Some GPU tests, or how to lose money buying a dual GPU card for A3

[VanZant 48]

It's a well known fact that performance with A3 doesn't scale very well. It has been so since the beginnings of time.

But how much?, I already thought that the relationship wouldn't be proportional, but today after a bunch of tests the conclusion was funnier than expected. Using SLI or dual GPU cards is mainly a waste of money and electrical power. Of course talking only about A3.

With dual GPUs not only you get an annoying PIP flickering, more stuttering, more frame lags, and even a more noticeable object popping. The more stress, the more useless is the second GPU until a point where the fps are exactly the same, literally.

This is the average resulting graph. I wasn't expecting this.

Core i7 2600k @4,5 Ghz

GTX 590 3GB

32GB DDR-1600

2x SSD Sata 3 RAID0

Since today my settings for A3 in the Nvidia control panel are single GPU. Some fps are lost above a certain threshold, but the benefits (PIP, stuttering, ...) are worth those fps.

[fumanwho 10]

It's a well known fact that performance with A3 doesn't scale very well. It has been so since the beginnings of time.

But how much?, I already thought that the relationship wouldn't be proportional, but today after a bunch of tests the conclusion was funnier than expected. Using SLI or dual GPU cards is mainly a waste of money and electrical power. Of course talking only about A3.

With dual GPUs not only you get an annoying PIP flickering, more stuttering, more frame lags, and even a more noticeable object popping. The more stress, the more useless is the second GPU until a point where the fps are exactly the same, literally.

This is the average resulting graph. I wasn't expecting this.

Core i7 2600k @4,5 Ghz

GTX 590 3GB

32GB DDR-1600

2x SSD Sata 3 RAID0

http://i.imgur.com/ndMuIk7.jpg

Since today my settings for A3 in the Nvidia control panel are single GPU. Some fps are lost above a certain threshold, but the benefits (PIP, stuttering, ...) are worth those fps.

This test doesnt tell us anything about how well arma 3 handles sli as the ai is calculated on the cpu, you have to use only settings that affect the gpu exclusively. Also there must be something wrong with your data collection script did you use some kind of averageing. Because its impossible to get straight lines, you need to check your script.

---------- Post added at 20:21 ---------- Previous post was at 20:20 ----------

woops didnt see average comment, could you post the raw data

[Dwarden 1125]

afaik the PIP flicker is because of incorrect profile / detection of arma 3 binary filename, may i ask why driver version of NVIDIA drivers you are using ?

the PIP isses should be gone for SLI if the driver detects proplerly arma3.exe (instead of obsolete arma3_dx11.exe or w/e was the very old filename (sry forgot already)))

[VanZant 48]

afaik the PIP flicker is because of incorrect profile / detection of arma 3 binary filename, may i ask why driver version of NVIDIA drivers you are using ?

the PIP isses should be gone for SLI if the driver detects proplerly arma3.exe (instead of obsolete arma3_dx11.exe or w/e was the very old filename (sry forgot already)))

327.23. I'm more focused on modding things and i don't check the vehicles very often, but as far as i know PIP has been flickering always.

This test doesnt tell us anything about how well arma 3 handles sli as the ai is calculated on the cpu, you have to use only settings that affect the gpu exclusively. Also there must be something wrong with your data collection script did you use some kind of averageing. Because its impossible to get straight lines, you need to check your script.

---------- Post added at 20:21 ---------- Previous post was at 20:20 ----------

woops didnt see average comment, could you post the raw data

The data source is variable and abstract. The average is what matters in this case, and at some point more than one GPU is completely useless. Plus the mentioned glitches.

Of course this is based on my hardware only.

Edited October 16, 2013 by VanZant

[fumanwho 10]

327.23. I'm more focused on modding things and i don't check the vehicles very often, but as far as i know PIP has been flickering always.

http://i.imgur.com/DrcvG3s.jpg

The data source is variable and abstract. The average is what matters in this case, and at some point more than one GPU is completely useless. Plus the mentioned glitches.

Of course this is based on my hardware only.

This does not measure gpu usage as you are stressing your cpu. You need to test against things that affect the gpu exclusively. I am pretty sure you know this already but i will just mention it anyways. The gpu is used for drawing. It is not used for ai.