[James Ross]

 Csantucci, on 24 January 2022 - 03:19 AM, said:

I have performed a comparison test using TechPowerUP GPU-Z 2.44.0 (Steve, thanks for the hint).

While the conclusion is probably correct, please stop using third party tools to measure this. In most cases, they'll only measure the total GPU memory usage, which is the wrong thing to measure on all Windows versions since Vista, and if they do measure the per-process usage it will either be a guess (Windows versions before 10) or identical to Open Rails (Windows 10 and later).

 Csantucci, on 24 January 2022 - 03:19 AM, said:

What I am wondering is only why there don't seem to be graphic malfunctions, even when the VRAM tops its max value.

Since Windows Vista, the Operating System can page GPU memory in and out the same way it does for CPU memory, which makes the "VRAM usage" in tools like GPU-Z a bad measurement for such experiments, as you saw here where it was capped.

You should be able to see this effect in the numbers we have built-in to Open Rails: GPU memory "committed" will increase beyond the limit of the graphics card, but "dedicated" will not. This is a similar difference to the CPU memory "private" and "private working set" numbers.

[Csantucci]

Hi James,
the problem is that my old computer, although Win10, does not display VRAM data (as I have already written somewhere else), even with your feature. So for me the only way to test is using external tools.

I thank you about the Video memory paging info, which was new to me.
If the committed GPU memory is even greater than a GB in my case, this indicates that the benefit of the code change is even bigger.

[James Ross]

 Csantucci, on 24 January 2022 - 05:57 AM, said:

the problem is that my old computer, although Win10, does not display VRAM data (as I have already written somewhere else), even with your feature. So for me the only way to test is using external tools.

In this case, because the differences are large enough, we can reach the same conclusion with your data, but in general we cannot trust data from external tools. I know that sucks, but there's a very real reason. :(

[ExRail]

 Csantucci, on 24 January 2022 - 05:57 AM, said:

Hi James,
the problem is that my old computer, although Win10, does not display VRAM data (as I have already written somewhere else), even with your feature. So for me the only way to test is using external tools.

Don't know if this has any use to your subject, I just jumped the gun on windows 7 not displaying Vram, I properly only read 1/2.
Just one more App to compare against.

GPU Shark on windows 7-64 it's using 'GL_NVX_gpu_memory_info' it seams.

GPU Shark: 342 MB Desktop
GPU Shark: 850 MB Open Rails 1.4-120 - Route Rattlesnake - no 4K(68MB) skybox
GPU Shark: 936 MB Open Rails 1.4-120 - Route Rattlesnake - With 4K(68MB) skybox
GPU Shark: 2736 MB GTA-V's Video settings report 2496MB in use out of 3071MB

[James Ross]

 ExRail, on 24 January 2022 - 12:11 PM, said:

Don't know if this has any use to your subject, I just jumped the gun on windows 7 not displaying Vram, I properly only read 1/2.
Just one more App to compare against.

GPU Shark on windows 7-64 it's using 'GL_NVX_gpu_memory_info' it seams.

GPU Shark: 342 MB Desktop
GPU Shark: 850 MB Open Rails 1.4-120 - Route Rattlesnake - no 4K(68MB) skybox
GPU Shark: 936 MB Open Rails 1.4-120 - Route Rattlesnake - With 4K(68MB) skybox
GPU Shark: 2736 MB GTA-V's Video settings report 2496MB in use out of 3071MB

Thanks, but this has the same issues as GPU-Z for this particular case - it is global (rather than per-process) and does not take into consideration the paging of modern GPUs.

[Eldorado.Railroad]

 Csantucci, on 24 January 2022 - 05:57 AM, said:

Hi James,
the problem is that my old computer, although Win10, does not display VRAM data (as I have already written somewhere else), even with your feature. So for me the only way to test is using external tools.

I thank you about the Video memory paging info, which was new to me.
If the committed GPU memory is even greater than a GB in my case, this indicates that the benefit of the code change is even bigger.

Carlo,

Great news and thank you for using an alternate tool. I would continue to use it and others tools like it. When the diagnostics provided do not work with your setup something has been overlooked. I am sure we are not alone in this dilemma.

So regardless of GPU paging to system ram, your test with GPU-Z proves that for the moving camera, you saw some evidence of GPU memory being managed. To stress again, this is for a MOVING camera, a camera that travels from one tile to the next. The test that I posted was for a STATIC, or fixed camera that stays on one location throughout the activity. In my post, I saw no evidence that anything was being managed, which is the same effect that you saw when "reduced memory usage" was disabled. I am wondering if you have a test setup that can mimic my test.

The following is directed at the naysayers for using external testing tools. Until a tool can be crafted that works with all currently supported O/S in OR, ignoring my results and discarding/discounting my hard work offers no proof that my methods are invalid. It is preferable to try other tools to verify that something is working than to accept on faith that is does what it claims. This is not the first time or the last that something has been coded that does not work across the board and it should be accepted that users will try to find a way around a problem. Indeed, GPU paging to system ram is proof that this external tool works very well, as it demonstrates that GPU memory is not being managed, the capped results. In OR when GPU memory is being managed then paging should not occur, which is what you saw. All it requires is a little common sense to interpret the results correctly. My graph clearly states total system + OR for both CPU memory and GPU memory. The raw data in text form clearly indicates when runactivity.exe starts and ends.

Thanks for your help Carlo and I think Serana has solved the GPU management for at least the moving camera.

Steve

[James Ross]

 Eldorado.Railroad, on 24 January 2022 - 03:58 PM, said:

Until a tool can be crafted that works with all currently supported O/S in OR, ignoring my results and discarding/discounting my hard work offers no proof that my methods are invalid.

I have already explained why it is impossible to craft a tool that works on all currently supported OSes, and why that includes external tools. If it was possible, we'd have done it inside Open Rails.

You are all welcome to use whatever tools you like when investigating by yourself, but when you are sharing on the forum or comparing with others the only valid numbers are those provided by Open Rails.

[alteo80]

good evening, I am happy to see that the opening of this post has generated so many answers and help. I have noted that the only measures taken into account would be those from openrails, nevertheless, I would like to share these figures from msi afterburner, concerning the video memory, 7800mb of use without the option checked, and with the memory reduction option, 2420mb. there is therefore an effect, however, it sometimes causes a crash with a dxsharp type message etc... When you change views.... Finally there is a real problem with the rendering process...But that's another subject. Good luck to all and thank you.