Jump to content
Sign in to follow this  
BWBriscoe

Nvidia performance article

Recommended Posts

57 minutes ago, Grand_Pubah said:

Just a quick question. I've been following this hoping to find the same fix. Can you tell me exactly what would be the exact name of the dll files you mentioned? the closest I could find is d3dx11_43.dll in the P3Dv5 main folder, but nothing like that in C:\Windows\SysWOW64 or the Win folders.

Thank you

So this is what I did,
Go to C:\Windows\SysWOW64
In that folder, there are lots of dll' files, some of them are the same as the ones in the P3D root folder. 
Now copy the matching dll's from C:\Windows\SysWOW64 into your root P3D folder, D3D12onWin7 folder.
Replace the files that are there, not adding anything new.
So I did copy from C:\Windows\SysWOW64
 

d3d11on12.dll

D3D12.dll

D3DCompiler_43.dll

D3DCompiler_47.dll

d3dcsx_42.dll

d3dcsx_43.dll

D3DX9_43.dll

d3ddx11_43.dll

DXCore.dll

dxilconv.dll

to directory D3D12onWin7 inside the main P3D directory.
Also under the "water and bathymetry" menu inside P3D turn off "Enable Bathymetry"
 
But please before you do this backup!
I struggled with this issue until I found this, it works on my PC. I hope this will solve the issue for you guys.

I9- 12900K- CPU @ 5.0GHz, 64 GB RAM @ 5200MHz, NVIDIA RTX 3090

Share this post


Link to post

@killthespam, I made copies of 3 DLLs in the main P3Dv5 folder before copying the equivalents from SysWow64. There were 2 DLLs in D3DI2onWin7 that needed replacing. Again, copies were made first.

Enable Bathymetry was already disabled. I launched P3Dv5. It took far longer than normal to launch but eventually did. My loaded aircraft was a Carenado PC12 and I opted for the Mooney Bravo.

I'm sad to report the same DXGI ACCESS_DENIED message appeared prompting me to close P3D. I chose Yes.

Rebooted the PC and tried again but this time at the Scenario screen switched to a default Bravo. Once loaded I tried again with a Piper Cub. Same result. I'll revert to the original files.

Thanks for trying to help though.

 


Ray (Cheshire, England).
System: P3D v5.3HF2 & v3.4, Intel i7-8086K o/c to 5.0Ghz, Nvidia GTX 1080 Ti 11Gb, Samsung 970 EVO M.2 SSD, 1Tb Samsung 860 EVO SSD, Asus Prime Z370-A mobo, 32Gb G.Skill DDR4 3000Mhz RAM, Win 10 Pro 64-bit, BenQ PD3200U 32” monitor, Fulcrum One yoke.
Cheadle Hulme Weather

Share this post


Link to post
1 hour ago, Ray Proudfoot said:

@killthespam, I made copies of 3 DLLs in the main P3Dv5 folder before copying the equivalents from SysWow64. There were 2 DLLs in D3DI2onWin7 that needed replacing. Again, copies were made first.

Enable Bathymetry was already disabled. I launched P3Dv5. It took far longer than normal to launch but eventually did. My loaded aircraft was a Carenado PC12 and I opted for the Mooney Bravo.

I'm sad to report the same DXGI ACCESS_DENIED message appeared prompting me to close P3D. I chose Yes.

Rebooted the PC and tried again but this time at the Scenario screen switched to a default Bravo. Once loaded I tried again with a Piper Cub. Same result. I'll revert to the original files.

Thanks for trying to help though.

 

Ahhh, I'm so sorry to hear that. I wished it did help.


I9- 12900K- CPU @ 5.0GHz, 64 GB RAM @ 5200MHz, NVIDIA RTX 3090

Share this post


Link to post
4 hours ago, killthespam said:

Ahhh, I'm so sorry to hear that. I wished it did help.

It begs the question as to why it works for some but not others. I’ve posted the problem on the LM forum but they don’t even acknowledge my post let alone suggest a fix.

When I have a problem with Active Sky, PMDG or FS Labs Concorde I invariably get a response from the developers. LM sells us their software but seemingly has no interest in helping us. Not very impressed! 

  • Like 2

Ray (Cheshire, England).
System: P3D v5.3HF2 & v3.4, Intel i7-8086K o/c to 5.0Ghz, Nvidia GTX 1080 Ti 11Gb, Samsung 970 EVO M.2 SSD, 1Tb Samsung 860 EVO SSD, Asus Prime Z370-A mobo, 32Gb G.Skill DDR4 3000Mhz RAM, Win 10 Pro 64-bit, BenQ PD3200U 32” monitor, Fulcrum One yoke.
Cheadle Hulme Weather

Share this post


Link to post

One thing I've noticed with P3Dv5 and DX12 is that the GPU statistic I monitor most closely--normalized total power--tends to spike up quite a bit higher than the traditional "GPU load" figure would indicate.  Even a 3090 bumps up into the neighborhood of 100% NTP when heavily loaded at 4K res--and that when constrained to 30 fps via VSync and hardware refresh rate.  Less than 100% "GPU Load" may not guarantee that the GPU isn't being overtasked.  That load figure only tells you how much of the GPU's available processor time is utilized, where NTP is a more direct indication of the real work being done, which is presumably higher when more cores are doing more complex calculations.  Might be worth a try to dial back the heavy GPU load to keep NTP between the lines.

My second (portable) sim PC uses a 9900K and a 1080Ti--I'm sticking with P3Dv4 on that one because I'm still not sure the 1080Ti is going to be up to the job in v5 without running into these DXGI problems, which I strongly suspect are associated with overtasking the GPU.


Bob Scott | President and CEO, AVSIM Inc
ATP Gulfstream II-III-IV-V

System: i9-10900K @ 5.2GHz on custom water loop, ASUS Maximus XII Hero, 32GB GSkill 3600MHz CAS15, EVGA RTX3090 XC3 Ultra
Samsung 55" JS8500 4K TV@30Hz,
3x NVME 2x SATA Samsung SSD, EVGA 1KW PSU, 1.2Gbps internet
SoundBlaster XFi Titanium, TOSLINK to Yamaha RX-V467 HT Rcvr, Polk/Klipsch 6" bookshelf spkrs, Polk 12" subwoofer, 12.9" iPad Pro
PFC yoke/throttle quad/pedals with custom Hall sensors, Coolermaster HAF932 case, Stream Deck XL button box

Share this post


Link to post
18 minutes ago, w6kd said:

I'm sticking with P3Dv4 on that one because I'm still not sure the 1080Ti is going to be up to the job in v5 without running into these DXGI problems, which I strongly suspect are associated with overtasking the GPU.

Bob, although I’ve only had v5 for 12 days I’ve never had a DXGI error when flying. My ACCESS DENIED problem is very different to running out of VRAM or demands on the GPU.

My error occurs with a C&D cockpit in a default aircraft and I select a different vehicle, also default. How can my 1080Ti be overtaxed in those conditions? Mine is unrelated to the problems you and others are experiencing.


Ray (Cheshire, England).
System: P3D v5.3HF2 & v3.4, Intel i7-8086K o/c to 5.0Ghz, Nvidia GTX 1080 Ti 11Gb, Samsung 970 EVO M.2 SSD, 1Tb Samsung 860 EVO SSD, Asus Prime Z370-A mobo, 32Gb G.Skill DDR4 3000Mhz RAM, Win 10 Pro 64-bit, BenQ PD3200U 32” monitor, Fulcrum One yoke.
Cheadle Hulme Weather

Share this post


Link to post
11 minutes ago, Ray Proudfoot said:

Bob, although I’ve only had v5 for 12 days I’ve never had a DXGI error when flying. My ACCESS DENIED problem is very different to running out of VRAM or demands on the GPU.

My error occurs with a C&D cockpit in a default aircraft and I select a different vehicle, also default. How can my 1080Ti be overtaxed in those conditions? Mine is unrelated to the problems you and others are experiencing.

I'd look at backing off on any overclock of the GPU and/or relaxing the timeout settings (TdrDdiDelay and TdrDelay in WDDM registry settings).  I know some people have gotten that error to go away by downclocking their card from a factory or user overclock,  suggesting that the GPU is timing out due to some sort of overrun during those acft swaps.  A lot of data is still getting sent to the GPU during an aircraft change.  It appears that even a momentary/transitory issue that overloads or otherwise delays the GPU from responding in the expected timeframe can trigger the error.

FWIW, I can't reproduce the error here--I can swap between acft, default or otherwise, without issue.  The only DXGI issues I ever had with a 2080Ti or my present 3090 were due to the (now apparently fixed) bugs in older beta versions of Trueglass.  I do still use the aforementioned relaxed timeout settings as a standard configuration tweak on all my sim setups since the days of P3Dv3 -- TdrDdiDelay set to 20 (default is 5) and TdrDelay set to 10 (default is 2).

 


Bob Scott | President and CEO, AVSIM Inc
ATP Gulfstream II-III-IV-V

System: i9-10900K @ 5.2GHz on custom water loop, ASUS Maximus XII Hero, 32GB GSkill 3600MHz CAS15, EVGA RTX3090 XC3 Ultra
Samsung 55" JS8500 4K TV@30Hz,
3x NVME 2x SATA Samsung SSD, EVGA 1KW PSU, 1.2Gbps internet
SoundBlaster XFi Titanium, TOSLINK to Yamaha RX-V467 HT Rcvr, Polk/Klipsch 6" bookshelf spkrs, Polk 12" subwoofer, 12.9" iPad Pro
PFC yoke/throttle quad/pedals with custom Hall sensors, Coolermaster HAF932 case, Stream Deck XL button box

Share this post


Link to post

@w6kd, it’s a professionally build PC and I’m pretty sure the GPU hasn’t been overclocked.

How much data is involved in selecting an avatar? Absolutely nothing surely? And it’s an ACCESS DENIED issue. Still don’t know access to what.

But consider how much data is involved changing from airport to airport multiple times in the same session. There’s no logic to it.

Anyway, I tend not to switch vehicles from inside the sim. The scenario menu allows for it so its a pretty low priority issue for me. Should they start happening inflight that’s another matter.

 Where do you set that  timeout value? I’ll check what mine is.


Ray (Cheshire, England).
System: P3D v5.3HF2 & v3.4, Intel i7-8086K o/c to 5.0Ghz, Nvidia GTX 1080 Ti 11Gb, Samsung 970 EVO M.2 SSD, 1Tb Samsung 860 EVO SSD, Asus Prime Z370-A mobo, 32Gb G.Skill DDR4 3000Mhz RAM, Win 10 Pro 64-bit, BenQ PD3200U 32” monitor, Fulcrum One yoke.
Cheadle Hulme Weather

Share this post


Link to post
36 minutes ago, Ray Proudfoot said:

@w6kd, it’s a professionally build PC and I’m pretty sure the GPU hasn’t been overclocked.

How much data is involved in selecting an avatar? Absolutely nothing surely? And it’s an ACCESS DENIED issue. Still don’t know access to what.

But consider how much data is involved changing from airport to airport multiple times in the same session. There’s no logic to it.

Anyway, I tend not to switch vehicles from inside the sim. The scenario menu allows for it so its a pretty low priority issue for me. Should they start happening inflight that’s another matter.

 Where do you set that  timeout value? I’ll check what mine is.

Many, probably most, high-end GPUs come out of the box already factory-overclocked.  If your 1080Ti runs a clock/boost speed over 1481/1582 MHz or a memory clock faster than 5504 MHz, then it has been overclocked at the factory beyond the nVidia specs for a 1080Ti.  All of the GPUs in current use here--two 980Ti, two 1080Ti, a 2080Ti, and a 3090 came overclocked to varying degrees from the manufacturer.  I had to downclock one of the 980Ti boards from its manufacturer default after about six months of use because it wasn't truly long-term stable at the factory O/C in some applications, despite being water-cooled.

Most likely that DXGI_ACCESS_DENIED error refers to memory access needed for the driver to write data to the VRAM.

The timeout value details are at https://docs.microsoft.com/en-us/windows-hardware/drivers/display/tdr-registry-keys
You won't normally find those registry keys on your system, as they are intended to be used in testing drivers, so they don't normally exist in a default Windows installation.

  • Like 2

Bob Scott | President and CEO, AVSIM Inc
ATP Gulfstream II-III-IV-V

System: i9-10900K @ 5.2GHz on custom water loop, ASUS Maximus XII Hero, 32GB GSkill 3600MHz CAS15, EVGA RTX3090 XC3 Ultra
Samsung 55" JS8500 4K TV@30Hz,
3x NVME 2x SATA Samsung SSD, EVGA 1KW PSU, 1.2Gbps internet
SoundBlaster XFi Titanium, TOSLINK to Yamaha RX-V467 HT Rcvr, Polk/Klipsch 6" bookshelf spkrs, Polk 12" subwoofer, 12.9" iPad Pro
PFC yoke/throttle quad/pedals with custom Hall sensors, Coolermaster HAF932 case, Stream Deck XL button box

Share this post


Link to post

@w6kd, Bob, thanks for the info and link. I'll check the GPU clock speed tomorrow.

But if my 1080Ti has behaved perfectly for 2.5 years with DX11 in v4.5 but only starts playing up with DX12 I don't consider the card to be the problem.

Microsoft or LM need to adjust their DX12 settings since it's pretty clear to me that's where the problem lies.

  • Like 1

Ray (Cheshire, England).
System: P3D v5.3HF2 & v3.4, Intel i7-8086K o/c to 5.0Ghz, Nvidia GTX 1080 Ti 11Gb, Samsung 970 EVO M.2 SSD, 1Tb Samsung 860 EVO SSD, Asus Prime Z370-A mobo, 32Gb G.Skill DDR4 3000Mhz RAM, Win 10 Pro 64-bit, BenQ PD3200U 32” monitor, Fulcrum One yoke.
Cheadle Hulme Weather

Share this post


Link to post
9 hours ago, Ray Proudfoot said:

@w6kd, Bob, thanks for the info and link. I'll check the GPU clock speed tomorrow.

But if my 1080Ti has behaved perfectly for 2.5 years with DX11 in v4.5 but only starts playing up with DX12 I don't consider the card to be the problem.

Microsoft or LM need to adjust their DX12 settings since it's pretty clear to me that's where the problem lies.

My suspicion is that DX12 just pushes the graphics subsystem closer to the margins than DX11.  With P3D v4 and DX11 I rarely see VRAM usage over 5 GB or NTP figures nearly as high as P3Dv5/DX12.  I'd also expect that the testing on the rendering engine is done with components running within the nVidia specs--as always, running beyond those specs puts the user into uncharted territory with no guarantees. 

It sure looks to me that, when using DX12, configuring things in a way that provides some performance margin in terms of power and memory bandwidth is how you get to stability--either through a better-performing GPU or by mitigating load via more conservative hardware/software configuration choices, or some combination of both.  Running at the same overclocked speeds as before or to the same 80-90% peak "GPU Load" figure in DX12 may not take into account that power delivery and VRAM are more stressed, and that actual core workload is higher.  My observations on NTP definitely have me wondering if power regulation on the GPU board is becoming a limfac.

It's good that we can now push more work to the GPU, but we have to remember that that's *exactly* what we're doing--more work placed on the same subsystem at the same speeds.  A good analog is what happens on a CPU when we put very core-intensive AVX instructions into the mix in a run of Prime95--at the same clock speeds the CPU will heat up more (indicative of doing more work) and often it can take an otherwise working non-AVX overclock out past the point of stability. 

So, bottom line, DX11 and DX12 look to be apples and oranges...direct comparisons or expectations about the performance/stability of one formed from observations of the other aren't holding up under real-use scrutiny.  The fact that I've seen both simmers and non-sim gamers alike getting to stability by downclocking their hardware or reducing GPU demand seems to support that theory.


Bob Scott | President and CEO, AVSIM Inc
ATP Gulfstream II-III-IV-V

System: i9-10900K @ 5.2GHz on custom water loop, ASUS Maximus XII Hero, 32GB GSkill 3600MHz CAS15, EVGA RTX3090 XC3 Ultra
Samsung 55" JS8500 4K TV@30Hz,
3x NVME 2x SATA Samsung SSD, EVGA 1KW PSU, 1.2Gbps internet
SoundBlaster XFi Titanium, TOSLINK to Yamaha RX-V467 HT Rcvr, Polk/Klipsch 6" bookshelf spkrs, Polk 12" subwoofer, 12.9" iPad Pro
PFC yoke/throttle quad/pedals with custom Hall sensors, Coolermaster HAF932 case, Stream Deck XL button box

Share this post


Link to post

I have been having a problem for the last month since windows update some CTD`s striped it back form all addons added since Christmas no joy, but I have tried the DX12 dll copy and past into the main P3Dv5.1. And it has worked a treat for me still testing have tried to CTD it pushed the settings up in flight still loaded no CTD rock solid went all the way with addons loaded and shaders re-loaded.

My settings.

 https://www.flickr.com/photos/154944674@N05/51072977906/in/dateposted-public/

https://www.flickr.com/photos/154944674@N05/51003792090/in/dateposted-public/

https://www.flickr.com/photos/154944674@N05/51073075447/in/dateposted-public/

https://www.flickr.com/photos/154944674@N05/51003791990/in/dateposted-public/

https://www.flickr.com/photos/154944674@N05/51003794725/in/dateposted-public/

https://www.flickr.com/photos/154944674@N05/51072980376/in/dateposted-public/

Everything on changed settings in flight to try to CTD it no go still loaded will test with the big boys FPS locked 30.

Edited by G-RFRY

 

Raymond Fry.

PMDG_Banner_747_Enthusiast.jpg

Share this post


Link to post
6 hours ago, w6kd said:

My suspicion is that DX12 just pushes the graphics subsystem closer to the margins than DX11.  With P3D v4 and DX11 I rarely see VRAM usage over 5 GB or NTP figures nearly as high as P3Dv5/DX12.  I'd also expect that the testing on the rendering engine is done with components running within the nVidia specs--as always, running beyond those specs puts the user into uncharted territory with no guarantees. 

It's good that we can now push more work to the GPU, but we have to remember that that's *exactly* what we're doing--more work placed on the same subsystem at the same speeds.  A good analog is what happens on a CPU when we put very core-intensive AVX instructions into the mix in a run of Prime95--at the same clock speeds the CPU will heat up more (indicative of doing more work) and often it can take an otherwise working non-AVX overclock out past the point of stability. 

So, bottom line, DX11 and DX12 look to be apples and oranges...direct comparisons or expectations about the performance/stability of one formed from observations of the other aren't holding up under real-use scrutiny.  The fact that I've seen both simmers and non-sim gamers alike getting to stability by downclocking their hardware or reducing GPU demand seems to support that theory.

If I encountered problems when loading 3rd party airports I could understand that might cause issues. But the card has worked flawlessly and during the last two weeks I've installed a lot of airports and gone to each to check they're okay. In 100% of cases they were.

My issue appears to be one very few people are talking about. Namely the inability of the GPU to cope with a change of vehicle within the sim. The fact I can load a different scenario including a change of vehicle without issue further convinces me there is a bug somewhere that only a few are experiencing but the Microsoft help on DXGI errors is painfully thin.

I'm attaching a screenshot of my GPU as seen by Afterburner. To m untrained eye it doesn't appear to be overclocked.

1080Ti.jpg


Ray (Cheshire, England).
System: P3D v5.3HF2 & v3.4, Intel i7-8086K o/c to 5.0Ghz, Nvidia GTX 1080 Ti 11Gb, Samsung 970 EVO M.2 SSD, 1Tb Samsung 860 EVO SSD, Asus Prime Z370-A mobo, 32Gb G.Skill DDR4 3000Mhz RAM, Win 10 Pro 64-bit, BenQ PD3200U 32” monitor, Fulcrum One yoke.
Cheadle Hulme Weather

Share this post


Link to post

P3D V5 is much more GPU bound than V4.. the biggest issue with DX12 is that in order to preserve performance it requires BY DESIGN to avoid sharing data between normal RAM and VRAM.

With DX11 if you ran out of GPU VRAM (Video RAM), the system was allowed to borrow normal RAM.. however this came with a big performance impact. Most people moved all the sliders to the right under P3D V4, and saw no errors but things became very SLOW.. this is because VRAM was starved and the GPU had to spend lots of time to handle operations between the GPU--> CPU --> Normal RAM -->CPU --> GPU in a constant cycle.

DX12 BY DESIGN, doesn't allow normal RAM usage, if VRAM is starved, it is the end of the process.. I know people might think this is dumb and tragic.. but in reality it is more efficient because in the future new DX12 native GPU cards can be designed to be more hardware and software intensive (because there is no need to worry about code to handle CPU --> RAM --> GPU, etc.), and such design will allow to produce much more accurate and realistic graphics.

Coming back to P3D, people had all these sliders to the right.. they were already VRAM starved.. installed V5 put the same settings they had under V4.. and kaboom.. VRAMs OOMS.. VRAM OOMS are very common with DX12, this is not an issue only present with P3D.. in addition P3D pushes our GPUs harder than any other video game, this is because an entire world has to be rendered in front of your eyes rather than "Scenes" or "Cut Scenes" or Limited world space as most video games which can have specific VRAM budgets to be meet avoiding any VRAM starvation.

Basically eventually all video cards in the future will be wired differently and DX11 will be totally deprecated.

@Ray ProudfootYour problem is very peculiar and particular.. I am having a very hard week at work at the moment (my non Flight sim Job), basically fending off an international hacking group trying to bring our systems down, I will contact you next week when things are calmer to provide you with some specific configuration under V5 to debug the DX12 error you are having, with this data we might find the cause or I can ask LM devs posible solutions, etc.

Regards,

Simbol

 

  • Like 2
  • Upvote 1

Share this post


Link to post
19 minutes ago, simbol said:

ProudfootYour problem is very peculiar and particular.. I am having a very hard week at work at the moment (my non Flight sim Job), basically fending off an international hacking group trying to bring our systems down, I will contact you next week when things are calmer to provide you with some specific configuration under V5 to debug the DX12 error you are having, with this data we might find the cause or I can ask LM devs posible solutions, etc.

Thank you Simbol, that’s much appreciated. I hope you can stop the hackers.

I’ve never had scenery sliders fully right. I prefer to see more AIG Ai aircraft than buildings. I can work around this problem UTL as a former programmer it does bug me and I’d love to know what’s causing it and more importantly, how to fix it.


Ray (Cheshire, England).
System: P3D v5.3HF2 & v3.4, Intel i7-8086K o/c to 5.0Ghz, Nvidia GTX 1080 Ti 11Gb, Samsung 970 EVO M.2 SSD, 1Tb Samsung 860 EVO SSD, Asus Prime Z370-A mobo, 32Gb G.Skill DDR4 3000Mhz RAM, Win 10 Pro 64-bit, BenQ PD3200U 32” monitor, Fulcrum One yoke.
Cheadle Hulme Weather

Share this post


Link to post

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
Sign in to follow this  
  • Tom Allensworth,
    Founder of AVSIM Online


  • Flight Simulation's Premier Resource!

    AVSIM is a free service to the flight simulation community. AVSIM is staffed completely by volunteers and all funds donated to AVSIM go directly back to supporting the community. Your donation here helps to pay our bandwidth costs, emergency funding, and other general costs that crop up from time to time. Thank you for your support!

    Click here for more information and to see all donations year to date.
×
×
  • Create New...