P3D multicore usage anomoly

[itsjase 52]

48 minutes ago, Bert Pieke said:

OR... what I do, is to use the Scanline X/2 mode in RTSS to effectively lock at 30 fps.

Not using the RTSS fps limiter though..

I believe that should give identical results because P3D isn't true fullscreen.

Do you also notice one core at 100% even when your frames are locked at 30?

[Rogen 1,087]

Vsyncing @ 30 Hz will reduce load, fairly signiticantly for the GPU, not so significantly for the CPU and that's just because of the way the ESP engine works.

It's all because a 30Hz vsync = 30 FPS (in P3D it's a wobbly 30 FPS though) and it's less work for the GPU to maintain 30 FPS than 60 FPS.

RTSS use is described pretty well in the linked thread.

Hopefully LM will be able to improve FPS performance in coming versions athough P3D is quite a bit ahead of where FSX was.

I took a flight from 11S in FSX-SE just to see waht it was like, OOM in no time and low FPS (although in FSX they would be considered OK) made me realise how far P3D has come, but then again so have the othere sims improved.

Cheers

[Bert Pieke 7,585]

13 minutes ago, itsjase said:

I believe that should give identical results because P3D isn't true fullscreen.

Do you also notice one core at 100% even when your frames are locked at 30?

As I stated above, the CPU usage does NOT drop when I use RTSS Scanline X/2, but it DOES drop if I turn on Vsync in P3D.   BTW, CPU usage also drops if I lock fps in NVI at 30...

In all three cases, fps end up at roughly 30 fps.

In summary, if I wanted lower CPU usage, there are at least two ways to accomplish that.. but my goal is smooth flight, and RTSS is the leading contender as far as I can tell. 

Edited April 19, 2019 by Bert Pieke

[Noel 4,501]

1 hour ago, Bert Pieke said:

In summary, if I wanted lower CPU usage, there are at least two ways to accomplish that.. but my goal is smooth flight, and RTSS is the leading contender as far as I can tell. 

I told myself I'd never waste time again fiddling w/ settings as I had reached the promised land of PERFECTLY SMOOTH, macro and micro stutter-free video performance about 2y ago now on my 6 y/o 3930K.  This thread had me testing out multiple possible scenarios using Sled's discovery.  In non-HT mode his method is remarkable.  I set up the same flight in the Phenom out of FTX KPSP in FTX SoCal.  I followed Sled's trick and indeed all six cores were way lower in their % utilization, and fairly balanced.  What was bizarre was seeing 6 cores hovering around 30-45% each.   Typically in that scenario for my normal HT enabled config the main thread will be around 80% and texture loaders vary w/ load while taxiing and flying.   I decided to combine Sled's setup with 30Hz monitor vsync.  Video quality was quite good, especially considering the low average CPU utilization.  Flat out remarkable--the low CPU use.  Much lower temps as mentioned previously.  But...video quality WAS NOT at all PERFECTLY SMOOTH!   Oh, it was pretty good for sure...until you really looked.   So i surmised, what if I can resume HT enabled, and do the Sled trick, AND do a hybrid approach again using monitor at 30Hz, vsynced.  I thought wow I might be able to get PERFECT SMOOTHNESS once again, but with much lower heat generated and w/ sufficient frame performance.   Yeah!  I tried, by following what i thought was the same routine of turning off and on P3D's unmasked load, but it would not do the balanced amongst 12 LPs as it well did w/ 6 PP's no matter what I tried!   I don't believe whatever causes this balanced appearance happens w/ HT on.  What I saw was 6 fairly balanced LPs, and 6 very unbalanced LPs, in fact some were at 0%.  

In any case, resuming my proven setup restored perfectly smooth video.   For ages, since the release of FSX and then P3D, I thought I had decent video quality.  But this method now, very simply put:  HT enabled, AM of 4084 for six core, VSYNC to 30Hz screen always delivers perfection in video performance as long as we can maintain 30fps.  This is why I will definitely purchase an I9-9900K which supports HT instead of the 9700K which does not.

Bert, have you done what I'm talking about?  

Sled:  get your display refresh and 30Hz and enjoy that awesome 8700K :o)

Edited April 19, 2019 by Noel

[itsjase 52]

2 hours ago, Bert Pieke said:

As I stated above, the CPU usage does NOT drop when I use RTSS Scanline X/2, but it DOES drop if I turn on Vsync in P3D.   BTW, CPU usage also drops if I lock fps in NVI at 30...

In all three cases, fps end up at roughly 30 fps.

In summary, if I wanted lower CPU usage, there are at least two ways to accomplish that.. but my goal is smooth flight, and RTSS is the leading contender as far as I can tell. 

That's interesting. turning on vsync in P3D or using NVI is exactly the same as using RTSS for me. No cpu usage drop. Only way I can get it to drop is by doing Sled's trick.

Edited April 19, 2019 by itsjase

[itsjase 52]

I've also made a super simple .bat file which does the core disabling/reenabling automatically if people are interested.

[Noel 4,501]

5 hours ago, itsjase said:

I've also made a super simple .bat file which does the core disabling/reenabling automatically if people are interested.

Can't afford the loss of processing power--wish I could!

[lownslo 592]

8 hours ago, Noel said:

I decided to combine Sled's setup...  Video quality was quite good, especially considering the low average CPU utilization.  Flat out remarkable--the low CPU use.  Much lower temps as mentioned previously.  But...video quality WAS NOT at all PERFECTLY SMOOTH!   Oh, it was pretty good for sure...until you really looked.

This has been my experience too.  Balancing core usage seems to be the correct approach, but what I see on my monitor and the level of detail I can dial up while running the first core at 100% is easily better.  I rarely experience long frames with my 100% first core config, but I do see them a lot more when I balance the core load.

9 hours ago, Rogen said:

Vsyncing @ 30 Hz will reduce load, fairly signiticantly for the GPU...

Using RTSS to sync at 30hz has allowed me to increase the settings of my 1080... much more eye candy!

Greg

[Bert Pieke 7,585]

If I had a 30 Hz monitor, I would try it, but mine only runs at 60 Hz..

[newtie 533]

48 minutes ago, lownslo said:

This has been my experience too.  Balancing core usage seems to be the correct approach, but what I see on my monitor and the level of detail I can dial up while running the first core at 100% is easily better.  I rarely experience long frames with my 100% first core config, but I do see them a lot more when I balance the core load.

I mentioned this way back in this thread. Just because the first core is running at 100% does not necessarily mean something is wrong and its gotta be fixed.

I'm erring with LM on this in that this is the way it is for a reason.

Cheers,

Mark

 

[Noel 4,501]

1 hour ago, Bert Pieke said:

If I had a 30 Hz monitor, I would try it, but mine only runs at 60 Hz..

No utilities that can force it to work do you know Bert?  It's really THE ONLY way I've had perfect smoothness...though some other approaches can get you in the ballpark ;o).  Bert:  take an easy to run plane to a very low demand setting and go for 60Hz to experience vsync to a monitor's refresh rate.  I haven't done this but it has to be the holy grail of smooth flying as long as you can stay at least 5 or so fps >60 at all time, or at least for your test flight.   Unlimited frames, no RTSS, just vsync to 60Hz screen and see how things go.  For my old CPU HT on is a must to achieve perfection at 30Hz, but maybe you can get away w/ HT off and get there.

Dang I'm enjoying the Phenom!  For the first 5 flights I didn't realize it but my dang JS calibration thru FSUIPC somehow got messed up so at rotation the dang thing would shoot up fast!  Finally figured out the problem and now it's a dream, and a good one at that!  Thanks!

Edited April 19, 2019 by Noel

[Sethos 4,336]

1 hour ago, Bert Pieke said:

If I had a 30 Hz monitor, I would try it, but mine only runs at 60 Hz..

Have you tried simply setting up a custom resolution, where you choose 30Hz instead of 60?

[Patrick Zwamborn 0]

8 hours ago, itsjase said:

I've also made a super simple .bat file which does the core disabling/reenabling automatically if people are interested.

Yes, sharing your .bat file (or text) is much appreciated. Thanks Patrick 

[SteveW 1,792]

1 hour ago, newtie said:

I mentioned this way back in this thread. Just because the first core is running at 100% does not necessarily mean something is wrong and its gotta be fixed.

I'm erring with LM on this in that this is the way it is for a reason.

Cheers,

Mark

 

Yes. Ultimately we can add in detail and there's more work to be done takes longer and slows fps down when it all takes longer than a frame to draw. No matter how it is paralleled up, we will add in more detail and maximise a core.

Can apps choose exactly the core or LP they desire, yes,  can they know about HT on or off, of course they can. Can the jobscheduler be helped to choose a particular LP, of course it can. Do apps do that, not much very rarely. A program like P3D can be built so that when it runs it will partition out it's parts, in order. That means the jobscheduler will move one-along for the next desired core specific task rather than load up the same core more. That's decided by the nature of the program, the jobscheduler just goes along with it. So we see the famous 100% core with unlimited no VSync, and we see it limited in VSync to something less than 100% if we chose reasonable settings.

When we move the program amongst cores after the program has laid out it's preferred arrangement of tasks, as P3D does, the jobscheduler is left responsible for moving all the threads, the app doesn't do that. The app can end up not necessarily so well off, or better, depends on the setup and the moves made.

Adding more cores allows background tasks to split out among those extra cores and they use up system resources along with the resources required for the sim. This does not mean the system can provide the resource with more cores, any faster, and can lead to performance issues.

Sometimes more fps can be seen at the expense of the reality bubble being filled less efficiently. So fps is not a factor in determining performance capacity within the setup.

[Noel 4,501]

3 minutes ago, SteveW said:

So we see the famous 100% core with unlimited no VSync, and we see it limited in VSync to something less than 100% if we chose reasonable settings.

Yep and I know I must remain a little below 100% to maintain great freedom from stutters, so I adjust settings accordingly.  Thank you SteveW for all of your shared insights it certainly helped me lose the annoying stutters I had going for the better part of 15y.