I checked the gpu usage with gpu-z. On my GTX1060 it is around 50% and memory used below 2 GB.
So it should make sense to run 2 wu's on that card. Is there a setup available?
And another question: for Einstein it makes sense to run processlasso to priorize the cpu usage for the gpu-wu's. Increases the gpu output pretty well since the gpu is not long in idle waiting for the cpu. The wu's are set up as to use 0.991 cpus. I assume this is a dummy value. Maybe the output can be increased here as well with processlasso.
Unfortunately it's no longer free of charge, but maybe someone has a free version available and can check this. If it works i will pay the 38 Euros for it.

Well I would be happy to receive any GPU tasks at all for my NVIDIA GPU.
But it says all the time 'got 0 new tasks' :(

So it should make sense to run 2 wu's on that card. Is there a setup available?

Have you tried app_config? Something like:

<app_config>
       <app>
           <name>mlds-gpu</name>
           <max_concurrent>2</max_concurrent>
           <gpu_gpu_usage>0.500000</gpu_gpu_usage>
           <gpu_cpu_usage>1.000000</gpu_cpu_usage>
       </app>
   </app_config>

And another question: for Einstein it makes sense to run processlasso to priorize the cpu usage for the gpu-wu's. Increases the gpu output pretty well since the gpu is not long in idle waiting for the cpu.

The BOINC client does this by default (on Windows at least) unless you turn it off - CPU only tasks run at base priority 'Low' and GPU/CPU tasks run at base priority 'Below normal'. See https://boinc.berkeley.edu/wiki/Client_configuration <process_priority> and <process_priority_special>

https://boinc.berkeley.edu/trac/wiki/ClientAppConfig

<app_config>
	<app>
		<name>mlds-gpu</name>
		<gpu_versions>
			<gpu_usage>0.50</gpu_usage>
			<cpu_usage>0.50</cpu_usage>
		</gpu_versions>		
	</app>
</app_config>

Big Thx for the responds, it's working now. GPU-usage has increased to 66% avg, 71% peak. More intresting is that the memory usage has not really increased, is always at 2012 MB. WU's finish and validate. Of coarse they take longer time, but there is also twice the output. Needs some time to compare the performance.
If someone wants to follow the results, it's cpu-ID: 5172

Two WUs are just locking each other on my 750Ti with 2Gb of video memory . Seems as not enough VRAM.

I forgot MLC gpu app uses a full cpu thread.

Try this app_config.xml too.

<app_config>
	<app>
		<name>mlds-gpu</name>
		<gpu_versions>
			<gpu_usage>0.50</gpu_usage>
			<cpu_usage>1.00</cpu_usage>
		</gpu_versions>		
	</app>
</app_config>

looks good so far, credit increased. I have two systems running with 2 wu's on gpu. Not more errors than usual.
I am curious how a GTX3060ti would perform. 4 wu's (8 GB Ram) and > 19 TFlops ...