Snapdragon 820 Previews

[dark zero]

OK, will only make one comment about this, but cant resist. This reminds me of the AMD fans who say "what can you expect, Vishera is an old architecture on an inferior node." Yea but that is the point, it is the job of a company to update and bring out new, competitive products. AMD actually has a better excuse than Intel, with their very limited resources. I dont know what is intel's problem in mobile.

I thought Bay Trail was a very nice advance, but now we get 14nm delays and uninspiring at best Cherry Trail. And still delays and mediocre products integrating modems. All the while, Apple and ARM just keep bringing out better chips and more attractive devices to further cement themselves in the market.

Bay Trail was decent.... compared to Cherry Trail who is a MASSIVE dissaster....
Even more, Moorefield is great.... however there is NO OS who are fully using it... and the only one who can (Windows 10), is not available for them....

 

[coercitiv]

Any info or estimate on die size for Kryo? Too early I guess.

 

[Andrei.]

Any info or estimate on die size for Kryo? Too early I guess.

Large.

 

[coercitiv]

Large.

Fair enough, we had to reach that day when silicon dies and soda portions are expressed with the same simplicity

 

[Nothingness]

Why do you always have to make it x86 vs ARM?

Come on, everyone can see you're the one who made the comparison by claiming Atom wasn't that far

In case you didn't notice I also posted Octane and PCMark results (and the OP has the Kraken score you're talking about).

But not in the post where you made your biased comparison (but see below why I'm not sure using the fastest score would be more interesting anyway). You're doing a great job putting information in this forum and I've thanked you several times in the past for that, but sometimes your posts show a heavy Intel bias and I don't see why I shouldn't point it.

Not that it matter because according to you javascript benchmarks are useless and WebXPRT favours Intel (still waiting for evidence on this) so it's pretty hard to come up with a comparison.

In my previous post, I pointed that the 820 JS scores are extremely good and I believe that if the browser was to be used on other ARM CPU the 820 wouldn't shine that much. Also I find it intriguing that an OoO core would do well on Geekbench3 while not doing so well on the non-stock browser. This is why I think JS benchmarks are misleading, or at least that I consider they are not enough to evaluate CPU performance.

Ryan and Andrei made it clear in their conclusion and I strongly agree with them:

As is always the case with these MDP previews, it’s critical to note that we’re looking at an early device with unoptimized software. And at the same time that we’re looking at a device and scenario where Qualcomm is looking to show off their new SoC in the best light possible. Which is to say that between now and retail devices there’s room for performance to grow and performance to shrink depending on what happens with software, thermal management, and more.

 

[Sweepr]

Come on, everyone can see you're the one who made the comparison by claiming Atom wasn't that far

The numbers are here, the difference isn't huge in (some) benchmarks like Basemark OS II 2.0, PCMark and WebXPRT. Neither Kraken and Octane running Chrome (except stock browser). Pardon me but considering how many people (including you) love to bash Atom, I was pretty surprised by the results. That's not to say Atom isn't in desperate need of an upgrade right now and I'm pretty sure there's a bigger gap between the two in other metrics, Intel dropped the ball by delaying Broxton.

 

[Thala]

Pardon me but considering how many people (including you) love to bash Atom, I was pretty surprised by the results.

That is maybe just maybe because you still judging by Javascript benchmarks?

 

[Sweepr]

That maybe just maybe because you still judging by Javascript benchmarks?

Sorry but other than Geekbench, Basemark and PCMark most results I found for the two are Javascript benchmarks. Can we get back to Kryo now? Otherwise I will request mods to lock this thread.

 

[Thala]

Sorry but other than Geekbench, Basemark and PCMark most results I found for the two are Javascript benchmarks.

It is not about what you found but how you interpret the results. You comparing a decent Javascript implementation for x86 in Chrome with a lackluster implementation for ARM. Despite both being Chrome these are essentially hand crafted code generators modules, where the x86 version has been heavily optimized by Intel. You essentially ignore the fact that there are much better Javascript engines for ARM available (e.g. stock browser).

I am not a fan of using Javascript as a tool for comparison for above reasons. You never know how much the code generation module has been optimized for the particular uarch. However at least choose a decent (or best known) Javascript implementation for each architecture.

 

[Sweepr]

I am not a fan of using Javascript as a tool for comparison for above reasons. You never know how much the code generation module has been optimized for the particular uarch. However at least choose a decent (or best known) Javascript implementation for each architecture.

Not saying that I disagree but you should tell AnandTech this, because they are still including results from different browsers in their reviews (for each platform).

 

[StrangerGuy]

From a nerd POV, its an interesting chip.

From a business POV however, they were already too late to the party because the high end Android device market has pretty much collapsed from the deluge of amazingly specced low-end devices and SoCs. $600+ Android devices were only a tiny 5.6% of the Android total by Q1 2015.

http://venturebeat.com/2015/08/01/600-premium-android-phones-are-getting-harder-and-harder-to-sell/

 

[Dribble]

From a nerd POV, its an interesting chip.

From a business POV however, they were already too late to the party because the high end Android device market has pretty much collapsed from the deluge of amazingly specced low-end devices and SoCs. $600+ Android devices were only a tiny 5.6% of the Android total by Q1 2015.

http://venturebeat.com/2015/08/01/600-premium-android-phones-are-getting-harder-and-harder-to-sell/

That's the problem for Qualcomm now, lots of people have got good at sticking some stock arm stuff together and ending up with something very cheap with decent performance. This has gone with cpu/gpu performance no longer being so critical in phones, and the drop in margins for selling phones. Hence any soc that's expensive and custom has a much smaller market then it used too.

That was part of the problem with the 810 - they had to make something faster and better then the cheap Chinese chips to be able to demand the premium they wanted, and in trying too ended up with something that throttled, but even if it had been perfect most phones would still use cheaper chips that were "good enough".

 

[kpkp]

3 top smartphone sellers have their own SoCs, Mediatek will be a strong competitor on the low to mid-high end where the Android volume is. Qualcomm midrange and low end SoCs are quite bad now, hopefully they learned something from the 810, this way they don't make the same mistakes with the 618/620 and we get solid mid-range SoCs for next year.

Andrei, weren't 618/620 supposed to be on the market before the 820? Any news on that?

 

[sm625]

I dont know what it is about web performance that these guys hate so much, but once again it is abysmal and they dont seem to care.

 

[Andrei.]

Andrei, weren't 618/620 supposed to be on the market before the 820? Any news on that?

They're coming in Q1 (basically a quarter later than promised). I think the Samsung A9 may be one of the first with the 620.

 

[Arachnotronic]

They're coming in Q1 (basically a quarter later than promised). I think the Samsung A9 may be one of the first with the 620.

620 should be a nice chip.

Hey, question, did you run 32 bit or 64 bit version of SPECint in the S820 preview?

 

[Andrei.]

64, 32b scores would have been higher.

 

[Arachnotronic]

64, 32b scores would have been higher.

Thanks for the clarification :thumbsup:

 

[Arachnotronic]

From a nerd POV, its an interesting chip.

From a business POV however, they were already too late to the party because the high end Android device market has pretty much collapsed from the deluge of amazingly specced low-end devices and SoCs. $600+ Android devices were only a tiny 5.6% of the Android total by Q1 2015.

http://venturebeat.com/2015/08/01/600-premium-android-phones-are-getting-harder-and-harder-to-sell/

I expect that percentage to come down. Mid-range Android phones are ridiculously fast and have lots of features.

And, if you're going to drop $600+ on a phone, might as well go iPhone.

 

[RussianSensation]

I have a feeling the Exynos 8890 will kick 820s arse overall. Some of the benches in AT's preview against the Note 5 are not impressive at all. While 820 is a huge improvement from 801/805/810 v2.1, the 7420's multi-core performance is very impressive. I have a feeling single core performance between 8890 and 820 will be close but 8890 will easily take it in the multi-threaded benches.

http://www.gsmarena.com/snapdragon_820_off_the_charts_on_antutu_scoring_131000-news-15468.php

This CPU gets absolutely clobbered in SPECint by the A9, often by 2x in some subteats including the infamous 176.gcc. Qualcomm has produced a Geekbench monster but it seems to not do so well when confronted by serious CPU workloads.

I'd argue that there is no such thing as serious CPU workloads on smartphones. Anyone interested in serious CPU workloads is using a Core i7 Quad+HT laptop or even a 6-10 core desktop. For most people buying Android smartphones, it starts with the price, then features like camera quality, battery life, microSD slot, screen size/quality, etc. Considering almost 95% of Android users do not buy $600+ flagship phones per the info above, it makes it more clear that performance is the least important factor for most consumers.

As far as Apple is concerned, because of work I have iPhone 5/5S/6/6S. For usability/productivity/media viewing, they all fail miserably compared to my iPad Air, IVB i7 15.6" laptop, desktop and their cameras in the evenings/low light conditions are garbage against Canon S110 pocket camera + my Olympus Micro 4/3rds. For basic text messaging, internet, phone calls, they are all good enough. I guess I don't care at all about smartphones after moving from Sony W760 to the iPhone 5 they all feel "the same".

It's getting harder and harder to get excited about smartphones since they are already good enough for basic tasks yet a decade+ behind when it comes to the real CPU/GPU performance + camera. With the companies chasing every last mm in chassis thickness while later trying to sell us ugly POS protruding $100 sub-2000 mAH cases, it seems until we get some revolution in battery technology, we'll keep getting stuck with phones that can barely last a day, way less for someone who uses it for productivity.

When I can take a smartphone, drop it into a dock that powers 3x1440P 144Hz OLEDs or even a single 4K monitor and can then play modern AAA PC games @ 60fps with a Bluetooth controller; have 4-5 days battery life in regular usage; have a camera which is at least as good as that of Sony Cyber-shot DSC-RX100 IV 20.2 MP, then I guess I'll get more excited. Otherwise, all these CPU+GPU benchmarks are just good marketing/power point/bar chart graphs without much real world revolution as far as actual user experience goes from what we already had in say 2012.

 

[ChronoReverse]

I'm most disappointed with the fact the CPU is a big.little design but still can't play in the same ballpark as Apple.

Apple has managed to create a faster and more power efficient CPU than anyone else without resorting to big.little. Surely they're not the only company with talented engineers? Do they have a patent on some method that nobody else can use?

 

[Andrei.]

Apple has managed to create a faster and more power efficient CPU than anyone else without resorting to big.little.

Until I actually dismantle an iPhone to measure its power the claim that Apple has more efficient CPU is just a hypothesis.

 

[ChronoReverse]

Until I actually dismantle an iPhone to measure its power the claim that Apple has more efficient CPU is just a hypothesis.

You're absolutely right but I think it's a reasonable guess since there are few if any indications that it's not true.


For instance, heat generation is an indirect indicator of energy use of a CPU. It doesn't tell if it's resulting from handling the load, leakage or whatever but ultimately that's still expended energy. I'm not one of those who disparage "turbo" type methods but it's readily apparent that Apple's CPU's are able to sustain load without the same levels of heat generation as others. It's possible that Apple has superior heat dissipation but at steady state, it's not going a big difference compared to other aluminium case phones.

Battery life is another indirect indicator. Now we have things like GPU, radios and the screen which are huge contributors to battery use but these can be relatively minimized if we're deliberate about it in an synthetic test. Airplane mode can be used to shut down radios during an endurance test and the screen can be dimmed to a minimum (or set to black for AMOLED in which case Apple has the disadvantage). Either way, iPhones have a significant battery capacity disadvantage in the first place but you're not going to find any Android phone matching the iPhone under load conditions even if only the Android phone gets to dim the screen! Actual usage is par but that's only because Android phones (and even Windows phones) have a battery capacity advantage.

All this is merely indirect evidence of course. But the thing is, I can't even think of indirect evidence to the contrary!


You're completely correct that it's just a hypothesis but if Apple's CPU's are in fact equal or even a little worse in efficiency compared to the best big.little CPUs available right now, this would still be an indictment on the big.little scheme which has yet to be proven to be worth all the additional complexity it brings. Certainly at load it loses out entirely and for low load situations it hasn't been readily apparent that it's advantageous to what Apple has done.

Snapdragon 820 may change my mind on this. I'm not too hung up on its absolute performance but it would need to demonstrate tremendous efficiency this late in the game.

 

[ChronoReverse]

I do admit that I'm frustated by big.little as I've been spending quite a bit of time tracking CPU usage on my Nexus 6P and adjusting CPU governor and thermal settings. The thread migration doesn't always work as expected and there are poorly written apps that will never move to the big core (wtf Google Camera). It doesn't help that sometimes weird things happen when a thread _is_ migrated leading to an application acting in ways otherwise wouldn't happen (again because it turns out it's not 100% transparent in practice).

It's simply hard to not feel that a super fast dual core would avoid the issues and have similar performance from the viewpoint of the user.

And the thing is, I'm not even against more slower cores. I find that Krait was a perfectly acceptable solution to the problem for its generation. But big.little has yet to mature (current implementations still leave a lot to be desired) and I'm not convinced that it will reach the point where it works transparently and efficiently at the same time.

 

[Andrei.]

I do admit that I'm frustated by big.little as I've been spending quite a bit of time tracking CPU usage on my Nexus 6P and adjusting CPU governor and thermal settings.

The 6P / S810 are not a good demonstration of big.LITTLE, the SoC is just simply bad in more ways than one. The 7420 has like twice over the perf/w and even the 5433 is far ahead.

In regards to Apple battery life etc, iOS itself is just incredibly more efficient than Android and I would bet a that has a lot to do with the fact that iPhones still compete with their little batteries. Again unless I get an iPhone and measure power in something like SPEC then I doubt large efficiency advantages.