AFAIK that is impossible. DOS games access the video memory directly and every video mode works differenty. If the software uses page flipping, you could try intercepting the functions (though that would be more complicated if the BIOS is skipped and the function is used directly), but if it draws directly to the visible page or through double buffering I don't thing there is any reliable way to calculate the FPS count.

It works this way:
You capture the output at the display rate that is 70 fps without frame drops. Then you analyse the difference of subsequent frames. Whenever the frame changes there is a spike in the difference compared to the analog noise background. The rate of those spikes are your fps.

Unless the game does VSYNC this method could interpret half-drawn frames (during tearing and such) as full ones, making this very innacurate. Also, in scenes where there is little movement it might be a bit difficult to tell what is a frame and what is pure noise. It is also interesting to note that this method can`t measure FPS higher than the refresh rate (wich is necessary for performance comparisons between graphics cards, etc.). And last, this would require another machine. Clumsy and innacurate, but to be fair this is probably the best thing possible.

Actually with a bit more analysis this can be done even for source material where vsync is off. I wanted to sketch just the basic principle.
I did this already with a capture of screamer rally for vetz using MSI on a Mystique and it worked out nicely.
This shows the few lines of code as well as the difference graph and fps graph.
ftp://78.46.141.148/pictures/vetz/fps_graph.png
Here the difference is zoomed in to see the analog noise line:
ftp://78.46.141.148/pictures/vetz/fps_graph2.png

Of course it is not suited for actual benchmarking where "hidden" frames also play a role. But this was not the question here.

That is a game i am very interested in. What frames did you get and which hardware did you use for screamer rally?

As the goal was just to get a capture using the MSI render interface I used my default benchtable. It is a P3 800 MHz on a Gigabyte GA-6BXU. As graphics card I tried a Matrox Mystique and Matrox Mystique 220. Both gave the exact same result as expected.
You can find the capture here: ftp://78.46.141.148/videos/screamerrally.mkv
This is hardware accelerated in DOS using Matroxs MSI render interface. The Mystique does not support bilinear filtering.

From the already linked fps graph you can see that the frame rate is between 30 and 40 fps, sometimes reaching 50 fps.
The actual recommended requirement is a Pentium 133 MHz.

I hoped there was a software solution. Seems as I don't have the right hardware to perform this task 😉 But thanks anyway.

so, no glide unfortunately.

40-50 fps is not a lot, considering the very fast cpu. It ''feels'' a lot faster on a voodoo3 and p3 700 but i am not sure, that's why i was interested 😀

Thanks!

Of course a Voodoo3 (1999) is a lot faster than a Matrox Mystique (1996) in the GPU limit.

yes, i know.

Just wondering which 3dfx card it would take to get the game at a steady 60fps and which cpu aproximately.

meljor wrote:

Just wondering which 3dfx card it would take to get the game at a steady 60fps and which cpu aproximately.

Not much, a Pentium 166-200 and a Voodoo1 and you'll get well above 30fps.

Links are not dead, is somewhere more info about this topic?

For datapath capture cards, vcs has the functionality elianda describes. Also elimination of tearing, though the other issues mentioned by alexanrs apply. It's also open source.

vvbee wrote:

For datapath capture cards, vcs has the functionality elianda describes. Also elimination of tearing, though the other issues mentioned by alexanrs apply. It's also open source.

So you want to measure DOS FPS you have to use Datapath and Epiphan cards only?

Is somewhere some example of output? I mean could he results back rendered to original picture as overlay to show some realtime Fraps like stats along with original picture? or its just text files, is there some framerate curve in the time etc?

For vcs, the code to display that information is there: https://github.com/leikareipa/vcs/blob/6e038a … filter.cpp#L184. How it ends up looking is up to you, but by default it's in the fraps format of yellow text at the top of the screen.

ruthan wrote:

vvbee wrote:

For datapath capture cards, vcs has the functionality elianda describes. Also elimination of tearing, though the other issues mentioned by alexanrs apply. It's also open source.

So you want to measure DOS FPS you have to use Datapath and Epiphan cards only?

My simple approach used an already captured stream for analysis and is not dependent on what kind of capture card you use.

vvbee wrote:

For vcs, the code to display that information is there: https://github.com/leikareipa/vcs/blob/6e038a … filter.cpp#L184. How it ends up looking is up to you, but by default it's in the fraps format of yellow text at the top of the screen.

I already was there more than once, but its too low level / incomplete for me, at the top of description are mentioned Datapath cards, so i through that is only for them and the nothing like final output sample video, some step by step process how to capture/evaluate framerate / detailed features description (it would be good motivation to get some new capture card) etc.. And at linked page - http://tarpeeksihyvaesoft.com/soft/ is again mentioned that it is for Datapath.
Its nice that there is mentioned how to build it or that it runs in virtual machine in Linux, but from whole 1 screenshot is clear that there are lots of other feature, btw does it run it Vmware / Virtualbox?

1My simple approach used an already captured stream for analysis and is not dependent on what kind of capture card you use.

That nice, because i though that it some fancy capture cards club thigh only, i know that in previous post is outlined that process in highlevel, maybe its enough for skilled guys, but i would new some step by step tutorial which tool to use and on which button click.. some screenshots, or how to video and example of outputs.. Because to be honest, from information so far i didnt know if out is just same csv text files with numbers or some screenshot with graphs, or video with overlay etc.

ruthan wrote:

I already was there more than once, but its too low level / incomplete for me, at the top of description are mentioned Datapath cards, so i through that is only for them and the nothing like final output sample video, some step by step process how to capture/evaluate framerate / detailed features description (it would be good motivation to get some new capture card) etc.. And at linked page - http://tarpeeksihyvaesoft.com/soft/ is again mentioned that it is for Datapath.
Its nice that there is mentioned how to build it or that it runs in virtual machine in Linux, but from whole 1 screenshot is clear that there are lots of other feature, btw does it run it Vmware / Virtualbox?.

Good documentation is great to have, but also takes considerable effort to create and maintain. I'd like to see wider adoption of vcs to justify the work on that front.

My experience is that the visionrgb hardware won't work in virtualbox, but that said, pci passthrough in virtualbox is experimental and possibly limited to linux hosts. Haven't tried vmware.

vvbee wrote:

Good documentation is great to have, but also takes considerable effort to create and maintain. I'd like to see wider adoption of vcs to justify the work on that front.

I would say that is chicken / egg problem.. you dont need of walls of text, but in my eyes most important for users is really to see some footage example, clarify which cards are and arent supported.. you should add more bold claims about features (for example that is only existing DOS FPS capturing automated tool etc).. and average users not really care how to build it from source code, when is binary already available.

BTW is mighty Phil, using your solution for his Dos framerate measurement? Because if he does it would be great for advertising..