Capture Latency

Along with its own rear facing camera sensor, the GS5 adapts a dual-mode autofocus system. Traditional AF designs in smartphones rely on testing contrast differences in order to determine when a point is in focus. With GS5, Samsung introduces a second mode: phase detection auto focus. Instead of relying (primarily) on contrast to determine focus, PDAF splits incoming light and compares intensity patterns to determine focus direction.

Low light scenes and scenes that can't generate appropriate phase info will default to contrast AF. PDAF appears to work across the camera's FOV although I did notice that the fastest AF times almost always happened in the center of the display.

Samsung claims a 300ms AF time thanks to its phase detection AF. In order to validate that claim I pointed a camera at a bunch of phones and measured AE/AF latency while preparing our ISO 12233 target shots from the previous page. 

I measured from the moment I tapped the focus target to the time the image stopped moving (I didn't rely on the AF lock indicator as some devices report focus lock prematurely). There's a bit more variance than I'd normally like in these numbers due to the nature of the capture, although I'm working on getting a higher speed camera to smooth some of that out.

Camera AE/AF Latency (Shooting ISO 12233 Target)

The GS5 definitely runs its AE/AF routine quickly, basically tying the M8 at the top of the charts here. The improvement over the GS4 is substantial, and there's even a big advantage over Apple's iPhone 5s. Note that if you move to lower light conditions you can see this number at least double, but that's something that impacts all of the devices here.

The Z1s is pretty frustrating because it has a great imaging system but an absolutely terrible camera UI. Focus speed is pretty bad compared to anything else here, basically on par with the iPhone 5/5c. 

The shortest time to focus I was able to record on the GS5 was 450ms in a different test scene, compared to 516ms for the M8. The GS5 can definitely be a hair faster but I found the M8 to be comparable if not slightly quicker overall.

Focusing is just one piece of the puzzle, I also measured capture latency as well. Here I'm looking at the time between when I tapped the capture button on the screen and when the camera UI was ready to take another shot. 

Camera Shot Latency (Shooting ISO12233 Target)

The GS5 remains solid, but here Apple actually pulls ahead. The 5c (and 5 by extension) are actually at the top of the charts here. Apple does some more work upon capture on the 5s, which is the only reason I can think of for the discrepancy here. Either that or the NAND on my 5s is in a dirtier state, impacting capture performance. 

Either way there's a huge improvement in capture speed compared to the GS4. Capture latency is one area where the Nexus 5 is absolutely horrible in. The latest updates made AE/AF reasonably responsive, but the capture latency kills the experience on the N5. 

This next chart combines the previous two values to give an overall picture of capture latency on these devices:

Camera Latency (Shooting ISO12233 Target)

Samsung's PDAF and ISP companion seem to do their job well as the GS5 is substantially quicker than the GS4 at image capture. That being said, HTC's M8 is slightly faster by comparison. 

Camera Architecture & Still Image Analysis That Fingerprint Scanner, Again
Comments Locked

296 Comments

View All Comments

  • devione - Wednesday, April 9, 2014 - link

    LG G2 = 143g
    Galaxy S5 = 145g
    Xperia Z1 = 170g
    HTC One M8 = 160g

    Apple iPhone 5S = 112g
    Sony Xperia Z1C = 137g
    One Mini = 122g
    Galaxy S4 Mini = 107g

    I think the Z1C stacks up pretty well, considering it punches well above its weight compared to the competiton.
  • doobydoo - Friday, April 11, 2014 - link

    It doesn't, though. It's heavier and slower than a 5S. Larger, too.
  • darkich - Tuesday, April 8, 2014 - link

    Superior review, Anand!
    I'd really like to hear what Raymond Soneira would have to say about your display measurements
  • dishayu - Tuesday, April 8, 2014 - link

    "I find that pretty much all the flagships offer some set of tradeoffs that prevent any one from being the perfect device (iPhone's screen size, GS5's materials, M8's camera). It's unfortunate because I'd really like to crown a single device the king of them all, but instead we're faced with a handful of differing optimization points."

    Try Xperia Z2. I don't really see any significant negative points with that device.
  • tipoo - Tuesday, April 8, 2014 - link

    Is there a way for you guys to test the display touch latency, as shown here? I'm interested in this, the M8 and the Note 3 before it finally made Android phones faster than the iPhones touch latency

    http://forum.xda-developers.com/showthread.php?t=2...
  • AnnonymousCoward - Sunday, April 13, 2014 - link

    Yes, I agree
  • HangFire - Tuesday, April 8, 2014 - link

    I'm glad to see more testing of the cameras and screens, it is too bad the M7 wasn't represented in the screen shots, as it is a very high density screen. I realize you can't test everything and a last-gen phone is not high on the priority list. On that note there is no excuse for no including a Nokia in the camera tests. My better half's 1520 cam makes this M7 user cry, it is that awesome, and that's not even Nokia's best.
  • ΔrNTA - Tuesday, April 8, 2014 - link

    Am I wrong in saying untill Qualcomm start making the SnapDragon 805, all phones exept the obvious one are already out of date until next year?
  • kyuu - Tuesday, April 8, 2014 - link

    Yes, you're wrong. The 64-bit thing is totally overblown. It's not going to make any practical difference until we get to a point where 4GB of RAM is a limiting factor in mobile devices. Which won't be for a long time.
  • kyuu - Tuesday, April 8, 2014 - link

    Oh wait, you said 805 not 808, my bad. 805 looks to be a largely incremental update, asides from the GPU. So yeah, in a sense I guess you're right. I don't think the difference is anything to be so concerned about that you'd delay a purchase for a half year or so over it, though.

Log in

Don't have an account? Sign up now