VRay rendering slower on a new computer with DirectX display

[Emmanuel]

I have been using an i7-5820K computer with 16Gb DDR4 RAM, runing Windows 8.1 for 8 years.
It's time to upgrade !

My new machine is an i7-13700K, with 32Gb DDR5, now running Windows 11 Pro 64. I am still using an old graphic card though because I don't play games any more and trueSpace does not make full use of the new generation graphic cores.

This new CPU is supposed to be 2.5 to 3x faster according to benchmarks though. Let's try it !

But what a surprise in trueSpace : when rendering a VRay scene on Modeler side with DirectX display mode, the new computer takes 37 seconds to complete the render while the old one took only 29 seconds ! WTF !
Checking the ressources monitor I can see that the CPUs are below 20% load !

But if I switch the Modeler window display to "no hardware acceleration" instead of Direct X or OpenGL, it works much faster : now only 8 seconds instead of 23. A ratio that reflects the performance upgrade of the new CPU.

I already reported on the forum that, on recent computer, we should use a "no hardware accelerated" window to launch VRay renders from. On the i7 5th generation the difference was already noticable. But now it is clear that the DirectX view ruins the new CPUs rendering performances !

I wonder how this is possible ? Why is the real-time display engine affecting the CPU rendering ? Does the graphic card behaves like a bottleneck ?

[bitkar]

i just tried on Win10 and with some scenes i did not get 100% CPU usage, but i got it the other way around - with DX on it was must faster :D ... I tried another scene and i got 100% on both cases and DX still faster (23s vs 26s)

:D there is something odd

[Emmanuel]

Thanks for the feedback.
On my "old" i7 920 I remember that VRay rendering from DirectX window was faster.

What is your CPU ? Your graphic card ?

[trueBlue]

I have been using an i7-5820K computer with 16Gb DDR4 RAM, runing Windows 8.1 for 8 years.
It's time to upgrade !

My new machine is an i7-13700K, with 32Gb DDR5, now running Windows 11 Pro 64. I am still using an old graphic card though because I don't play games any more and trueSpace does not make full use of the new generation graphic cores.

This new CPU is supposed to be 2.5 to 3x faster according to benchmarks though. Let's try it !

But what a surprise in trueSpace : when rendering a VRay scene on Modeler side with DirectX display mode, the new computer takes 37 seconds to complete the render while the old one took only 29 seconds ! WTF !
Checking the ressources monitor I can see that the CPUs are below 20% load !

But if I switch the Modeler window display to "no hardware acceleration" instead of Direct X or OpenGL, it works much faster : now only 8 seconds instead of 23. A ratio that reflects the performance upgrade of the new CPU.

I already reported on the forum that, on recent computer, we should use a "no hardware accelerated" window to launch VRay renders from. On the i7 5th generation the difference was already noticable. But now it is clear that the DirectX view ruins the new CPUs rendering performances !

I wonder how this is possible ? Why is the real-time display engine affecting the CPU rendering ? Does the graphic card behaves like a bottleneck ?

Where is the "no hardware acceleration" option?

Look at the PlayerD3D.log
Does it have something like:
Device manager R_DeviceManager : Direct3DX SDK version 27 is suitable for this version of DirectX
Device manager R_DeviceManager : Using driver level version 900
Device manager R_DeviceManager : Maximum vertex shader profile is vs_3_0
Device manager R_DeviceManager : Maximum pixel shader profile is ps_3_0
Device manager R_DeviceManager : Approximate available video memory is 4293918720 MB

You can see exactly how long it takes VRay to render in the Vray.log:
LOG: Rendering Time: 0:00:10, RsRenderer::render(1144)

[Emmanuel]

Where is the "no hardware acceleration" option?

In modeler you can choose between 3 API display modes : OpenGL, DirectX, and Wireframe mode. The wireframe mode is using no graphic card acceleration.
The modelers' realtime display mode affects render speed in different ways, depending on the CPU generation.

[trueBlue]

Thanks
For me, Wireframe Display is 9 seconds slower than Direct 3D Solid Render Display
Compare.scn
Render Scene in Model only view
Direct 3D Solid Render Display
LOG: Rendering Time: 0:01:00, RsRenderer::render(1144)

Render Scene in Model only view
Wireframe Display
LOG: Rendering Time: 0:01:09, RsRenderer::render(1144)

Render Scene in Model view and Floating Workspace View
Direct 3D Solid Render Display
LOG: Rendering Time: 0:01:00, RsRenderer::render(1144)

Render Scene in Model with Floating Model Camera view active
Direct 3D Solid Render Display
LOG: Rendering Time: 0:00:30, RsRenderer::render(1144)
This setup was the fastest

[Emmanuel]

Interesting !
What is you CPU ?
What is your GPU ?

I am currently using an AMD RX550 4Gb DDR5 but I remember my old Nvidia was providing faster D3D view rendering.
Possible that AMD chips/drivers are not good for trueSpace.

[trueBlue]

Intel(R) Core(TM) i7-6700K CPU @ 4.00GHz 4.00 GHz
64-bit operating system, x64-based processor
Windows 10 Home

NVIDIA GeForce GTX 970
4096 MB GDDR5

Read the PlayerD3D.log
You should be able to see if there is any issues

Also, I vaguely remember having to install DX9 even though I have DX12 when I first got this PC

[bitkar]

Thanks for the feedback.
On my "old" i7 920 I remember that VRay rendering from DirectX window was faster.

What is your CPU ? Your graphic card ?

Ryzen 7 1700x + Radeon RX580 8G

[Emmanuel]

I've installed an old Nvidia 8600GTX with only 256Mb RAM and DX mode is faster now.
So this issue is GPU drivers relative !

I also noticed that OpenGL rendering is always very slow as soon as the GPU drivers are loaded.
If Windows is using default display drivers, trueSpace's OpenGL runs fine.

We can figure that trueSpace's VRay rendering is using the CPU but through the graphic card drivers ! So the graphic card driver is somehow involved in the computation. Note that VRay performance will not speed up with faster GPU. Graphic chip or graphic RAM does not affect trueSpace render ; only the driver code does !

Pretty interesting...