FCAT: The Evolution of Frame Interval Benchmarking

[Jaydip]

120Hz monitor should showcase lees tear because very few games can maintain > 120FPS at decent settings @1080P.

 

[3DVagabond]

There obviously is some latency induced by it. Though all it does it make a fast frame slowed down to be closer to the frame length of the prior frame, so it shouldn't induce a lot of latency, but we'll have to wait and see.

Do you know whether it actually adjusts frame by frame, as you describe? Or, does it just take an overall avg. and space the frames that way? I've never seen where that is explained. If you have, could you please point me to it?

 

[bystander36]

Do you know whether it actually adjusts frame by frame, as you describe? Or, does it just take an overall avg. and space the frames that way? I've never seen where that is explained. If you have, could you please point me to it?

It has never been explained, only that it is frame metering to make the average frame difference consistent, and it seems to be flexible enough to adjust to your current FPS, so it has to adjust based on small frame samples, or it would have a huge problem when FPS shifts.

The Fraps latency numbers also show that it is comparable to Crossfire still, so it has to adjust based on small sample sizes, if not based on the previous frame, or large amounts of latency would be shown in the fraps data.

 

[3DVagabond]

It has never been explained, only that it is frame metering to make the average frame difference consistent, and it seems to be flexible enough to adjust to your current FPS, so it has to adjust based on small frame samples, or it would have a huge problem when FPS shifts.

The Fraps latency numbers also show that it is comparable to Crossfire still, so it has to adjust based on small sample sizes, if not based on the previous frame, or large amounts of latency would be shown in the fraps data.

FRAPS measures latency?

 

[willomz]

If you had FRAPS data and FCAT data then you could measure the amount of time spent in the pipeline and thus spot any delays.

The FRAPS data suggests that work enters the pipeline at a regular rate and FCAT suggests that it leaves the pipeline at a regular rate also.

 

[3DVagabond]

If you had FRAPS data and FCAT data then you could measure the amount of time spent in the pipeline and thus spot any delays.

The FRAPS data suggests that work enters the pipeline at a regular rate and FCAT suggests that it leaves the pipeline at a regular rate also.

From what I can make out 2 statements have been made that we really don't know if they are accurate or not. We don't know how nVidia's frame metering works, frame by frame or not, and we don't know if it adds latency.

Edit: Sorry, these are 2 people making these assumptions. Not just willomz. I edited the post.

 

[willomz]

All I'm proposing is that someone with the data could investigate, that's why I used the word 'if'.

However given smooth, regular input and smooth, regular output it would 'suggest' that what is happening inbetween is also regular.

If you have alternate short frames and long frames then you would need to delay the short frames and thus make all the frames long in order to have smooth output. However if you can get everything to line up from the beginning then you don't need to introduce any delays.

Thus the FRAPS data, the Present() data IS important. If the Presents happened at irregular times and then the actual frames displayed came out regularly then THAT would imply that an artificial delay was occurring.

 

[3DVagabond]

All I'm proposing is that someone with the data could investigate, that's why I used the word 'if'.

However given smooth, regular input and smooth, regular output it would 'suggest' that what is happening inbetween is also regular.

If you have alternate short frames and long frames then you would need to delay the short frames and thus make all the frames long in order to have smooth output. However if you can get everything to line up from the beginning then you don't need to introduce any delays.

Thus the FRAPS data, the Present() data IS important. If the Presents happened at irregular times and then the actual frames displayed came out regularly then THAT would imply that an artificial delay was occurring.

I'm not saying anything about there being more or less latency. Just that it was stated that nvidia's frame metering works on a per frame basis, when we don't know that, and that FRAPS measures latency, which it doesn't. These were put forward to show that nVidia's frame metering doesn't add latency (or very little at least), when in fact they don't. They are basically made up.

While I understand what you are saying, it doesn't show the amt. of latency nVidia's frame metering adds either. Since you can't defeat the frame metering, I don't know how we can measure it's effects on latency.

AMD has said they never did frame metering because it adds latency. They are going to go as far as making it optional for those who want minimum latency. This doesn't speak to nVidia's approach though either. We don't know so people just choose to believe what they want to.

 

[GaiaHunter]

I'm not sure if we all are talking about the same latency.

I was talking about input latency, which if I'm not mistaken was cited as a reason to not use vsync.

 

[bystander36]

I'm not saying anything about there being more or less latency. Just that it was stated that nvidia's frame metering works on a per frame basis, when we don't know that, and that FRAPS measures latency, which it doesn't. These were put forward to show that nVidia's frame metering doesn't add latency (or very little at least), when in fact they don't. They are basically made up.

While I understand what you are saying, it doesn't show the amt. of latency nVidia's frame metering adds either. Since you can't defeat the frame metering, I don't know how we can measure it's effects on latency.

AMD has said they never did frame metering because it adds latency. They are going to go as far as making it optional for those who want minimum latency. This doesn't speak to nVidia's approach though either. We don't know so people just choose to believe what they want to.

I wasn't suggesting an exact method, only that if smoothing was done, the result would be a short frame would be altered to match closer to a previous longer frame.

And if smoothing is done, at least some latency has to be introduced. However, that amount is only going to be added to frames that were were unusually fast.

 

[BTRY B 529th FA BN]

Excuse me. Sorry for interrupting, but does anyone know where I can download FCAT from?

 

[bystander36]

Excuse me. Sorry for interrupting, but does anyone know where I can download FCAT from?

FCAT is a special card for gathering frames output from a video card. It is not something you can buy, as far as I know, and the software that goes with it, is to help gather information about those frames. Again, not something you can buy.