Witnessed the improvements over performance for the past decade, the one thing I miss the most is battery life. I'm not even talking about Nokia, the first two generations of iPads, on pre-iOS 7 firmware, could last for a week with mixed tasks that are not extremely power consuming. I also love the Galaxy Note 3 and 4 on their original firmware that they could last for a whole day even when playing games like ports for GTA series. Now, flagships (like my s8) can't last longer than a day even without playing games. We already have enough performance to sustain most tasks needed so it's time to focus on efficiency...
Performance should be a factor. She charge our phones every day or every second day, but computing performance is still increasing. So why bother talking about efficiency, which doesnt change in reality much. You getting more powerful devices, but to restrain heat you have to reduce manufacturing process or make batteries bigger than people want slim phones, very difficult tasks and for years to come it will not change, we will keep charging our phones every few days, but performance will increase, so yeah forget efficiency, performance what is matters
funkboy, 09 Apr 2018I always thought it would be smart to have: - at least one A7n class "big" core to handle ... moreNope. One big core isn't enough as most of the tasks are SMP 2 in their nature so you need two & that is optimal. Truth is integer tasks scale really bad in SMP while Floating Point (FP) one's scale fairly good. As in the current state & difference between VFP units and NEON SIMD's between big & little core's and operating power consumption it's always better to execute FP tasks on the little core's especially ones written for more (than two) SMP threads but those are very rare. In the end current scheduler logic prevents us from issuing tasks specifically like that. So the two big cores plus four little ones is still the best possible setup.
It's actually very simple. Big OoO core's are 1.74~2.2x faster in integer tasks compared to in order one's so they are needed all do they consume 3~5x more power. Problem is that silicon vendor's are pushing the clocks beyond silicon sane limits.
A75 is 50% bigger than A73 which also means it uses 50% more power it's also 25% faster (MHz per MHz) doing so.
Power consumption is doubled on 2.6 GHz compared to 2GHz so;
A75 @ 2GHz = 96% performance of A73 @ 2.6GHz with 75% power required.
I always thought it would be smart to have:
- at least one A7n class "big" core to handle relatively rare complex jobs that can take advantage of the extra instructions (NEON, intense FPU math, gaming, etc.)
- at least one tiny A35 core to be the only thing running in the phone while it's in standby doing background jobs & running the baseband stack
- & four A5n class "little" quick efficient cores for everyday jobs while the phone is in active use
& everything should be able to dynamically scale the clockspeed of individual cores like on amd64, preferably over as wide a range as possible.
In its minimal configuration, a 6-core SoC like that would be smaller & cheaper than current normal 8-core A53 chips, approximately as fast, & use significantly less power.
This should be possible with A75/A55/dynamiQ right?
It's more about using mixed approach. Good performance with good power efficiency always wins. The smarter SoC design is a key of achieving it alongside with no aggressive to high clock's. Unfortunately all vendors persuade the benchmark performance then optimized real use one pushing clocks rather than balancing power/performance metrics to used lithography. Also both OEM's and vendors don't do a really good job optimising power/performance trough software with S9 Exunos being last example of that.
Mi Max 2 loses sole in comparation to Mi Max, lack of big OoO core's on S625 simply can't compensate for it's increased efficiency. The end result is that Mi Max 2 has only 25% higher SOT with still a little bigger battery compared to the Mi Max but it also loses when it needs to be pulled hard 40% on CPU integer tasks. All do FinFET structure brings improvement in leakage limit enabling that target clock limits can be pushed bit higher (on CPU-s & GPU's) compared to planar it still cannot be used for RF nor mixed components & some of those like cellular radio modems are still big power consumer's which is one of the main reasons we don't see a higher efficiency improvement in general usage switching to FinFET. The S650 is a best example up to date of smart design approach while still keeping the costs down by both not choosing top lithography process availability at the time & having only components that it really needs.
For instance bringing the current S845 design to 22nm FD-SOI as it is would cut the price for it 40% down. Cutting retarded not needed components as two A75 cores & L4 cache would future increase that to 50%. That's twice the reduction in price. Limiting the CPU/GPU frequencies down to 2GHz top for CPU (big cores) & 600 MHz for GPU would reduce CPU performance for 20~22% and 20~24% for GPU while increasing CPU efficiency beyond the 22nm FD-SOI/10nm FinFET process difference & keeping it the same for GPU. When you add in the mix RF & mixed FD-SOI component in to the mix the one based upon 22nm FD-SOI would actually be considerably better regarding power consumption (needles to say separate FD-SOI RF components can pared to any FinFET containing pure digital ones). For comparison half cut strategy on FinFET cutting the GPU in half & doing the same thing for core's and cache like described for FinFET would lead to similar price reduction & would not had CPU performance penalty (tho for the purpose of differentiating it from top flagship one's the lower clock limit would be implied so it would be the same o at the end) but GPU penalty would be much higher, whole 50%. So much about better approach in designing the hardware it self from me. I used Qualcomm just as example same thing apply regardless of vendor.
For those who voted Efficiency go buy an ultra low performances cpu with a huge battery capacity lol and good luck
With sufficient performance, battery life and efficiency are all that matter
Just grab motorola z force, xperia xz2 and you have speed and enough to power all day
The faster a soc is the worse off battery life is. I don't care how fast a phone is: if it doesn't have adequate or above avg. RAM it will run less than smoothly. The Snapdragon S9 uses up almost all of the 4gb onboard, leaving about 1.3 gb of free ram available. This is partially due to Android 8.0 running the show, and also due to Samsung's own OS skin. I predict that with each new Android software, more and more ram will be required, despite what a particular soc is bringing to the table. I feel efficiency is the number one priority of handset consumers at the moment. Cheers all. :-) .
Unlike Camera vs Performance, this comparison shouldn't be heavily favored towards one as bit of both is required in a good phone. It actually depends on what performance and efficiency are in this case? Like a phone with sluggish-UI would def. be worse than one losing juice in night or evening. On the other hand, a phone with good operating performance (even without good gaming performance) can surely be aided by efficiency.
Btw flagships are not cost-effective these days. They were when I had Nokia N95, I used it for 4 years after getting it 2x price of the regular smartphone. But these days a single phone is considered 'old' even after an year or two. Giving it to son won't change its market usefulness.
Anonymous, 08 Apr 2018If you want both just buy any Sony phone.Right...
Kiyasuriin, 08 Apr 2018Marvelous.
The question is, can you withstand the amount of math you have to bear? The very f... moreDon't worry about me, I love math and I'm already accepted in school. Going to college next year in Computer Science with Mobility and Security (not the college you might think about. Our education system has an interim infrastructure between high school and university)
ProJames, 08 Apr 2018Sorry. I was targetting big corporation programmers (Windows comes to mind with it's poor reso... moreMarvelous.
The question is, can you withstand the amount of math you have to bear? The very first WIDESPREAD language is C language. There are older than that. But C is the base for everything.
You will need to think as Developer and bear the responsibility that details matter.
Especially on the mobile industry.
Performance & efficiency can't go hand in hand unless there's absolutely perfected AI-infused chipsets which can determine to perform according to a specific task. For this, we need a new architecture which is intelligent enough to handle basic tasks with minimal performance and power draw. If there are 8 cores in a processor, only two should be used when performing basic tasks like WhatsApp etc.. One dedicated to handle background task and one to handle other tasks.
For me. I used to use Samsung S6, but charging every almost half a day was a quite a real pain, but then I looked here the phones that has quite power and good battery life and are quite small to handkle with one arm and I bought Sony XZ1 Compact as my daily driver for comunication, internet and listening to music with my Sennheiser HD598 with FiiO DAC and I'm very happy with my choise and I'm very gratefull that GsmArena helped me a lot with my decition, thank You all for Your tests and great work. Keep up and best wishes to GsmArena team and all of its fans. Peace.