Posts: 1.821
Threads: 13
Joined: Mar 2011
Reputation:
55
Location: Florida, USA
Got your PM. Will get started now.
OS: Linux Mint 17.2 64 bit (occasional Antergos/Arch user)
(I am no longer a Windows user)
CPU: Intel Pentium G3258
GPU: Nvidia GTX 650 Ti
Posts: 1.821
Threads: 13
Joined: Mar 2011
Reputation:
55
Location: Florida, USA
03-07-2015, 05:53 PM
(This post was last modified: 03-07-2015, 05:55 PM by DaTankAC.)
Dragon Quest VIII on my G3258.
extra SW rendering threads : avg fps
0: 41.1
1: 39.9
2: 38.8
3: 37.5
and same results down to 16. No real slowdowns with lots of extra threads. But of course, I'm running dual-core so results may not be most accurate.
These results were based on WAIT_ON_GS_STILL_CPU_INTENSIVE
OS: Linux Mint 17.2 64 bit (occasional Antergos/Arch user)
(I am no longer a Windows user)
CPU: Intel Pentium G3258
GPU: Nvidia GTX 650 Ti
Posts: 1.821
Threads: 13
Joined: Mar 2011
Reputation:
55
Location: Florida, USA
With default file of GSThread_CXX11.h I left the #define FULL_WAIT_LESS_CPU_INTENSIVE alone.
0: 42.8
1: 40.8
2: 40.2
3: 40.3
4: 39.9
16: 38.2
In both tests, MTVU decreases my fps by around 5%.
To add, with my former cpu (FX4300), MTVU still caused a loss of fps. Unlike when I used to be a Windows user, where MTVU was a good gain in fps, about 20% on EE limited games.
OS: Linux Mint 17.2 64 bit (occasional Antergos/Arch user)
(I am no longer a Windows user)
CPU: Intel Pentium G3258
GPU: Nvidia GTX 650 Ti
Posts: 6.069
Threads: 68
Joined: May 2010
Reputation:
167
Location: Grenoble, France
Thanks for the data. I was expecting worse behavior.
2C is clearly not enough for extra thread on SW renderer.
I need to check MTVU. The code replaces the n ring buffers between MTGS and extra threads. I'm curious of the impact to replace the 2 MTVU/MTGS-core ring buffer.
I need more data for 4c/6c cpu