believe? A pity we won't be seeing a dual A15 + dual A7 in phones any time soon as this might prove very capable in mid-range devices.
D
Deleted member 13524
Guest
The additional 4xCortex A7 come at little area/power cost and much larger e-peen.. I mean Antutu scores and marketing prowess.
And FWIW, the reported max clocks for the production MT6592 are those same 2GHz you're asking for.
I think they're aiming for the Snapdragon 800's 30k Antutu scores, and then build all the marketing around that.
And FWIW, the reported max clocks for the production MT6592 are those same 2GHz you're asking for.
I think they're aiming for the Snapdragon 800's 30k Antutu scores, and then build all the marketing around that.
Its in use in an amlogic soc, the AML8726-M8
Rockchip also recently announced a Mali450 soc
I think hisilicon is coming out with one shortly too.
I think the fact that the 450 is getting more traction than T-series (even though 450 was announced long after T-series) suggests to me that T-series is struggling. Is Samsung the only one with T-series socs ?
According to the following, the AMlogic chip is clocking their 450MP8@250 Mhz.
http://www.tomshardware.com/news/Mali-450-MP8-AmLogic-ARM,22332.html
Yeah, I'd read that the 6592 purportedly clocks up to 2GHz. Doesn't alter the fact that I'd be just as happy with a quad-core chip running at the same speeds! Quad-core Cortex-A7 at 2GHz would have pretty good performance, I'd have thought.
The amount of times I have to roll my eyes when I read about people getting excited about Antutu scores is ridiculous. Anybody who thinks a quad-core Cortex-A7 is going to get near the performance of a Snapdragon 800 in real world use is smoking lots of illicit substances.
You're probably right about the marketing push of Antutu scores, however. Not a bad idea when people don't really understand the technology and just look for bigger numbers. Ah, take me back to the days of Netburst...
10GHz!
I agree with mariner's statement upstream:
This makes a lot of sense to me, and yet we already know it will never happen. Bleargh.mariner said:
Nebuchadnezzar
Legend
I don't see why all the hate for these 8 cores, there's plenty of usages for them.
I'm not saying 8 cores must be a bad idea, but it's at best a great gamble to make a chip with 8 slow cores, where single thread performance might suffer. IMHO even one fast core with 7 slow cores is probably better (though scheduling would be a headache).
Similar threads
- Sticky
