NVIDIA Tegra Architecture

Interesting. Didn't know they would make Support libraries like that. The App page doesn't tell you much about what exactly it does though. I can't play with it anyway, as I don't have any Samsung Android devices here.

 

It didn't make any difference, scores are the same
Those libs are quite a bit outdated, last update dated by April 2, 2014, I doubt they contain any optimizations for A57

 

Thanks for that. Reduced the sunspider score from 11xx to 7xx on my Snapdragon Note 4.

 

Yes true. Let's hope Google updates it in the future.

Great.

 

Do you have Exynos or Snapdragon based phone?

 

Snapdragon. N910F

 

It explains the gains
Could you please test 7-zip benchmark which I've compiled with gcc4.6? (Anandtech uses this bench for desktop CPUs) It does not require the root access, you can push the exec file via adb to device - "adb push 7za /data/local/tmp" or via the terminal with "cp /sdcard/7za /data/local/tmp", need to type "./7za b -mmt=4" to launch it (-mmt=4 - the number of threads), here is some additional info - http://www.7-cpu.com/ . I've added results for Tegra K1 and Snapdragon 800
Here is results for Exynos 5433 - http://forum.ixbt.com/topic.cgi?id=8:24730-15#335

 

That Samsung chrome download have been bugging me for months on what exactly it is.

 

http://www.anandtech.com/show/8670/google-nexus-9-preliminary-findings

The important part is the Basemark OS battery test. I estimate display power to be around 1.5W, so that leaves about 6.6W for the CPU bound test and SoC power. Lets say 6W to be optimistic.

 

Going by those Basemark battery scores Denver TK1 does look significantly more power hungry than the TK1 found in the Shield tablet.

Slightly disappointing that it's basically a slew of browser based benchmarks. Can't really tell how power efficient Denver is until we have a better idea on how it performs. Hopefully there will be a more thorough low level analysis in the full review.

OlegSH, Here you go:

http://i.imgur.com/cwsX891.png

1,2,4,8 and 16 threads: http://pastebin.com/6ypLVu92

Not sure how to interpret the results. How does it compare to A57?

 

I read other reviews around the web before coming around to that one. It certainly stands out as being the most... effervescent.

Edit: Gizmodo, The Verge, and AndroidPolice thought the N9 was mediocre. Engadget was positive but unmoved. In terms of things related to the Tegra, the first 3 sites all noted intermittent slowdowns/lag while doing normal things. Verge and AP also found that their units got hot during low-key usage.

Maybe this will be all fixed with more OTA updates over the months. Who knows.

 

In theory, if you were to create custom apps from scratch with Denver as the sole SoC target (ie automotive industry), would that be the ideal case scenario for Denver's code transformation?

I can't help but think that the lack of cheap & plentiful 20nm wafers in 2014, and thus the ability to get a jump on Qualcomm in supplying a high-end 64-bit SoC were the driving force behind Denver, rather than a revolution in perf/watt vs A57+A53.

 

GSM Arena was very impressed with both the performance and battery life of Nexus 9:

http://www.gsmarena.com/htc_nexus_9-review-1160p10.php

They achieved nearly 10 hours of web browsing time and nearly 12 hours of video playback time: http://www.gsmarena.com/htc_nexus_9-review-1160p3.php

A user at XDA developers forum has purchased a unit and his first impressions are "Very smooth. Opened 10 Tabs in Chrome. Opened some other apps like tweetings, friendcaster, google+, inbox switched fast between them without any micro lag.".

 

Anandtech's preliminary writeup until their review is done is good enough for me. That's another problem we have for that market: serious sites aren't yet dealing with the ULP mobile world and experienced reviewers that touch the topic still a sad minority.

 

Thanks, at 4 threads quad core A57 configuration is 13% faster overall ( total score = compression mips + decompression mips divided on 2 ) mostly because of significantly faster compression rate of A57 due to better OoO, while Krait is a little bit faster at decompression thanks to much higher frequency. The most interesting part is single thread and 8 threads performance. A57 cluster works in conjuction with A53 cluster with 8 threads in Exynos 5433, therefore scores are much higher with 8 threads - 8075 for total, compressing - 5599 MIPS, decompressing - 10550, while single threaded A57 perf is on par with Core 2 Duo at the same frequency - 1905 for total at 1.9 Ghz, which is pretty good for smartphone processor

 

It seems nobody cares to use normal benchmarks like 7-zip, PCMark and ANDEbench PRO on Android, I wonder why. Even some emulators tests would be much more usefull then some shitty Basemark OS results

 

Hard to imagine Motorola and Google would release the product if they were seeing 4 hours of battery life during development and testing.

Must be an outlier.

 

The guy who claimed 4-5 hours of battery life had the brightness settings cranked up while spending a significant portion of that time running very CPU-intensive and/or GPU-intensive benchmarks and while also zipping through one webpage after another after another to test the device.

Phone Arena, GSM Arena, and Anandtech have all measured between 9-10 hours of WiFi web browsing battery life when used in a more realistic way.

 

I think it's starting to sound obvious that Denver is still in the middle of a large software tuneup.

From Wired:

Edit:

"only"

 

Clearly something was not right for him if games went from being "extremely jerky" to playing "incredibly well" with one OTA update.

FWIW, the users on XDA developers forum who have the tablet so far are reporting excellent real-world performance.

Here is what CNet had to say: "Navigating the OS is swift and switching between apps is so quick it's almost eerie. Performance also remains impressively smooth when many apps are open in the background. At one point I had over 20 apps running and I was still able to play large games without a hitch." That said, there is no doubt that some apps may need to be updated or optimized for the new Android "L" OS, and even the OS itself needs some work to iron out the kinks.