lol @ Qualcomm

[shabby]

If you haven't read the new 5x review go do it now http://www.anandtech.com/show/9742/the-google-nexus-5x-review/3 scroll down to Sustained Performance, this i believe should be the new benchmark of smartphones.
Every time qualcomm releases a new cpu its 30% faster, 25% more efficient, has faster ram, faster gpu...

Snapdragon 808 can only keep its two A57 cores at their peak frequency for two minutes before throttling both down to 633MHz and putting the A53s up to their peak 1.44GHz. After twelve minutes the A57s are just shut off entirely, and you're left with a cluster of 4 A53 cores at 1.44GHz.

All those fancy new features are only good for 12 minutes?

 

[postmortemIA]

in order words, specs are fake. that's why 4xxx and 6xx series are better bang for your bucks.

 

[lopri]

The phone does not stop functioning after the 12 minutes. A53s take over whenever necessary, which is how big.LITTLE works by design. It would have been certainly better to have more consistent performance throughout varying workloads, 12 minutes are more than enough for plenty of mobile needs that require heavy lifting.

Games are not going to push 4 x A53s and Adreno 418 on 1080p to the limit anytime soon. That would alienate majority of users out there. That leaves 4K video encoding as a weak point in real life usage scenario. And people often do not take videos longer than 10 minutes because chances are their arms will be fatigued before the SOC gives up.

I am not defending Qualcomm here. I am saying that Google have found a nice balance between power and performance in the Nexus 5X.

 

[Mopetar]

Hopefully they do a lot better with their new custom core that they've been working on as their foray into big.LITTLE has been disastrous.

Have they done this kind of analysis on other SoCs thought? It's a little unfair to criticize Qualcomm for this if the competition isn't any better.

 

[lopri]

Hopefully they do a lot better with their new custom core that they've been working on as their foray into big.LITTLE has been disastrous.

Have they done this kind of analysis on other SoCs thought? It's a little unfair to criticize Qualcomm for this if the competition isn't any better.

Not to mention that "Race-to-Idle" is what has driven mobile SOC developments. I have just seen a relevant article on the front page. (http://www.anandtech.com/show/9751/examining-intel-skylake-speed-shift-more-responsive-processors) Smartphones are not tortured to run multiple instances of Folding@Home.

 

[shabby]

The phone does not stop functioning after the 12 minutes. A53s take over whenever necessary, which is how big.LITTLE works by design.

So i guess the thermal issues were part of the design? Compared to the sd800/805 that were tested and you can see just how big of a joke these big.little soc's are. I wonder how long the a57 cores will last on the 810, 5 minutes before shutting down?

 

[senseamp]

Small plastic phones aren't designed to run Basemark continuously. They aren't HPC supercomputers. Vast majority of people need full CPU in short bursts for responsiveness, and a big core does that fine. 12 minutes is excessive. No well written mobile phone program should need to max out a SD808 CPU full throttle for for 12 minutes.
People obsess about benchmarks, but you should know that mobile benchmarks are giving you the peak performance of an SOC, not sustained performance.

 

[lopri]

It is interesting that many reviews recently discovered the importance of sustained performance on a smartphone yet they have nothing to show for it other than benchmarks. Furthermore, these reviews do not seem to have any idea how to qualify a sustained performance in the context (i.e. a sustained performance of xx minutes are useful and/or necessary if users want to do y) , other than offering a fairly generic notion of "the longer the better."

Yet it is not entirely surprising that use cases for such scenarios are rare - we are dealing with a mobile device with limited power source, after all. Laptops are used for surfing the web on the go. Desktops are used to serve those web pages. I would not be inclined to run a server on a laptop, no matter how efficient and performant it is. Running on a battery has its limits. Users are unlikely to use a smartphone to run a task that they know will 1) take long, and 2) is power-hungry.

Games are an exception, a rather big and important one at that, but I have not run into a review that says games X, Y, and Z suffer from the relatively "lower" sustained performance of a given SOC in comparison to another. That is hardly surprising considering that games are often more responsive to GPU portion of the SOC (let alone limited by display) and mobile SOCs are still progressing at a break-neck pace.

What is surprising is that reviews repeatedly double-down on this newfound importance of sustained performance with nothing else to show but for benchmarks and marketing slides. It seems to me like a rather uncritical attitude to take. (or worse)

P.S. I have found that the 5X runs into trouble when recording 4K videos for long period of time (i.e. 10 minutes+). It seems like a genuine problem stemming from the limits imposed by the SOC, and that means I would advise against the 5X if one likes to record and play 4K videos. (likewise, slo-mo videos)

 

[Mr. Pedantic]

It is interesting that many reviews recently discovered the importance of sustained performance on a smartphone yet they have nothing to show for it other than benchmarks. Furthermore, these reviews do not seem to have any idea how to qualify a sustained performance in the context (i.e. a sustained performance of xx minutes are useful and/or necessary if users want to do y) , other than offering a fairly generic notion of "the longer the better."

Yet it is not entirely surprising that use cases for such scenarios are rare - we are dealing with a mobile device with limited power source, after all. Laptops are used for surfing the web on the go. Desktops are used to serve those web pages. I would not be inclined to run a server on a laptop, no matter how efficient and performant it is. Running on a battery has its limits. Users are unlikely to use a smartphone to run a task that they know will 1) take long, and 2) is power-hungry.

Games are an exception, a rather big and important one at that, but I have not run into a review that says games X, Y, and Z suffer from the relatively "lower" sustained performance of a given SOC in comparison to another. That is hardly surprising considering that games are often more responsive to GPU portion of the SOC (let alone limited by display) and mobile SOCs are still progressing at a break-neck pace.

What is surprising is that reviews repeatedly double-down on this newfound importance of sustained performance with nothing else to show but for benchmarks and marketing slides. It seems to me like a rather uncritical attitude to take. (or worse)

P.S. I have found that the 5X runs into trouble when recording 4K videos for long period of time (i.e. 10 minutes+). It seems like a genuine problem stemming from the limits imposed by the SOC, and that means I would advise against the 5X if one likes to record and play 4K videos. (likewise, slo-mo videos)

This is what the Nexus 5X does:

This is what I'd expect:

 

[lopri]

^ Indeed. ()

 

[core2slow]

The galaxy s3 was the last phone I owned with a snapdragon. I remembered how laggy it was compared to my then international exynos galaxy s3, but not only that, battery life was horrendous as well. Judging by the S810 fiasco, nothing has changed with Qualcomm...and I doubt things will change with the s820, either.

 

[senseamp]

808 is a 20nm planar part, where as A9 is a 16nm FinFet part which is on a different Perf vs Power curve. 808 power management is doing the best with the Silicon it has to work with. The alternative would be to run at 25fps from the beginning instead of running 50% faster for first 12 minutes and then thermally throttling. But then the user who just wants short bursts of high performance for responsiveness would be forced to live with a lower sustained TDP level of performance.

 

[sweenish]

This should surprise no one who has been following Qualcomm's failings this year.

Here's hoping the next gen is better, or Intel or someone else is able to step up.

 

[Yuriman]

This is what the Nexus 5X does:

This is what I'd expect:

Seems to me that Apple left performance on the table, then. Their CPU/GPU can run at max clocks indefinitely, which means they probably could have implemented a very significant turbo.

 

[Bryf50]

SoC manufactures need to start changing the way they report clock speeds. It should be like Intel with a base clock that can be sustained for long periods and a boost clock. Manufacturers should then build the thermals of their device to match.

 

[slackerinabox]

I don't know enough about cpu design to understand that, so what does this mean in lay terms?

Are the qualcom cpus trottling excessively to conserve battery? Does this effect performance, or is this just rush to idle in action? If this is done to conserve battery life, then why is battery life still pretty anemic? It genuinely seems as if this generation is a step back performance and energy sipping wise, how did that happen? Lower margins, stiffer competition, and the need for dirt cheap SOCs? Is this just the gigahertz race again? Are mobile gpu's theoretically faster, but running at their max potential would mean horrible battery life, right? THen aren't these just overclocked last gen cpus? Does the rush to 64 bit have anything to do with this? Was apple so in the lead that, to catch up, the competition had to screw the pooch and mess up the details?

Something feels weird these last 2 generations, efficiency took a backseat to absolute performance, but even then max performance seems only to be theoretically attainable (like my s6 edge blew away benchmarks but stuttered, over heated, and had awful battery life when not in under-clocked low power mode). So what is the point then?

And what does this mean about Apple's a9? is it running at max potential all the time, or is there seriously untapped potential left out there to scale for a few years (mhz bumps and cache adjustments, minor changing until the A11). Could the A9X on the iPad pro be something that intel should be very afraid of?

 

[sm625]

If you haven't read the new 5x review go do it now http://www.anandtech.com/show/9742/the-google-nexus-5x-review/3 scroll down to Sustained Performance, this i believe should be the new benchmark of smartphones.

Not really. The problem with the Snapdragon 808 isnt that it has low sustained performance. It simply has low performance. The loss in performance after 12 minutes of heavy use would be completely and totally forgiveable if it were able to match an iphone 6s at Kraken. But it cant. Not even close, not even if you ran the benchmark at midnight, outside, in January, in Minnesota. It's just not a good design.

 

[dawheat]

Not really. The problem with the Snapdragon 808 isnt that it has low sustained performance. It simply has low performance. The loss in performance after 12 minutes of heavy use would be completely and totally forgiveable if it were able to match an iphone 6s at Kraken. But it cant. Not even close, not even if you ran the benchmark at midnight, outside, in January, in Minnesota. It's just not a good design.

Eh it's a mid-range SOC that is in phones starting at $200, though obviously the Nexus 5X starts higher at $379. Not sure why anyone would expect it to match an iPhone 6 in anything, even for a short burst.

I'm ok with it in lower mid-range phones as it gives decent burst performance and I'd take it over the 615 any day.

 

[senseamp]

SD808 is built with a 20nm planar process. It's more comparable to the A8. It's not as efficient as 16nm FinFet that A9 is using because FinFet lets you tune the transistors to run at a lower voltage without making them too leaky. SD820 will be a 16FF chip that will be more apples to Apples with A9. It may still be worse or better, but you can then compare at same process node.

 

[Mopetar]

Isn't the lack of throttling with the 6S due to the fact that the resolution it needs to drive is a lot smaller because I could swear the 6S+ did show some throttling in the same test simply because it has to do more work.

Perhaps I'm just misremembering, but could you link to the articles where those charts came from?

 

[ChronoReverse]

Not really. The problem with the Snapdragon 808 isnt that it has low sustained performance. It simply has low performance. The loss in performance after 12 minutes of heavy use would be completely and totally forgiveable if it were able to match an iphone 6s at Kraken. But it cant. Not even close, not even if you ran the benchmark at midnight, outside, in January, in Minnesota. It's just not a good design.

First, the 808 isn't up against the iPhone 6S, it should be up against the iPhone 6 (the 810 is up against the 6S).

Second, Kraken is a javascript browser benchmark that is affected FAR more by the browser being used than the CPU. Even a flawed benchmark like Geekbench is better and shows a more even picture vs the iPhone 6: http://browser.primatelabs.com/geekbench3/compare/4138100?baseline=4138100


Finally, we already know that this year's Snapdragon's are bad. It doesn't help that TSMC messed up their 16nm so it's all on 20nm (whereas Apple gets to be 14nm/16nm). I suspect the 810 would at least be competitive with Samsung's Exynos at 14nm/16nm even if it would not match the homerun the A9 is.

 

[Andrei.]

Not to mention that "Race-to-Idle" is what has driven mobile SOC developments. I have just seen a relevant article on the front page. (http://www.anandtech.com/show/9751/examining-intel-skylake-speed-shift-more-responsive-processors) Smartphones are not tortured to run multiple instances of Folding@Home.

I think you misunderstood that piece. Race to idle has never existed in mobile SoCs. It's a myth.

 

[poofyhairguy]

I suspect the 810 would at least be competitive with Samsung's Exynos at 14nm/16nm even if it would not match the homerun the A9 is.

Any SoC that is a stock ARM design isn't going to be a homerun. It takes a custom core to have a real homerun nowadays. ARM can't keep up with Apple or Intel, but Qualcomm hopefully can with the 820.

 

[ChronoReverse]

Any SoC that is a stock ARM design isn't going to be a homerun. It takes a custom core to have a real homerun nowadays. ARM can't keep up with Apple or Intel, but Qualcomm hopefully can with the 820.

That's what I meant. That 810 would have been just decent instead of the mess it is instead if it had been put on the same process. And that it still wouldn't keep up with the custom cores.

 

[Bryf50]

Any SoC that is a stock ARM design isn't going to be a homerun. It takes a custom core to have a real homerun nowadays. ARM can't keep up with Apple or Intel, but Qualcomm hopefully can with the 820.

Exynos 7420 did pretty well with stock ARM. Maybe not a "homerun" but it's where it needs to be performance wise.