Video Options effect on performance

[pyj99]

Hi everyone,

I just wanted to share the results of some extensive tests I did on what impact has each Video Options effect on performance.

My Goal

My goal was twofold:

1. To make up my opinion wether the CPU or the GPU is the bottleneck in RTW
2. To achieve the best trade-off between visual effects and performance

My Test Setup
I read from some reviews that benchmark were difficult to make with RTW as the computer decision may change from one test run to the other. I decided to take away this problem by using only the scripted battle introduction in the prologue campaign which lasts ~2 min.

As it is scripted I can have a coherent test run. The tests are incomplete however as this battle intro doesn't use effects, the units size are fixed and the animations are always desynchronized.

My system is an Inspiron 9100, P4 3.0 GHz, 512 Mb of Ram, Mobility Radeon 9800 256 MB
I recorded the framerate per seconds with fraps

I run the game at LCD native resolution of 1680*1200/32

Test Results

I started my test with all All options on Highest except Units on High and
All boxes activated.
Then I decreased the options one at a time leaving the others on highest.


Min Max Avg Delta
Base situation 24 44 33.723
All options on Highest except Units on High
All boxes activated

AA Low 22 44 33.524 -1%
AA None 22 45 33.12 -2%
Terrain Medium 24 43 33.775 0%
Building Medium 27 52 37.267 11%
Building Lowest 27 50 37.364 11%
Grass Low 25 49 36.754 9%
Grass None 25 49 37.408 11%
Unit Medium 23 43 33.728 0%
Units Low 23 47 35.064 4%
Effects Medium 23 43 33.099 -2%
Effects Low 21 44 33.304 -1%
Vegetation Medium 22 43 33.359 -1%
Vegetation Low 22 44 33.681 0%
No Detailed Unit Shadows 22 45 34.919 4%
No Unit Shadows 23 48 37.584 11%
No Unit Shaders 21 43 31.366 -7%
No Glints 20 45 33.245 -1%
No Smoke/dust 22 42 33.343 -1%

It is very interesting to see that Building, Grass and Unit shadows have the most impact on performance while to my tastes Buildungs and grass visual effects are not very impressive.

It is also interesting to show that AA has no noticeable effect on performance in this setting.

I then wanted to test with all options on highest including units details

Units on Highest 16 44 27.43 -19%

You see there a significant dip in performance in particular in the minimum frame rate. This is likely because of the fact that on highest RTW uses only 3D polygons for units while at lower settings it uses 2D cards when viewed from afar.

I then wanted to check were was the bottleneck CPU or GPU and I overclocked my graphic card by ~20% to see if there was an impact on performance

All options on highest (even units) 16 48 30.97 +13%
All options on highest (units on high) 25 49 38.22 +13%

As you can see the performance gains were far less than expected (I gain ~20% when overclocking on 3Dmark2k5). In particular there was no real impact on minimum framerate.


Preliminary Conclusions

Some video options bring only small visual improvements while having a huge impact on performance

If you have a MR9800 (~ Ati radeon 9800 pro for desktops) It seems that RTW is in fact much more CPU limited than GPU.

I wil continue to investigate the matter when talking about unit sizes

bye for now,

PYJ99

[Nelson]

Nice work, pyj!

What was the unit size you used? I wonder if 1024x768 would bring a noticeable improvement. I never run at 1680x1200.

I have always suspected that the TW games have been CPU hogs. So many AI decisions for each soldier are a real load. I have a 9800 pro/128m with and an AMD 2800 XP cpu and I have left the shadows off for a while. I will not be able to run Rome with huge units and all the graphics turned up until I upgrade my cpu and my video. Not anytime soon I’m afraid. In fact, I'm not sure what it would take to max everything at this point on the tech curve and still have a lowest rate in the 30+ range. It is, after all, the lows that make or break the experience.

[pyj99]

@ Nelson
Thanks

In the scripted battle of the prologue, the unit sizes are fixed (I think to normal 40 hastati inb a unit)

From memory; the number of men in the battle is ~1100-1200

100 for Julii
400 for Senate
600 for Barbarians


Test results on Unit Size

NB the following test resukts are not comparable with the previous ones as they come from a custom battle.

The purpose was to test the effect of Unit Size on performance

Custom Battle Rome vs Gauls 20 units vs 20 units All Highest excluding Unit details on high
Frames Time (ms) Min Max Avg Delta
Huge 3577 206090 10 23 17.356
Large 2969 146121 11 24 20.319 17%
Normal 3786 174112 12 27 21.745 25%
Small 7473 323438 17 27 23.105 33%
Desynch 5166 280295 12 23 18.431 6%
Overclock7571 388201 10 26 19.503 12%
GPU


As you can see the effect of diminishing unit size is much less important than expected.
In particular, the impact on the minimum framerate is negligible...

Maybe the impact of the number of units there is on the battlefield is more important than the number of men
I will have to test that.

It is however interesting to see that on Huge number of units the GPU is beginning to make a small difference on the average fps.
However it has no impact whatsoever on the minimum framerate.

So if you want to play with options max out on huge, it is useful to have a better GPU than the MR9800.
However it would not be sufficient alone to guarantee you a huge fps.


PYJ