40 minutes ago, GSalden said:

when using 3 screens :

How come your 3 screens aren't in a continuous image? I thought you were using that so-called miracle fish-eye distortion fix in P3D?

Pete

 

Edited August 7, 20187 yr by Pete Dowson

@Pete Dowson
they have!
 

 

Edited August 7, 20187 yr by Chris Bell

On 8/6/2018 at 10:08 AM, GSalden said:

Rob, with all respect : why did V4.2 not have these issues and V4.3 does ?

You tell me, I can't replicate but I'm only running a single GPU, so not much info I can provide.  I'm still trying to figure how:

1.  moving your GPU to another slot and you recover 2X FPS?

2.  How your dynamic FFTF is working at all, let alone a setting of 0.01 (that should trigger the death of AG and loading of high res terrain tiles)

3.  Your three screen configuration has me completely baffled 2 + 1?

4.  Your running multiple AI aircraft products at the same time?  Did we get to see how these are configured, % of traffic?

5.  Don't really know what test you are using as a reference point and is it always consistent? (same time of day, same route/path, etc.)

I can somewhat understand no VC as responding to SLI better with SSAA, but some of your FPS numbers seem to defy physics and yet you're having performance issues with AS Frankfurt , FTX Germany , True Earth Netherlands and FT Amsterdam?  

Did we ever get to see you graphics settings ... all 5 pages?

Are you using and external shader tools (PTA, ENVSHADE, etc.)?

Any texture swaps (clouds, water, etc.)?

Personally I'd drop SSAA, remove Dynamic FFTF, switch to single GPU, remove any Prepar3d.cfg edits you've made, clear skies and no weather, disable all AI traffic, and start your testing from there?  Assuming your EXE.XML (both) and DLL.XML (both) are clear and/or no duplicate entries?  Then progressively test each item by adding it back and see where that takes you ... test them individually and NOT together or accumulative.

Cheers, Rob.

EDIT: OR if you feel it's P3D V4.3 at fault, you can easily downgrade the client only to P3D V4.2 and test again.

 

46 minutes ago, Rob Ainscough said:

You tell me, I can't replicate but I'm only running a single GPU, so not much info I can provide.  I'm still trying to figure how:

 

Dude??? Gerard is your protege, Fix it!!! 🙂     ...I mean you started this 16 core, 3 gpu, 4 8k monitor stuff ..... now clean it up.  <vbg>

7 hours ago, Pete Dowson said:

How come your 3 screens aren't in a continuous image? I thought you were using that so-called miracle fish-eye distortion fix in P3D?

Pete

 

It is 

This is s screenshot . Left view is on a 40"monitor . Middel and right are 32" monitors and in NVSurround. So looking at the screens the views are a continuous image without distortion.

 

 

6 hours ago, Rob Ainscough said:

You tell me, I can't replicate but I'm only running a single GPU, so not much info I can provide.  I'm still trying to figure how:

1.  moving your GPU to another slot and you recover 2X FPS?

The SLI was not kicking in correctly it seemed. Tests are done with 1 GPU

2.  How your dynamic FFTF is working at all, let alone a setting of 0.01 (that should trigger the death of AG and loading of high res terrain tiles)

No issue with loading AG and no problems with loading ( no blurries ). The 0.01 helps with max framerate on the ground. The 7940X has more than enough terrain loading cores, However, I also tried multiple times without FFTF Dynamic and the outcome is the same. Multiple people ( like Nico Kaan, Frank Libretto, Henk Timmermans and some people reporting here at the forum ) also suffer from the framerate drop over time, so it is not just me.  

3.  Your three screen configuration has me completely baffled 2 + 1?

See my answer to Pete's post

4.  Your running multiple AI aircraft products at the same time?  Did we get to see how these are configured, % of traffic?

No, I am testing with : 1 no AI  2 PSXseeconTraffic  3 Traffic Early Acces . So not at the same time.  PSXseeconTrraffic : 15% parked aircraft at Schiphol + Live traffic ( it does not use flightplans and therefore is easier on the system ) . Traffic Early Acces : 15% .

5.  Don't really know what test you are using as a reference point and is it always consistent? (same time of day, same route/path, etc.)

Same saved flight without weather program so I start and end in the same conditions. I also have the same flight saved again under another name to be sure.

I can somewhat understand no VC as responding to SLI better with SSAA, but some of your FPS numbers seem to defy physics and yet you're having performance issues with AS Frankfurt , FTX Germany , True Earth Netherlands and FT Amsterdam?  

If I fly from Amsterdam to Frankfurt the drop is way less. Probably as I am climbing out of Amsterdam there is not much TE NL scenery to be seen. 

Did we ever get to see you graphics settings ... all 5 pages?

Will do that today ( FYI 3x 1920 ; no 4K )

Are you using and external shader tools (PTA, ENVSHADE, etc.)?

Tried with PTA and later went back to default Shaders. No difference.

Any texture swaps (clouds, water, etc.)?

Tried with ASCA clouds / Envtex skies + water and also with default textures. No difference. 

Personally I'd drop SSAA, remove Dynamic FFTF, switch to single GPU, remove any Prepar3d.cfg edits you've made, clear skies and no weather, disable all AI traffic, and start your testing from there?  Assuming your EXE.XML (both) and DLL.XML (both) are clear and/or no duplicate entries?  Then progressively test each item by adding it back and see where that takes you ... test them individually and NOT together or accumulative.

I have always used 4x SSA/16AF/ texture size 1024 and never had any issues.on a single GPU ( 50-90% ) load range during a flight.

Cheers, Rob.

EDIT: OR if you feel it's P3D V4.3 at fault, you can easily downgrade the client only to P3D V4.2 and test again.

That I understand, but P3d is coming from 4.3 and going towards 4.4. Therefore I am anxious to find out what is causing this.   

 

 

Here at startup with FT Schiphol + True Earth Netherlands .

See the framerate. To me it does not look that my system cannot handle it .

 

Think you shall test to downclock the cpu , the problem you describe, problem with gpu slots and other like usb , intigrated networkcard i seen when the OC was on the edge.

have my 7920X up and running in P3D-V4.3 with same brand of gpus same settings in p3d as with the 8700k

tested lot of different OC profiles for the 7920X , run HT off 4.9ghz sync all cores that work better then mixed clocks for me.

the 8700k sytem at 5.4ghz HT off is smother not overload the cpu ( usage 85-90% with bursts to 100% , on the 7920X the usage ar higher) run Vsync unlimited.

the 8700k is approx 15% faster ( cpu+memory clock) that do the smothness (my personal thougts) on the other hand the loadtimes is faster with the 7920X.

On both systems i not seen any fps drops over time done 6 hour flights on both system ( only fly with FSL A319 )

Edited August 8, 20187 yr by westman

Core 0 is at 4.8 Ghz

Core 1 is at 4.7 Ghz

Core 2-13 at 4.4 Ghz

Heat : max 56 degrees Celcius during a flight

On 8/6/2018 at 6:49 PM, Rob Ainscough said:

Some objects shouldn't be terminated like AI.  

But just so you know, "letting go" (dealloc of resources) is a time consumer, the less one does it the better the FPS at the cost of more memory.  Normally dealloc or GC (garbage collection) happens during idle time, if you have no idle item then it may not happen.

I don't have PSXseeconTraffic so I can't comment, but if it's similar to UTLive with AI creation via SimConnect then I would expect FPS drop to happen over time as AI schedules (departures/arrival) increase filling the global with AI routes.  Pending how the schedules are define some airports may have a higher arrival/depature rate at specific times than other airports, you'd need to know the AI schedules ... also, what is your AI density and how complex are your AI models?

Cheers, Rob.

Hi Rob,

Why should AI objects not be terminated? Gerard is talking about AI objects from PSXseeconTraffic. I am the developer of PSXseeconTraffic and for your information  PSXseeconTraffic works quite different from UTL although both use SimConnect. 

PSXseeconTraffic creates an AI object in P3D with a SimConnect call SimConnect_AICreateNonATCAircraft (to reflect a real live aircraft from information that is coming from an external traffic feed program) and changes its position, heading, pitch and bank 25 times per second via SimConnect calls until it is out of the lateral and vertical ranges defined by the user. Then PSXseeconTraffic removes the AI object from P3D via SimConnect. So there are no AI schedules or AI routes involved at all.... The AI object is an object like any other. 

Now it looks that P3Dv4 does not (always) free the resources, resulting in a drop in fps in a long flight.... ? 

Regards,

Nico Kaan

Edited August 8, 20187 yr by kiek

Testing at Schiphol right now with SLI

1 no AI —> 58 fps on rw24 with ASP4 running and bad weather 

2 no AI —> same as above but looking ftowards the entries airport —> 38

3 As 2 , but PSXseeconTraffic running with 15% parked aurcraft —> 29

4 same as 3 but the Sim paused . The PSXseeconTraffic aircraft keep on taxiing , taking off and landing —> 37

2 hours ago, GSalden said:

This is s screenshot . Left view is on a 40"monitor . Middel and right are 32" monitors and in NVSurround.

Yes, so it is two P3D Windows, not one. The big problems with stretching the one window over, say, 180 degrees, are that:

(a) the max width zoom you can get in only about 150 degrees, and

(b) the fish eye distortion at either side in unbearable even with 150 degrees. 

I really don't see how minor changes to shaders and the like can fix the distortion, and it certainly can't give the 210 degrees I need.  Isn't that what the miracle fix is supposed to provide: no fish-eye?

Stretching over two monitors for something like a 120 degree FOV is relatively easy. I had a 100 degree view across my 10 foot screen with one projector before I got the curved 210 degree screen. I dealt with the distortion at the edges by changing the rectangular projection to a rhomboid, and tilting my point of view up a little.

Pete

 

The wide window is NVSurround and the fisheye is corrected.

Also the left view has the fisheye corrected.

These are 2 camera’s.

48 minutes ago, GSalden said:

Testing at Schiphol right now with SLI

1 no AI —> 58 fps on rw24 with ASP4 running and bad weather 

2 no AI —> same as above but looking ftowards the entries airport —> 38

3 As 2 , but PSXseeconTraffic running with 15% parked aurcraft —> 29

4 same as 3 but the Sim paused . The PSXseeconTraffic aircraft keep on taxiing , taking off and landing —> 37

After 1h + 20 min the framerate ( option 3 testing ) has gone down to 21.

Just planespotting

Edited August 8, 20187 yr by GSalden

Gerard, are you saying opt 3, your AI Traffic is reducing your fps as you fly, when it did not in 4.2?