Mediatek 8xA7 @ 2Ghz ?

I forgot to add that the only thing that worries me is the implementation; heck even Samsung didn't get it right with its first attempt. Let's see what Mediatek will make out of it.

 

I don't know if Mediatek can even afford to do their own custom implementation at all, they could very well be using a TSMC hard macro configuration provided by ARM and their partners.

 

There's nothing special to it, it's a basic Linux kernel feature. You don't need any support for it.

 

To me, this seems aimed at benchmarks more than real-world applications.

That said, A7s are tiny, so it might not cost much more than a "regular" quad-A7 setup.

 

a little OT in this thread, but after your comment about the tiny A7 [ in fact ~0,5mm² @ 28nm ]:

I would like to see a Federation-implementation of an 8 core A7 system.
Federation means that you combine two in-order cores to form one somewhat OOOe Core.
With OOOe the federated "quad A7" CPUs could be quite fast at 2GHz.

Federation means this here:
http://www.cs.virginia.edu/~skadron/Papers/boyer_federation_taco.pdf

a few more papers are available.

 

Of course does it depend on the benchmark, but given you get better results in benchmarks with some value, how many chances do you have it's not going to also show in real-word applications exactly?

If the implementation is more or less as some of us expect you have at 2GHz peak vs. the first 6589 iterations with quad A7@1.0GHz quite a healthy performance boost. The only other question mark would then be if and by how much power consumption has increased with two quad A7s at 2.0GHz. Since more bets in that market are actually about perf/mW and not just sterile performance, they might have another winner.

As a reminder these SoCs are dirt cheap; folks should keep that in mind.

 

To be honest I'm having trouble thinking of a cell phone application that might use more than 4 threads, except perhaps for games.

 

I don't even think there are any games with more than 4 threads simultaneously and I don't think there's any benchmark either, but would like to stand corrected. The point of the 8 cores in this case should be rather perf/mW than just performance by itself.

You have in a MT6589T today 4 A7 cores ranging from 400+ to 1.5GHz. You cannot clock them independently afaik, meaning that despite A7 cores having relatively low power consumption overall you cannot use any within that quad as a power saving core; now having two quad core clusters you could theoretically clock any amount of cores of the two quads up to 2GHz for complicated tasks and any amount of cores of the second quad at very low frequencies.

If implemented as I imagine it you could get in the 6592 higher performance than any of its 6589 predecessors while definitely not burning more power.

 

Frankly I'm very sceptical about the two clusters for perf/power reasons. I suspect the disadvantage of running tasks on a too-high clocked clusters is pretty minimal in real world. If you have some complicated task requiring full clock, the power consumption is probably mostly dictated by that anyway, so the other tasks running on high-core clocked won't contribute that much to overall power consumption (after all the core those smaller tasks are running on will idle a lot of the time, so it is certainly not drawing max power). There's other reasons why I don't think it's a particularly good idea (more complex scheduling) and unlike A7/A15 pairs I think there's other reasons why it's 8 cores (though yes I can't see any other than for some benchmark purposes neither).

 

Does this new dynamic feature requires a special Linux kernel or is it there already?

 

WTF is an eight core MP4

I believe even a 544MP4 when I see it, it sounds more like typical chinese marketing bullshit where they're counting Lord knows what. An at least 4x times increase in graphics performance between the 6589 and the 6592 sounds completely unrealistic to me. Sign me up for a MP2@~250MHz if they're really going multi-core.

***edit:

MT6589 = SGX544MP1@250+MHz
MT6589T = SGX544MP1@350+MHz

a viable scenario based on that:

MT6592 (1.7GHz CPU) = 544MP2@250+MHz
MT6592T (2.0GHz CPU) = 544MP2@350+MHz

 

Sony is reportedly bringing models with the MT6589 later this year.
I could see a 2x2 big.LITTLE SoC with a decent performing Rogue being quite competitive with the Snapdragon 600 (perhaps giving S400 a run for its money?), so Mediatek might actually become more widespread.

To think that some 4 years ago all they had to compete with the OMAP3 and Snapdragon was some piss-poor SoCs with an ARM9 @ 400MHz.. they're really stepping up in the industry.

 

A 2 fold increase at a time makes far more sense unless Mediatek all of the sudden changes pace and wants to go for higher than budget offerings.

Sure but you have to admit that my scenarios aren't exactly absurd either and yes purely speculative always.

By the time you see 8xxx in devices that 2*2 big.LITTLE config with its frequency will be mainstream and below. Here's my line of thought:

6589T = CPU 4*A7@1500, GPU= 11+ GFLOPs
6592T = CPU 2*(4*A7)@2000, GPU = 22+ GFLOPs
8135T = CPU 2*A15@1600 + 2*A7, GPU = ~45 GFLOPs

 

I am more interested in its networking side of things. They claim to support FDD/TDD LTE, WCDMA , TD-SCDMA, GSM etc.

Does anyone know which IP they will be using. As far as i know, this is the only LTE SoC not coming from QualComm, Samsung and Huawei. With the later two being an Mobile Network Equipment Provider therefore owning some of the 4G IP.

 

MT8135
Big.little - 2*A15,2*A7 paired with IMG G6200.
Integrated Wi-Fi, Bluetooth 4.0, GPS, FM.

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

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