Need help understanding real world context of ARM SoC performance in tablets

[arandomguy]

Short version - See title. Don't have understanding of benchmark numbers in relation to real world differences for CPU and GPU performance in different tablets.

Long detailed -

Not sure if this is the best section for this, since SoCs aren't strictly just the CPU, but there seems to be most discussion in this forum regarding SoCs.

Here is my issue. I can see the reviews and benches done between devices using different SoCs but unlike with PC parts I don't have a context of what those numbers actually mean in real terms.

Two different SoCs in two devices compared for example -
Exynos 5420 - http://www.anandtech.com/show/8197/samsung-galaxy-tab-s-review-105-84inch
Tegra K1 - http://www.anandtech.com/show/8296/the-nvidia-shield-tablet-review

I can see the Tegra K1 has a significantly faster CPU and GPU but I just have no context in terms of what that means in real usage.

Anandtech uses a lot of web browsing oriented benchmarks but I'm not really sure what they mean. The K1 is faster but in what ways does this actually improve usage? I don't recall noticing any browsing performance issues on both a Nexus 4 or Nexus 5. Does this mean they've reached that good enough phase and the numbers are more academic? Or are there certain websites that showcase the browsing performance differences (more media and image heavy?)? Flash content, which is resource heavy, isn't really a factor on mobile Android devices.

Tablet usage might be more application heavy. So what are some examples of application usage that would be noticeably better on the K1?

Or looking at GPU performance and gaming as something specific and what is strongly affected by performance differences on the desktop. Comparing a GTX 980 to a 750ti performance numbers I know that the former would run more demanding games at higher settings at 1080p while some games would effectively be the same. Even something specific like Crysis 3 would look better and run smoother at 1080p but League of Legends would not be much different. With the K1 vs Exynos I can see there also is a large performance difference but other then throwing out faster 3Dmark numbers what are real examples of better gaming?

Personally I've only used (owned) a Nexus 4 and Nexus 5 extensively. I don't really recall any large practical performance differences between the two. While at the same time the one performance limited usage scenario I had still could not be accomplished by either. Basically brute forcing non typical video decoding (such as 1080p hi10p, h265, 4k) which at this point don't have actual full hardware support (and in some cases maybe won't ever). Can some of these current SoCs do that?

 

[Qwertilot]

Its a good question

If you consider that various things with a7 (Arm not Apple) based things in them have reviewed without real complaints I think there's a strong argument that for most people the high end ones are already somewhat past good enough.

Obviously if you're trying to drive towards gaming (as various things are) its a bit of a different matter.

Also, of course, sundry Samsung things routinely launching using both Exynos and Snapdragon just depending on if they need the 3G modem from the snapgdragon or not.

 

[Roland00Address]

Atom Clovertrail uses the same architecture as the 2007/2008 intel atoms but just optimized more for power. Aka this is "crappy atom" but make it dual core and not single core and increase the clockspeed 25% and have a decent gpu that could accelerate stuff on the gpu once the software got there and a lean OS and the 2008 to 2012 atom is not that bad, not great, but not as bad as the 2008 version.

Atom Clovertrail is what the tablets from 2012 used.

Atom Clovertrail is >= Cortex A9 depending on clockspeed and the software implementation.

Intel Baytrail is what tablets from late 2013 and 2014 used. I need to relook it up but it is roughly 80% faster. The 1.8 ghz atoms are comparable to performance to the nice cortex A15 and the Qualcomm 801 and 805 skus.

Cortex A57, Denver 64bit dual core are even faster. We are starting to talk about very low tdp core speed like sandybridge and haswell. Remember the core architecture can have a big range in performance with ghz speeds like 1.4 ghz to 4.4 ghz a difference of 315% and differences in hyperthreading and cache. Cortex A57 and Denver are competing with are closer to that 1.4 ghz in performance of sandybridge and haswell and nowhere near the 3.0+ghz you see in the high end desktop pentiums, core i3s, i5s, and i7s.



Lets put it this way, these low tdp phone and tablet chips are getting close to the performance of core 2 duo desktops from 2006 and 2007, and laptop parts of the core 2 generation that you see in very expensive laptops like $1.5 to $2.5k in 2008 and 2009. Not the $500 to $1500 laptops but the very very nice laptop chips

And on very specialized tasks, like cryptography these arm processors are night and day faster by an order of magnitude over the core 2 generation. These tests such as cryptography are what you see that make things like an nvidia tablet or an apple ipad score so much better in a multi OS benchmark like geekbench 3. This is due to have specialized hardware to do these tasks on the chip soc instead of trying to used the general cpu to do it. Using the general cpu to do it and not fixed function hardware makes the cpu cry.

(now Geekbench is bad for other reasons but I digress)