Mediatek 8xA7 @ 2Ghz ?

[Exophase]

I don't see why all the hate for these 8 cores, there's plenty of usages for them.

I wouldn't say that there's zero use for them, but it's questionable if it warrants the overhead of having two separate CPU clusters, including separate L2 cache, coherency link, etc. Maybe if the asynchronous domains get put to good use but I'm skeptical.

I think MediaTek does deserve criticism for basically exploiting media coverage of terrible benchmarks like AnTuTu. I'm skeptical that MediaTek would even be making this SoC if these benches didn't exist (ones that take a bunch of awful benches then make it worse by running several instances of them in parallel). Promoting these benchmark numbers makes you part of the problem.

Also, shame on Engadget for report that this is the first true octa-core, unless Exynos 5420 really didn't fix coherency :| They're probably just lifting the more accurate claim of MediaTek's, that it's the first octa-core Cortex-A7.

 

[Nebuchadnezzar]

I wouldn't say that there's zero use for them, but it's questionable if it warrants the overhead of having two separate CPU clusters, including separate L2 cache, coherency link, etc. Maybe if the asynchronous domains get put to good use but I'm skeptical.

They're still incredibly small and cheap. Doubling a quad A7 is probably more beneficial than going 4xA7 + 1xA15 (The alternative for same die size). It helps in all things loading (App loading, browser loading), even if it's not that useful for continuous loads (Gaming).

 

[Exophase]

They're still incredibly small and cheap. Doubling a quad A7 is probably more beneficial than going 4xA7 + 1xA15 (The alternative for same die size). It helps in all things loading (App loading, browser loading), even if it's not that useful for continuous loads (Gaming).

I could not disagree with you more on your claim of 8xA7 being better than 4xA7 + 1xA15. I'm highly skeptical that loading apps scales so well on cores, except maybe if it's compiling code with ART or something.

Even MediaTek's own usage charts for web-site core utilization don't show a case where 8xA7 would be faster, and they were undoubtedly looking for a shining example here. Look at the utilization graphs, I would not qualify the utilization as meaningfully > 6 cores at any point, which 4xA7 + 1xA15 could generally cover.

And at any rate, why would MediaTek even be offering a 2xA15 + 2xA7 SoC if they didn't find that configuration superior for at least something?

 

[Deleted member 13524]

Also, shame on Engadget for report that this is the first true octa-core, unless Exynos 5420 really didn't fix coherency :| They're probably just lifting the more accurate claim of MediaTek's, that it's the first octa-core Cortex-A7.

AFAIK, Exynos 5420 is not capable of activating all 8 cores at the same time.

 

[Nebuchadnezzar]

AFAIK, Exynos 5420 is not capable of activating all 8 cores at the same time.

Of course it is, I think you're confusing it with Samsung Mobile probably never updating the Note 10.1 2014 and Note 3 kernels to actually do so. In that sense Mediatek is correct in saying they have the first working 8-core.

 

[Deleted member 13524]

Of course it is, I think you're confusing it with Samsung Mobile probably never updating the Note 10.1 2014 and Note 3 kernels to actually do so. In that sense Mediatek is correct in saying they have the first working 8-core.

So you're saying the Exynos 5420 is capable of using all the cores at the same time, but no one has seen it doing so because there's a software update somewhere that no one has that would allow it.

Sorry, but that's practically the same as not being capable of doing it.

 

[Exophase]

Does anyone know if MT6592 is actually capable of running the two clusters at different voltages or at least clock speed? I assumed it would be, but unlike real big.LITTLE there's no reason why this has to be the case.

So you're saying the Exynos 5420 is capable of using all the cores at the same time, but no one has seen it doing so because there's a software update somewhere that no one has that would allow it.

Sorry, but that's practically the same as not being capable of doing it.

I don't think it is. One of them can (and probably will) be improved with software updates, the other can't be. MediaTek is describing their hardware, not software.

 

[Nebuchadnezzar]

but no one has seen it doing so because there's a software update somewhere that no one has that would allow it.

Where did you pull that nonsense out of?

There have been plenty of demonstrations of the devboards showcasing HMP and ARM using the chip to promote big.LITTLE over the last several months. I confirmed this back several months ago before the 5420 even went public.

What Samsung's Mobile division and their incompetence in software does with the actual products they ship the chip in is their own problem unrelated to the hardware.

Does anyone know if MT6592 is actually capable of running the two clusters at different voltages or at least clock speed? I assumed it would be, but unlike real big.LITTLE there's no reason why this has to be the case.

They'd be stupid if they didn't. They already confirmed they're powering down clusters for power-management, and that logically would imply shutting down the cluster regulator too.

 

[Exophase]

They'd be stupid if they didn't. They already confirmed they're powering down clusters for power-management, and that logically would imply shutting down the cluster regulator too.

The ability to power down clusters separately doesn't imply anything about having separate voltage and clock rails. You can power down individual cores in a cluster too, but they certainly aren't in separate DVFS domains. It doesn't have to mean shutting down a regulator, it could just mean power gating.

Having separate clocks means needing another PLL (not necessarily a huge deal), having separate voltage rails means another regulator somewhere either internal to the SoC or on the PMIC (much bigger deal). I wouldn't take it for granted that this is the case for a cheaper SoC. It doesn't help that MediaTek hasn't advertized heterogeneous multiprocessing.

 

[Deleted member 13524]

Where did you pull that nonsense out of?

From your own post, which I quoted.

There have been plenty of demonstrations of the devboards showcasing HMP and ARM using the chip to promote big.LITTLE over the last several months. I confirmed this back several months ago before the 5420 even went public.

Devboards can enable features that could not ever be enabled in consumer products. First, the available power consumption is in a whole different league and then it could have additional hardware that didn't fit into the consumer products.

 

[Nebuchadnezzar]

Devboards can enable features that could not ever be enabled in consumer products. First, the available power consumption is in a whole different league and then it could have additional hardware that didn't fit into the consumer products.

Enable features? Additional hardware? Are we even having a technical discussion here or just throwing vague words around?

Devboards are running the exact or similar platforms mirrored in the consumer counterparts. In this case the public dev boards are actually lagging behind the consumer platforms and are based on each other. There is no such thing as additional "features" or "hardware", the SoC is the same and comes in the same package. And if by available power consumption you mean available thermal headroom, sure, but that doesn't change anything in terms of the features or feature capability of the SoC.

The ability to power down clusters separately doesn't imply anything about having separate voltage and clock rails. You can power down individual cores in a cluster too, but they certainly aren't in separate DVFS domains. It doesn't have to mean shutting down a regulator, it could just mean power gating.

Having separate clocks means needing another PLL (not necessarily a huge deal), having separate voltage rails means another regulator somewhere either internal to the SoC or on the PMIC (much bigger deal). I wouldn't take it for granted that this is the case for a cheaper SoC.

A cluster's common power domain is handled on the SoC level which is very often just regulator driven. And an additional buck converter on a PMIC is less of a deal than you think it is, many suppliers such a Maxim offer more than reasonable solutions for it.

It doesn't help that MediaTek hasn't advertized heterogeneous multiprocessing.

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

The MT8135 is the first implementation of ARM’s big.LITTLE architecture to offer simultaneous heterogeneous multi-processing. As such, MediaTek is taking the lead to improve battery life in next-generation tablet and mobile device designs by providing more flexibility to match tasks with the right-size core for better computational, graphical and multimedia performance,”

 

[Exophase]

A cluster's common power domain is handled on the SoC level which is very often just regulator driven. And an additional buck converter on a PMIC is less of a deal than you think it is, many suppliers such a Maxim offer more than reasonable solutions for it.

Very often just regulator driven.. you're basing this on all of what, two examples from Samsung? There's nothing stopping MediaTek from using power gating per-cluster in their SoC. And we're not just talking an additional buck converter but one with comparable peak load as one of the most demanding, they're not all created equally..

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

They haven't advertised it as a capability of THIS CHIP. Separate power domains would enable a limited form of HMP, unless your idea of heterogeneous requires different uarchs. That, in my opinion, is more substantial than merely having 8 A7 cores.

 

[Lazy8s]

Samsung Mobile said they weren't updating their 5420 based devices for simultaneous eight CPU core operation because it wouldn't be a good balance of power/thermals under current implementations.

 

[Deleted member 13524]

Enable features? Additional hardware? Are we even having a technical discussion here or just throwing vague words around?

Yes, additional hardware compared to what's present a cellphone's PCB fed by a single 3.3/3.7V power source. For example, a dev board can have additional/more precise/more expensive/more power-hungry voltage regulators that could handle having all cores enabled in Exynos 5420 without causing spikes in temperature or power consumption. Or even just a heatsink.

Apparently, according to Lazy8s I was right and consumer devices with Exynos 5420 will not get any software update to enable all 8 cores at once, for the most predictable reasons (not incompetence from the software team).
Again: it's practically the same as not being able to do it.

And yes, I was trying to keep an intelligible discussion even as an electrical engineer, but your apparent hostility urges me to stop here, so I'll just leave it at that.

 

[Nebuchadnezzar]

They haven't advertised it as a capability of THIS CHIP. Separate power domains would enable a limited form of HMP, unless your idea of heterogeneous requires different uarchs. That, in my opinion, is more substantial than merely having 8 A7 cores.

Heterogeneous MP implies that the CPUs are different which as you say they are not, so I don't even know why you brought that up for this chip. You're trying to describe aSMP such as Qualcomm is doing. The two are substantially different.

Yes, additional hardware compared to what's present a cellphone's PCB fed by a single 3.3/3.7V power source. For example, a dev board can have additional/more precise/more expensive/more power-hungry voltage regulators that could handle having all cores enabled in Exynos 5420 without causing spikes in temperature or power consumption. Or even just a heatsink.

Can may maybe example theory.

That's fine and dandy but that's just not how it is. The reality is that the phone/tablet platforms have even higher performing PMICs than these dev boards because the boards are targeting a low price point. Samsung uses their own in-house PMIC in the phones with twice the voltage resolution and claimed low dropout versus a cheaper alternative from Maxim on the dev boards.

What Samsung PR is claiming about heat is as worthless as them still claiming that there's absolutely nothing wrong with their first b.L implementation even though everybody knows they're lying through their teeth.

Backporting GTS on a current Android kernel, or upstreaming various drivers from third-party vendors to a non-LTS version which does support GTS, and going through validation for all of that, is however a very valid real reason and problem.

 

[Arun]

I agree with Nebu about the reasons why Samsung isn't exposing simultaneous 8-core, but I disagree about the likelihood of Mediatek having different voltage rails on the MT6592 for the two clusters; or rather on the likelihood of shipping implementations exposing them if they are there.

The octa-core marketing gimmick and HEVC makes it look like a high-end solution, but in many ways this is very clearly still a low-end platform where every penny matters. The Mali-450MP4 is a good indication of that. It's very strong for its intended market but shouldn't mistaken for something it's not IMO...

 

[Exophase]

Heterogeneous MP implies that the CPUs are different which as you say they are not, so I don't even know why you brought that up for this chip. You're trying to describe aSMP such as Qualcomm is doing. The two are substantially different. Can may maybe example theory.

Heterogeneous can and has been used in the past to refer to clock speeds, but you're right, I should have said asynchronous instead because those are the current marketing terms.

Now with that bit of semantics out of the way - MediaTek has not advertized any asynchronous multiprocessing capability of this chip. Given that Qualcomm isn't especially shy about this capability on their SoC it seems like a big omission on MediaTek's part, unless they think it'll just confuse the ignorant userbase they're deliberately marketing to.

 

[hatter]

Some MT6592 benchmarks

http://timesofindia.indiatimes.com/...ower-your-next-phone/articleshow/26833119.cms

Single core performance looks poor, graphics average, general performance good.

All said, this could have been a nice chip with a better graphics processor.

 

[loekf]

Are eight cores better?
The answer is a clear no. Number of cores doesn't matter. The actual technology used inside the cores is more important. This is the reason why Snapdragon 800 is significantly faster than MT6589 even though it has four less cores. Similarly the chip in Note 3 is also faster than MT6592. And theoretically, A7, the processor that powers iPhone 5S is faster than every other mobile processor even though it has just two cores.

Wow... reviewers who actually dont believe the multi core hype. Can you also say that Android is not really optimized for multiple cores ?

 

[Ailuros]

http://semiaccurate.com/2014/04/01/mediatek-goes-first-last-mt6595/

Not nice at all if true.