Investigating Warframe Performance (Cpu Frequency/core)

[qeuu]

Purpose:

To determine the effects of CPU frequency and thread count on Warframe's framerate.

 

Method:

The tests will be performed in the simulacrum by spawning 20 juggernaut behemoths and fighting them with a hysteria Valkyr. Each test will be done for roughly 1 minute, then will be ended by returning to the console within the simulacrum.

The tests will be run at varying CPU frequencies, and varying core/thread counts(given to Warframe)

The tests will be run on a computer with:

i7 5820k @1.2GHz, 2.4GHz, 3.6GHz and 4.3GHz

Nvidia gtx 770 2GB

16 GB DDR4 2400MHz RAM

Tests will be run at 1920X1080 in borderless windowed mode, max settings with PhysX, tessellation, depth of field, and motion blur turned off. 

Tests will be run with multi-threaded rendering on.

 

Results:

Data Tables only:

 

Data Tables and Graphs:

The tests can be seen as the drops in frame rate, and are separated by periods on very high frame rate or no frame rate. Tests are in the order they appear in the tables.

 

Discussion:

So a few notes about the data. I did use MSI afterburner to record the frame rates during the tests. Although I was unable to get accurate average frame rates from the data, so the average frame rate was estimated. Also the scenario that was created seemed to be a lot more intense than what is normally encountered in gameplay.

 

So, looking at the data we can see that in general higher frequency results in higher maximum frame rates, but doesn't seem to always higher average frame rates. This could be due to cores throttling to lower speeds since the CPU is not being utilized enough. This would lengthen the response time between frames.

 

Regarding core/thread count, it seems that for the test any of the dual core and single core tests hanged when killing all the remaining enemies by returning to the spawn terminal. Although frequency doesn't seem to have much an effect on this.

 

Regarding core count, from what I can tell having more cores increases maximum frame rate, but adding hyper threads reduces maximum frame rate while making the frame rate more stable.

 

Regarding maximum frame rate outside of testing. This was the max frame rate when there were no enemies in the simulacrum. So for every configuration except for one core at 1.2 GHz resulted in the GPU bottlenecking the frame rate.

 

There seem to be anomalies in the data. Mainly regarding the frame rates with all CPU cores and threads (the default configuration) as they seem to be lower than the configuration with 4 core and 4 threads. Although this could also be explained by the CPU throttling to save power.

 

Conclusion:

Based on the results obtained it seems that having more cores increases frame rate but having more than 4 cores given to Warframe doesn't seem improve frame rate. Giving Warframe hyper threads makes the frame rate more stable although this also reduces average frame rate.

_____________________________________________________________________________________________

 

If I have more time I will go back and try and get calculated average frame rates for the tests, as I did save the  logs from most of the tests.

 

For anyone wondering why I decided to do this, it is a continuation of: Testing multi threaded rendering and thread count

 

So does anyone have any comments to make about this?

[Magnulast]

There are a few questions regarding this manner:

 

A: Did you optimize the cache?

 

B: Was this run on a Raid(2x/3x SDD), SDD, SSHD, or HDD?(label proper speed via rpm)

 

C: Was there any background programs running that uses in-game information directly?

 

D: What Nivida driver are you using? What OS are you using?

 

Thanks for the information, it helps me understand where, some of the issues may lie with my CPU...when using some in-game recording  programs.

[Abenoki]

Awesome, so this turned out exactly as I had hoped! This illustrates the CPU limitation of warframe pretty well, it also shows that its threaded a bit better than i expected which is nice.

 

Ideally we see here that Warframe for 60fps consistent would like 4 cores and clock speeds of 4ghz+ (in regards to intel at least) so going from an i5 to an i7 isnt going to give you much in the case of warframe vs overclocking your current quad (assuming its a recent quad and not 10yrs old) to increase you min/max and avg FPS. CPU under 2ghz really arent going to like this game, it can be playable but 2ghz+ min would be nice. 

 

Those gunning for 120fps youll want a decent GPU combo with i5 (or higher) @ 4.5ghz+ 

 

AMD users keep in mind youll have to have around 500mhz over an equivalent intel CPU to reach similar performance levels most likely so an i5 @ 4.5ghz+ for pegging around 120fps would require an AMD CPU to hit 5ghz or more. AMD dual cores will struggle in warframe which i can attest to from personal experience with a friend's dual core machine and a gtx660.

 

Thank you for doing this and also letting me know that you did and Id like to see more data if you decide to do more testing.

Edited November 28, 2015 by Echoa

[qeuu]

There are a few questions regarding this manner:

 

A: Did you optimize the cache?

 

B: Was this run on a Raid(2x/3x SDD), SDD, SSHD, or HDD?(label proper speed via rpm)

 

C: Was there any background programs running that uses in-game information directly?

 

D: What Nivida driver are you using? What OS are you using?

 

Thanks for the information, it helps me understand where, some of the issues may lie with my CPU...when using some in-game recording  programs.

 

A: Not right before the test, but it was optimized recently

 

B: I ran this off on my 500GB samsung 840 EVO.

 

C:I was running MSI afterburner to log my frame rate and GPU usage. Also it was running through steam, even though i use the non-steam version of warframe, so the steam overlay was active.

 

D:I was using version 355.98. I'm running Windows 7 Professional (64 bit)