Help me improve my VR performance.

[Sandman1330]

I'm having the same issue as Spectre.

 

 

I'm running a FX-6300 and GTX 1060, so I'm obviously going to run into a CPU bottleneck.

 

 

However, MSI Afterburner shows CPU usage on each core to be typically around 50%. No cores are maxed out, and the GPU is only running 50-60%. FPS in the rift vary from 90 up high, 45 on average, and as low as 15 down low near towns, etc. (All settings on lowest)

 

 

My gut feel here is that, while only really utilizing 2 or 3 cores, the load is jumping from core to core. I think it's jumping around so fast that it isn't registering the core at 100%, since each core is really only getting short spikes at max before shifting the load to another. Perhaps this is a method DCS/Windows/AMD uses to distribute heat and load?

 

 

I have tried setting affinity to only 3 cores, but the difference was negligible. Now that I think of it, I'm going to try it again and see where the usages fall out...

Edited March 31, 2017 by Sandman1330

[Sandman1330]

So, the results:

 

 

I set myself up in the Huey, in the free flight quickstart mission. Stayed on the ground (this is usually where I have the worst FPS).

 

 

With all 6 cores engaged, I saw around 22FPS and fairly even load across all cores - 50-70%.

 

 

I then parked one core per module (cores 1, 3 and 5 were parked) in order to see if removing the problem of shared cache, etc within modules would make a difference. Core utilization went up to the 70s for cores 0, 2 and 4.

 

 

I then disabled core 4 to see what happens with only 2 cores - now at 80-90% for both core 0 and 2. There was no significant drop in FPS up to this point (indicative of DCS' use of only 2 cores).

 

 

Finally, I disabled core 2, running only on Core 0 - utilization not surprisingly finally went near 100, and there was an FPS drop - from 22 down to 15-18.

 

 

Interestingly, after re-enabling all the cores, the load seemed to stay on cores 0, 2 and 4 (contrary to before disabling them, where it was spread throughout). It would make sense to push most of the load to these cores, due to the shared module architecture of the FX, but it didn't do it until after I'd fiddled with affinity. FPS was unchanged anyway.

 

 

So, after this (brief) bit of testing, I'm left to conclude that something, somewhere, is spreading the load between all the cores. The main thread for DCS seems to jump rapidly between cores, such that no core actually shows 100% but rather it is evenly distributed.

 

 

This may only apply to my specific hardware config (AMD FX, Windows 10), but it could also explain why Spectre is not seeing 100% on one core, even though he suspects a CPU bottleneck.

 

 

If someone else has done more extensive testing or has a better theory, I'm all ears!

Edited March 31, 2017 by Sandman1330