AMD SMT ON vs OFF: TLOU test
Started by romano




1 posts in this topic
romano
Member


0
19 posts 2 threads Joined: Apr 2021
06-04-2022, 04:31 PM -
#1
There were some findings about Intel CPU's like 9900k, where disabling HT caused improvement in RPCS3 performance. But I did not find any info regarding AMD's SMT, so I decided to make a quick test on a single game with my 5800X. As I will explain later, results were rather surprising.

For a setup, I had:
  • v0.0.22-13648,
  • spu loop detect. ON, approximate xfloat, thread sched. = "OS"(for now, even though I run Win7),
  • SPU and shader threads at auto, 
  • accurate RSX ON, ZCull at precise,
  • read and write color buff. ON, wake up delay 600,
  • rest should be default.

Test was done with CPU's manual OC at 4.45ghz - all cores fixed.

Now the patches were like this:
   

I ran things 2x to make sure cache load did not affect things much. Benchmark readings were at the scene from beginning where girl wake up from bed after phone call and look at the door. I also tested turning around etc.

Here is the shot with SMT enabled:
   

And here with SMT disabled:
   

Now at this point, ~10.5% improvement seem like a no brainer. WRONG! And this is what screenshots do not tell. 
First, despite CPU being 8 core, gameplay experience with SMT disabled is atrocious. I will repeat it, atrocious! 

Game will experience constant pauses every few seconds that are very noticeable and can last up to whole second. HDD trashing is pronounced and is likely cause of pauses, despite game being run several times in row. I even saw pausing without disk IO. Experience is next to unplayable. I then tried to set preferred SPU's and shaders to 2 just to see quickly - nope only lowered FPS and introduced graphical bugs. I also had crashes without SMT and game felt much more prone to them.

On the other hand, having SMT enabled was a day and night. ZERO pauses, smooth turning around or moving, even with a few FPS less it felt so much smoother and better. 21fps felt just like it could be 30 or more(of course in that room those were at the lower bottom, 25-28fps were common in different spots). Not a single crash(but I ran it only briefly, though without SMT that would be enough).

VERDICT:
So the bottom line is, at least with my HW and OS, experience shows that SMT(and possibly HT) have a huge impact on gameplay experience - in a positive way. FPS loss from it is irrelevant and does not tell the full story. 
Whether I could achieve same quality experience without SMT with more tweaking, maybe different OS,... is possible I guess. But this one proved me that FPS gains are not everything and having CPU with HT/SMT is a huge benefit for RPCS3, both for smoothness and stability.
[email protected], 16GB RAM, GTX1060-6GB, Win7
Ani
Administrator
*******


16
4,264 posts 105 threads Joined: Aug 2017
06-04-2022, 11:57 PM -
#2
Yes, disabled SMT on AMD CPUs hurts performance, it is known
    Desktop: Ryzen 7 5800X,   Radeon RX 6800 XT, 2x8G DDR4 3600MHz, Manjaro Linux
     Laptop: Ryzen 9 5900HX,  Radeon RX 6700M,   2x8G DDR4 3200MHz, Manjaro Linux
Old Desktop: AMD FX-8350,     Radeon R9 280X,    2x4G DDR3 1600MHz, Manjaro Linux


Forum Jump:


Users browsing this thread: 1 Guest(s)