Low FPS in C2 games is more crippling than it should be?

Hmm, interesting results from Colludium - Did you every try it out on Chrome 38?

Something else I'm a little worried about is that this test doesn't account for the significant input delay that I also experience when this GPU lag is happening. Ashley, could you ever reproduce that?

sqiddster

I tried to revert but v39 is now the latest release. It's worth adding that I'm running the x64 version of chrome. Here are my complete chrome beta / stable results:

Chrome v39 (Version 39.0.2171.27 m (64-bit))

19.1 - 25.2 ms; however, there are spikes to a max of 39.7 ms every 5 seconds (lasting 2 seconds)!!

Chrome Beta (Version 39.0.2171.27 beta-m (64-bit))

19.3 - 25.7 ms; the same spikes to a max time of 43.1 ms every 5-6 seconds (lasting 1-2 seconds).

Edit to add - judging by the version numbers, it seems that I should expect very little difference between these tests (QED).

Mmmmmm!!

Actually while Chrome 38 gives me a terrible dt variation (typically 8 - 45ms), Chrome Canary seems a lot better, ranging only 21-24ms. It also subjectively looks better - it's not perfectly smooth in Canary (obviously, since at 45 FPS there's no choice but to skip some frames), but it doesn't have the eye-straining extreme juddering which Chrome 38 has with its wide dt variation. So it may be that it's already been fixed and we only need to wait for Canary to come through (currenty v40).

Can anyone with affected games give them a spin in Canary and say if it seems any better? You can install it side-by-side with normal Chrome: https://www.google.co.uk/intl/en/chrome/browser/canary.html

Canary might have also improved the input latency so you can test that with it too.

BTW in the test I posted earlier the rectangle count is completely irrelevant, it's only aimed at measuring the dt variation.

Canary showed a definite improvement, both ingame and in the throttle test.

There are still some severe issues though - I get a lot of spikes and jerkiness even at relatively high FPS's, and the input lag definitely isn't solved.

As Squidster mentioned, disabling webGL greatly improves performance. I tried it and it helped a lot. Is that an issue with webGL or is it because effects are disabled? I'm using Chrome version 38 with Windows 8.1 32 bit.

Chrome Canary

19.7 - 23.8 ms, no time spike every 5 sec apparent this time.

When I run my embryonic game I see no improvement in the slight jerky performance and occasional hesitations compared to v39 Chrome (even though I no longer create or destroy any sprites during run time...).

sqiddster are you seeing the same issues in Firefox? Ashley's demo ran so much better for me than it does in chrome.

ArcadEd I just tried the game in Firefox, and while Firefox can't run the game nearly as well as Chrome can (at a comparable resolution) it feels perfectly smooth all the way down at 40fps - whereas Chrome feels pretty bad at anything sub-60fps.

Good info, so sounds like it could be a Chrome issue, which means it's in Node Webkit, correct?

Yep.

The question now, is finding a repro for the jerkiness and spikes, which also seem to occur during GPU lag only.

I find that if the frame rate drops in my game, the player is able to pass through collision - has anyone else found this?

The spikes might be the same causes of jank as in the sbperftest demo which is already in a Chrome bug report. I think it would be best to see what they can do to improve that demo and see how that impacts other games.

Tests like sbperftest are essential for bug reports, because it makes objective measurements and is 100% repeatable. That's difficult to get with a full blown interactive game.

sean Noonan , yes, that is something I notice too, and I think that is more likely to happen with the "Jump-thru" behavior and less often with the "Solid" behavior.

FYI my game has just developed a new stutter. Despite saying 120fps every once in a while it does this super stutter thing, and the only thing I'm doing is moving using 8 direction movement with some paralax (all other code is dormant and nothing else is different in the seconds before and after the stutter except a few biggish sprites (500px wide) on one of the paralax layers.

Only happens when I export. :\ Hasn't happened in crosswalk, but html5 export definitely.