Cool VL Viewer forum

A few hints for dedicated AMD GPUs on Windows and good settings i found working well.

Use a recent driver version, at least "AMD Software: Adrenalin Edition 22.7.1." which added massive improvements to OpenGL performance. Newer tends to be better.

Enable "OpenGL Triple Buffering" and disable VSync (called "Wait for vertical refresh"):


Don't bother with HBCC Memory Segment, its just slower, keep it disabled.

If your GPU is new enough (RX 6xxx) and your motherboard supports it, you should enable the "PCIe Resizeable BAR" support. This usually takes some form of BIOS Setting that must be enabled, and a setting in the AMD driver, where it is called "Smart Access Memory".



If you have a GPU that is too old (e.g. RX Vega), you may tweak some registry setting to still allow it to be activated. This thread has some hints how to do it:
https://forums.guru3d.com/threads/resizeable-bar-support-issues.435389/

Thank you for the hints, Kathrine.

Is there a setting in AMD Windows drivers to enable threaded OpenGL rendering, like there is for NVIDIA ?

If the answer is yes, then this setting should be enabled as well for best performances.

Also, and for all GPU brands, as long as you are running the viewer on a desktop computer (i.e. when the power consumption is not an issue), you should consider using the "performances" mode of the driver, where the GPU and VRAM frequencies will be kept to the max: it helps getting faster rezzing when you enable the viewer smart frame limiting feature, at which point the driver could otherwise consider the load is low enough to lower the frequencies, causing the CPU to wait longer for the GPU to finish rendering each frame, and thus getting less remaining "free" time (i.e. the remaining time within the available the time slot corresponding to the voluntary frame limitation) to use for processing network messages and data and rezzing stuff. It also avoids delays in the frequencies stepping up when moving from a render-light scene/place to a render-heavy one, that could otherwise cause small frame "hiccups"/dropouts. For the same reasons, the CPU governor (Linux) or performance mode (Windows) should be kept on "performances".

The newer AMD drivers use multiple threads internally, the older ones do not, so no, there is no explicit setting.

And yes, sure, you can tweak the driver settings for more performance still, depending on the exact card / cooling / power supply. Be it undervolting, overclocking, performance settings.What exactly stable and workeable depends on the card. Some older models like Vega have a lot of undervolting/overclocking headroom, while others are already close to the maximum.

I was not speaking about overclocking, but just about the performance profile set for the driver, since I remarked on your screenshot that this was set to "Power saver" in your driver, with extremely low (minimal) frequencies for VRAM and GPU.

Ah, no, its not set to power safer there. Its actually set to "performance" but thats outside the right edge of the screenshot.

The frequencies listed on the screenshots are with idle desktop, not with the viewer running.

Well, with NVIDIA, even when idle (rendering just the desktop, that in my case does not make any use of the GPU since I do not even use a compositor), when in "performance" mode, the frequencies are (almost, for the GPU, fully for the VRAM) maxed out...

Performances:
Adaptive: