Forza 5 [XO] *large pics inside*

Which would point to a serious bug, but here's a closer look at something like the FORZA logo.



The first two are the duplicate frames, the third is the next frame.

The differences that we see between the first and second frame is clearly the result in the shutter of the video camera also recording some parts of the third frame.


I checked other parts of the video when the timer does show up using the exact same method.
Flawless 60fps on those areas that I could reliably count the clock every frame. No duplicate times whatsoever on the other ~4 or so areas, so it's probably the initial area giving the game problems.

If they ended up being able to give digital foundry a stream to analyze we'd see a pretty good 60 fps lock with some dips in certain areas of the track.

 

That doesn't explain the other areas I outlined, namely the speedometer area -- the blue speed bar and the red line indicator below are different between all 3 frames.

 

I hope this image illustrates what I mean.

The second frame is a blend between the two frames. It shouldn't be surprise that it doesn't look like the first or the third. Some components will look like the first frame, and some will look like the third frame due to how video cameras and shutters work.

Video cameras do not "record black".
The presence of even a brief third frame will cause the portions of the second frame to look like the third frame, but not totally like the third frame.
Notice the similarities of the second frame and the output that I can create from "blending", if you will, the third frame and the first frame together.

Details like the forza logo and the blue roadblocks should be a clear indicator that the second duplicate frame is a frame that captured two independent frames within it's shutter time. (namely the first and the third).

If the second frame is truly a different frame that has been output from the Xbox One, the devs have a much bigger problem on their hands, as in a BUG.

 

no F3 and 4 he said resources to keep FPS and physics refresh as high as it is caused them to not implement. F5 is time constraints

 

I'm fine with the way they evolve Forza over the course of a generation.

 

Darn, 2 pages mostly about our 60 fps videos over at Gamersyde. That's why I like you guys
Anyway, I'm not too sure if you can use our videos for such specific task. We use 5 years old Sony FX7 HDV cameras, recording at 1440x1080i 29.97fps. I then use a deinterlacer (usually yadif, but it can also be tomsmocomp+separatefields) to achieve 60 fps. So while it should have 60 different images in a second, you never what magic could happen in avisynth that could do some thing or another.
I'm currently evaluting a new camera that does true 1080p60, there still will be ghosting since the shutter will never be perfectly sync'd with the tv, but at least I won't have to do anything to the image afterwards

 

Ah. Thanks for that info blimblim, and the videos.

 

Welcome to B3D. I love your webpage...keep on rocking in direct feed

 

Actually I've had an account for years, but sadly I've lost my password, and the email I've used to create it. So, new account it is

 

Good info and great feeds!


I found some of the frames to be free of any ghosting/blending between two frames, leading me to believe that your video was actually shot at 720p instead of interlaced 1080i.

Your video was very clear even at 60FPS (and allowing me to count 55~60 discrete clock readings in a second, without even blending between two frames in many cases) so it led me to believe that this method was possible. Are you sure you shot this at 1080i?

edit: scratch that. If your deinterlacer threw away different fields, then it should be irrelevant if you shot it at 1080i. I found some evidence of this so yes that probably happened.

The three frame I uploaded here does give me a impression that the source wasn't interlaced.
If the source really was interlaced, I probably shouldn't get clear as day readings as each frame was sampled over 33.3 ms while the screen was refreshed at 16.6 ms.


And As I said, I checked other areas where there were also clock readings and applied the same method.
(namely ~36s, ~1m 5s, 1m 30s, 1m 50s) and I found 60 discrete clock readings in 60 frames, indicating that in these areas Forza 5 indeed is achieving 60fps very well.

Here are the numbers in the other areas.
No duplicates as far as I can see. They also are extremely predictable making my life recording them very easy.

37.006 37.023 37.039 37.056 37.073 37.089
37.106 37.123 37.139 37.156 37.173 37.189
37.206 37.223 37.239 37.256 37.273 37.289
37.306 37.323 37.339 37.356 37.373 37.389
37.406 37.423 37.439 37.456 37.473 37.489
37.506 37.523 37.539 37.556 37.573 37.589
37.606 37.623 37.639 37.656 37.673 37.689
37.706 37.723 37.739 37.756 37.773 37.789
37.806 37.823 37.839 37.856 37.873 37.889
37.906 37.923 37.939 37.956 37.973 37.989

1.08.006 1.08.023 1.08.040 1.08.056 1.08.073 1.08.090
1.08.106 1.08.123 1.08.140 1.08.156 1.08.173 1.08.190
1.08.206 1.08.223 1.08.240 1.08.256 1.08.273 1.08.290
1.08.306 1.08.323 1.08.340 1.08.356 1.08.373 1.08.390
1.08.406 1.08.423 1.08.440 1.08.456 1.08.473 1.08.490
1.08.506 1.08.523 1.08.540 1.08.556 1.08.573 1.08.590
1.08.606 1.08.623 1.08.640 1.08.656 1.08.673 1.08.690
1.08.706 1.08.723 1.08.740 1.08.756 1.08.773 1.08.790
1.08.806 1.08.823 1.08.840 1.08.856 1.08.873 1.08.890
1.08.906 1.08.923 1.08.940 1.08.956 1.08.973 1.08.990

1.33.006 1.33.023 1.33.040 1.33.056 1.33.073 1.33.090
1.33.106 1.33.123 1.33.140 1.33.156 1.33.173 1.33.190
1.33.206 1.33.223 1.33.240 1.33.256 1.33.273 1.33.290
1.33.306 1.33.323 1.33.340 1.33.356 1.33.373 1.33.390
1.33.406 1.33.423 1.33.440 1.33.456 1.33.473 1.33.490
1.33.506 1.33.523 1.33.540 1.33.556 1.33.573 1.33.590
1.33.606 1.33.623 1.33.640 1.33.656 1.33.673 1.33.690
1.33.706 1.33.723 1.33.740 1.33.756 1.33.773 1.33.790
1.33.806 1.33.823 1.33.840 1.33.856 1.33.873 1.33.890
1.33.906 1.33.923 1.33.940 1.33.956 1.33.973 1.33.990

1.50.006 1.50.023 1.50.040 1.50.056 1.50.073 1.50.090
1.50.106 1.50.123 1.50.140 1.50.156 1.50.173 1.50.190
1.50.206 1.50.223 1.50.240 1.50.256 1.50.273 1.50.290
1.50.306 1.50.323 1.50.340 1.50.356 1.50.373 1.50.390
1.50.406 1.50.423 1.50.440 1.50.456 1.50.473 1.50.490
1.50.506 1.50.523 1.50.540 1.50.556 1.50.573 1.50.590
1.50.606 1.50.623 1.50.640 1.50.656 1.50.673 1.50.690
1.50.706 1.50.723 1.50.740 1.50.756 1.50.773 1.50.790
1.50.806 1.50.823 1.50.840 1.50.856 1.50.873 1.50.890
1.50.906 1.50.923 1.50.940 1.50.956 1.50.973 1.50.990


It seems that only the first few seconds had any issue hitting 60fps.
Starting area issues/ game just loaded perhaps?

Getting a good 60 fps result probably wouldn't be possible if there was a fundamental error in this method.


Would you mind giving us your 1080i source?
Maybe that would give us even more accurate results.


edit: added more figures from other timeframes.

 

Our cameras can only do 480i or 1080i so there is absolutely no doubt that it's 1080i. Anyway you are in luck since I still have one of our forza 5 raw files around:
http://www.gamersyde.com/leech_30412_1_en.html

 

wow, great. hoping virtualdub works with m2t files.

 

Nope, not directly. Use avisynth with ffms2, then open the avs in virtualdub.

 

I checked the file and it's it's pretty clear we need to deinterlace the source.

On close inspection, it appears that the camera interlaces the video by actually having a 60fps feed,
and interleaving two images to form one image and outputs it in 30fps format.
Which means that every frame from the original source actually contains two frames, one containing 0~16ms, and the other one containing 16~33ms.

Once we isolate them, we can see the original 60 frame per second feed with one minor issue, and that is that it only has half the resolution.
However, as long as the resolution is good enough that we can see the timer readouts clearly and is able to identify them from each other, we should be okay and good to go.



For our purposes, I applied the following

deinterlacing mode: discard fields
field order: double frame rate, top field first
from virtual dub, so we have minimum post-processing.
This way we only get the refreshed lines and discard the "old interlaced frames".

Luckily we also have timed sections on this runthrough too, and the result is pretty nice.

I know there is a loss of quality in this "edited" version but the point is to be able to get a good readout instead of picture quality.
As long as we can read the seconds and differentiate them from each other/identify that they're the same readout, I believe we're good.

I slowed down the frame rate all the way down to 1 fps, so we can see clearly every readout. It looks like the game is running silk smooth 60 fps in this video.

http://youtu.be/8Fkh89o2y3k

readouts are clear, and come at good intervals, and you can read every single time without any issues.



Here's a sample from 6~8 seconds, using the same methods, although I can't control how you deinterlaced them, but the result should be the same nevertheless.
http://youtu.be/YGs3wMdEuHw

There are noticable "stops", if you will, at
6.927, 6.961, 7.061, 7.094, 7.127, 7.244, 7.511, and 7.927 if you slow it down.

On the other hand, if we take the same video, only at 1:07~1:08
http://youtu.be/BnAnZXa0tB0
we find 60fps done pretty nicely and it does show in the video.




I'm not trying to prove that Forza 5 isn't 60 fps or anything like that, but instead this proves Forza 5 to be pretty stable, and in the worst case, close to achieving 60fps from the particular samples we take here.

 

We know for a fact that the Forza 5 build was only 75%~ complete at E3 and that it was still pre-beta. We should be happy that they are pushing 60 fps solid even before they started the beta process.

 

Is this a joke?

I mean really, do you really come to these conclusions without any kind of evidence to support them?

Forza 5 has a brand new physics engine, graphical engine, lighting engine, 3D laser scanned tracks, every car at autovista level details and running on 60 fps at 1080p on brand new hardware within two years of development.

Just a prettier Forza 4? Yeah right....

 

7.127 - 7.094 = 33ms = 2VBL = drop frame

 

Strangely the clock progresses only 16ms after some of the other apparent frame drops, looking at those numbers from strange on the previous page.

 

Doesn't or didn't AMD 7000 series gpu have or had a microstuttering problem?

Or is this unrelated?

 

yes it's pretty curious. Sometimes the jump in comes the next frame after the duplicate, but sometimes it jumps after two frames or more. Maybe it's related to how the code schedules the pipeline when it can't lock onto 60 fps?