Mediatek planning 10 and 12 core chips for late 2015!

First of all it’s the apps that need to be optimized in order to use all those cores. It could be awesome if they kept their BIG.Little configuration but added 4/2 “even smaller” cores that hardly used any power so that technologies like Active Display form Motorola or always-on voice commands listening could be easily implemented.

Mediatek is a powerful CPU but their problem lies in the GPU. Both Qualcomm & Nvidia have better GPU i think.

The main point of this one is probably just for destroying benchmark tests.

10 and 12 cores? Will my next server run on Mediatek SoCs?

I really don’t mind this “core wars”, but if they manage to do a 12-cores SoC composed by a very (VERY) low-energy quadcore for basic operations and a powerful octa only for an intense use, it might have some sense… octacores right now are only a waste of battery power, better keeping them on rest and use them only if they are really needed!

How ever you look at this it’s plane stupid and absurd. First of all today’s software uses first two cores in 95% cases. Future software could use 4 up to 8 cores. Today more than 8 cores is absurd on anything that is not the network server with lot request’s to pull in the future you will need even less general purpose cores as most of the computing will be done on integrated FPGA as programmable gate area’s are around 4x more efficient than GPUs & well suited for variety of tasks not just hi parallel. So you don’t need 12 cores or you will ever need them. Thing that MTK needs is integrating more GPU clusters & this is not hard to do but somehow its hard for them to figure out. In the next gen they could switch to newer interconnect, adopt (license) MALI 8 gen & go with DPL DDR 4. Iddon’t see them (MTK) as someone ever relatively creative so I don’t expect them to go with like hyper memory cube or external connect lines IBM used on it’s last gen even bought are free open standard. What they could do (& they dont) is per cluster/core phase input optimization (even NV got this sane with X1 & kraits used it from the start) & FPGA with DSP capabilities integration in start for LTE & then developing it for usage with everything

else. Main problem with FPGAs is that they are hard to program (but that is changing thanks to Chronos) & lack of support in Linux (& Windows) for micro kernel’s. Microsoft got this & is working on implementation, Linux can import it from BSD when Linus is done with it’s stubbornness.

What a joke, even 8 cores is too much. I’d rather have a super powerful and power efficient quad core

12 cores and 1gb ram

How ever you look at this it’s plane stupid and absurd. First of all today’s software uses first two cores in 95% cases. Future software could use 4 up to 8 cores. Today more than 8 cores is absurd on anything that is not the network server with lot request’s to pull in the future you will need even less general purpose cores as most of the computing will be done on integrated FPGA as programmable gate area’s are around 4x more efficient than GPUs & well suited for variety of tasks not just hi parallel. So you don’t need 12 cores or you will ever need them. Thing that MTK needs is integrating more GPU clusters & this is not hard to do but somehow its hard for them to figure out. In the next gen they could switch to newer interconnect, adopt (license) MALI 8 gen & go with DPL DDR 4. Iddon’t see them (MTK) as someone ever relatively creative so I don’t expect them to go with like hyper memory cube or external connect lines IBM used on it’s last gen even bought are free open standard. What they could do (& they dont) is per cluster/core phase input optimization (even NV got this sane with X1 & kraits used it from the start) & FPGA with DSP capabilities integration in start for LTE & then developing it for usage with everything

else. Main problem with FPGAs is that they are hard to program (but that is changing thanks to Chronos) & lack of support in Linux (& Windows) for micro kernel’s. Microsoft got this & is working on implementation, Linux can import it from BSD when Linus is done with it’s stubbornness.

Andi you forgot to add Qualcomm are having serious issues with their 810 chipsets.

First of all. It’s hard to make processor cores any faster than they are today without a major breakthrough as in quantum processors (which I don’t see coming the next couple of years). Also, today, if elevating the clock of the processors, the temperature and power drain goes through the roof. No nice way to go. Then you have to convert to Mototolas custom low-power-always-on chips which is nice but requires a whole new way of programming them (they work more like graphics processor shader language than ordinary processor micro ode. As *A LOT* of programmers out there are not capable of correctly programming and using multi threading correctly without introducing bugs and race conditions, there are no easy way to increase the power of the processors without a whole bunch of programmers learning their stuff and beginning to crank out better software written for the new processors.

Now…..

Writing software for massively parallel processors and graphics processors are so way above most developers that I cringe and blush with embarrassment. I thus see the only way forward as more processor cores and kicking the developers in the nuts and tell them to learn their stuff.

So… Are the developers going to learn to write parallel/threading software all by themselves or do they need the enticement of faster code? Well… Programmers are a massively lazy bunch. If they can lean back (as we have done for years on end, just waiting for the processor speeds to make up for the slow, ugly code), we sure will…. It’s really a story about the hen and the egg. If no multi core processors (and by that I mean *multi*, not merely one or two more) appear, then no improvement will come. So, while it may be true that your flappy birds or calendar app does not need to have more processor cores, these 8 or 12 core processors will pressure developers into learning to develop for them or to make development tools that automatically fix the threading and racing issues (thereby enabling half assed, lame, mediocre developers to actually not be phased out). So if the processor manufacturers stay with the quad core T-Ford kind of old ass processors, nothing will change. Instead we might actually see well written software that uses more cores and achieve higher performance.

It still a lot of people that thinks that we can crank out more power per core without going to a completely new processor technology (not merely generation, this requires a radical new technology shift). What to do when a single office worker can’t deliver more. Well, then. We hire more workers… What to do when we discover that the old boss cannot manage the workers properly. Well, we hire a new one or get passed by the next company that actually manages to orchestrate the workers (processor cores) properly.

Bottom line: We need the new processors with more cores even if a lot of people just play flappy bird…

First of all. It’s hard to make processor cores any faster than they are today without a major breakthrough as in quantum processors (which I don’t see coming the next couple of years). Also, today, if elevating the clock of the processors, the temperature and power drain goes through the roof. No nice way to go. Then you have to convert to Mototolas custom low-power-always-on chips which is nice but requires a whole new way of programming them (they work more like graphics processor shader language than ordinary processor micro ode. As *A LOT* of programmers out there are not capable of correctly programming and using multi threading correctly without introducing bugs and race conditions, there are no easy way to increase the power of the processors without a whole bunch of programmers learning their stuff and beginning to crank out better software written for the new processors.

Now…..

Writing software for massively parallel processors and graphics processors are so way above most developers that I cringe and blush with embarrassment. I thus see the only way forward as more processor cores and kicking the developers in the nuts and tell them to learn their stuff.

So… Are the developers going to learn to write parallel/threading software all by themselves or do they need the enticement of faster code? Well… Programmers are a massively lazy bunch. If they can lean back (as we have done for years on end, just waiting for the processor speeds to make up for the slow, ugly code), we sure will…. It’s really a story about the hen and the egg. If no multi core processors (and by that I mean *multi*, not merely one or two more) appear, then no improvement will come. So, while it may be true that your flappy birds or calendar app does not need to have more processor cores, these 8 or 12 core processors will pressure developers into learning to develop for them or to make development tools that automatically fix the threading and racing issues (thereby enabling half assed, lame, mediocre developers to actually not be phased out). So if the processor manufacturers stay with the quad core T-Ford kind of old ass processors, nothing will change. Instead we might actually see well written software that uses more cores and achieve higher performance.

It still a lot of people that thinks that we can crank out more power per core without going to a completely new processor technology (not merely generation, this requires a radical new technology shift). What to do when a single office worker can’t deliver more. Well, then. We hire more workers… What to do when we discover that the old boss cannot manage the workers properly. Well, we hire a new one or get passed by the next company that actually manages to orchestrate the workers (processor cores) properly.

Bottom line: We need the new processors with more cores even if a lot of people just play flappy bird…