もう地獄には余地はない – PropCombine Performance Test Results

This is an investigation into prop-combining for nmrih mapmaking, and its effects on fps. Combining props may result in a performance increase, and this is interesting to investigate. Especially seeing so many nmrih players have potato pcs lol.

Defining the problem

Here are some results concerning props and propcombining, and performance. I thought people here might be interested.

As many of you mappers probably already know: fps in the source is heavily influenced by prop drawcalls. There is a drawcall for every prop and each of its materials. In many cases, performance may be boosted by reducing the number of props visible. One method for achieving this is to combine props; propcombine.

The question arises: can we combine props for nmrih, and get fps gains doing so?

Test method

We will be using TeamSpen’s propcombine postcompiler to combine props. The bsp format in nmrih is not natively supported, so we have to use another source game’s hammer as proxy for combining purposes.

• Link here.

After defining the areas to combine props in, and running vbsp+bspcombine, I reduced the prop count in an area by half. Copy pasting the combined props from the proxy game’s map to the working nmrih map gets us two versions of the same map: one with separate props and one with combined props.

結果

The results are interesting and, personally unexpected: I expected fps to slightly improve but the opposite was found. a ~15% fps DECREASE was found! (in simulated potato mode, in insertion2.

Above is the original map’s fps in potato mode, 44fps.

Here’s the map fps after propcombine, 38fps.

Conclusion and discussion

What we can learn from this is that combining random props without thought does not necessarily increase performance.

I do know it is possible to massively boost performance by combining props. 例えば, a boost of at least triple fps was achieved in the carpark, and in the traintunnel, in the map brooklyn_2.

What constitutes a good candidate for combining, vs a bad one? I am unsure at this moment. Perhaps the combined ratio of 2->1 in insertion vs ~ 10->1 in brooklyn_2 has an effect: a threshold of the number of props combined after which gains are had, but if it is fewer performance is worse.