MediaTek Helio X20 spotted on Geekbench, decimates the best SoCs around

You really need to do a real one by one test from the source (optimized compiling it) & exclude any crypto present in geekbench3 so that Cisc & RISC could get any sense & even then it would be just a possible performance metric.

You see ARM is still a rather pore on the SIMD aria along with compilers & tools. Something that whose a rather long time built for X86 platform, SIMD extension variety along with specialized tool chains, compilers & match libs are fronts where ARM simple can not (still) compete with X86. It’s kind of sad that ARM is selling extra fee (& expensive one) for their optimized match libs as they won’t get far this way, on the other hand ARM GGC is far behind wile LLVM with its wide engagement & lost of back end funding (Aple, Samsung…) along with lots of branches (Clang, Zapcc…) show a lot’s of potential they will actually need 2~3 more years of development to take a lead.

looks like MTK beats Q’com in CPU department while loses over the memory.

what would be the final effect in real life?

i mean, opening a web page full of JavaScript for example, which one would be faster?

Or a page full of GIF or YouTube videos?

sorry man, I’m really not into that stuff so don’t get mad if i tell you that i’ve read only a quarter of your comment. i appreciated your time though.

i simply wanted to know what regular people like me should expect in real life without going through the technology behind it.

that helped me better.

thanks for the reply

In practice the software, the connectivity and even the storage has a huge impact on browsing so the end result depends less on the SoC.

X20 is not really a direct competitor for SD820, should be a bit cheaper and X30 would aim to compete with SD820. Remains to be seen how aggressive Qualcomm is with SD652/650 pricing and if Mediatek tries to undermine those and SD820 with it’s pricing.

The integer score was quite a bit better than expected here, from what MT8173 and Kirin 950 have shown us, expectations were around 2100 not over 2400. Here the X20 vs Kirin 950 http://browser.primatelabs.com/geekbench3/compare/4632468?baseline=4630896

It does follow the same logic, i am not complaining that there aren’t 4 big cores since 4 would have to be clocked a lot lower to fit in the TDP.. You do actually need more than 2 cores, you might not need 8 small ones with the 2 big ones ,perf wise, but there should be a significant benefit power wise. The A53 are also rather small so the cost is reasonable.Mediatek claimed this kind of power savings by using 3 clusters

Don’t spill your bullshits on me.

Read & get some education.

http://www.moorinsightsstrategy.com/research-paper-do-8-cores-really-matter-in-smartphones/

Their is zero browser engines that can use more than two cores. First one will be Mozilla Servo when it gets out next year & if it gets out.

S6520 won’t be a lot slower, it will be a bit slower in most cpu based tasks but still more than fast enough for usage. GPU is important & especially to some of us that use emulators to play with. At the end & multimedia & other high parallel job’s are very important to most people. & the last bullshit about how you don’t need a bigger gpu I live to you as you quoted MTK. Even if your app doesn’t full potential of the gpu 4 clusters at 300 MHz will actually consume less energy than 2 same clusters @ 600MHz. That & similar statements are for cheap basterd’s that make a cheap SoC’s wile trying to convince you how their shit is better than others people gold.

Well they used actual SoC’s that are currently available. Every SoC made this year will have a thermal throttling issues & will go over 2.5W if you know how to tax it including undercloked MT6735 @ GHz & S410. Actually they will go over 2W jest with CPU’s fully utilized & you can test this with a profiler & let’s say 7 Zipp let it compress more than 5 min with a screen on. GPU is a lot bigger cluster vs CPU core & it spends a lot more power naturally.

Now let’s talk a little about golden values of silicon when it comes to to power consumption. As the material leaking starts at 400MHz, it’s sustainable up to let’s say 1GHz after that leaking is huge & power consumption jumps almost progressive. As I explained you that even quad core A53 will brick a power limit (in severe use) their is not much to talk about. Every simple manufacturer rapes the Silicon pushing frequency & trying to save on a die size to minimize cost & max the profits. This is not really a good engineering. MTK studs up among others by pushing this even more than others. When it comes to to GPU’s Qualcomm stil have edge as they still go with monolithic single (large) cluster design. This approach costs more to develop but gains are that it scales 100% performance vise to a solution where you add more than one cluster, for example first T760 cluster gives 100% & every other around 68~70%. Naturally this approach will consume & less power (one cluster) for same lv of performance. I wouldn’t call the A520 a weaker GPU then T880 MP4. All Adrenos have a problem with any kind of surface texture transrorma & it shows in this particular test & even more in Antutu 3D Anarchy driven with lots of water surfaces, old ATI design flow. 3D Mark is let’s say best one for comparation currently available. Driver’s are another story but I won’t go into that this time let’s just say how & Mali & Adreno are on a bad side there.

The S652 is a direct competition to X20 & it’s a better product & will be higher priced as always but those 10~15$ doesn’t change much in a final product pricing.

Thing is that S652 is best balanced SoC announced up to date that will give a best user experience across the board; faster than S810 in CPU tasks (most folks would agree with me this is enough), good multitasking abilities, good enough for FHD gaming, rather good multimedia capabilities including 4K video recording & cetera thanks to QDSP 680, good camera IPS, certainly better cellular radio than competition (at least Aple knows what he is paying for).

MTK had a design victory this year with MT6732 instead of keeping that & adding cellular radio to the SoC (as integrated part) they made a cheaper & much worse 67×5 series. Even worse part is that MT6732 & MT6752 parts are discontinued without any support any more.

So I ask you once more would you actually pay 15$ more for a product that is better balanced to your possible needs & will be nicer to the battery life & dose have a secure open stock suport (CAF) for at least 2 years (as you won’t use that phone probably more than 2 years)?

It’s actually a simple question & don’t think that I am a large fun of QC (I actually hate them).

you are the man!

do you have a master degree in engeneering or so?

I can tell that other than the passion you also have knowledge.

it’s nice to know something more about these stuff even though I never go into this depth.

Happy holidays, see you around.

Actually their is a need for another dedicated multi purpose cluster but not as aditive or inter cluster one. Up until now it whosent been possible to implement one as their whose not a ARM v8 (64bit) licensable (or any other) solution that whole power/cost optimized enough to justify it. Now with Cortex A35 this change it a lot. When I say dedicated cluster I mean one made out of Two A35’s clocked up to 400 MHz with additional DSP blocks that will be able to address all needs of a device in a inactive mode (messages & cetera along with wite noise and storage IO’s & other light tasks as well as music playback) leaving bigger ones in a sleep mode wile enough powerful for waking up process to go smooth. In a active use state of the device they still can perform a peripheral off loading tasks from other bigger clusters as storage IO controllers, for music playback & cetera.

For a main tasks and mainstream the two quad clusters in a big litle config still stays a best solution considering & multitasking needs. The next engineering task would be a creating a better & more efficient mid range performer as for instance Krait cores whose in a ARM V7 (32 bit) architecture or Cortex M7 still is. Using only one quad cluster with a scaling up conditional between 3 main states (golden 400MHz, balanced 1~1.2GHz & max more than 1.2GHz) with let’s say values of 82% or more max load for a period needed to make a blink of an eye would be a best that it can be achieved right now considering user experience quality/power consumption/performance metric. Along with a small dedicated cluster I mentioned of course.