What is acemd3?
It's the next-generation application that we are deploying. It will supersede the current short and long queues ("acemd2"). The app has several advantages and an entirely new codebase. For one, it runs on RTX cards.
What should I do to receive acemd3-based workunits?
Check in your Project preferences page that "ACEMD3" is enabled. Chances are that it already is.
Where do I find the current list of apps?
Here under the heading "New version of ACEMD"
What driver/card/OS combinations are supported?
Windows, CUDA80 Minimum Driver r367.48 or higher
Linux, CUDA92 Minimum Driver r396.26 or higher
Linux, CUDA100 Minimum Driver r410.48 or higher
Windows, CUDA101 Minimum Driver r418.39 or higher
See also here, but note that the driver/cuda matchings may be approximate. When in doubt, upgrade your driver.
Can I use it on multi-GPU systems?
In general yes, with one caveat: if you have DIFFERENT types of NVIDIA GPUs in the same PC, suspending a job in one and restarting it in the other will NOT be possible (errors on restart). Consider restricting the client to one GPU type only ("exclude_gpu", see here).
I have an RTX card and WUs failures contain this error text "[...] Error compiling program: nvrtc: error: invalid value for --gpu-architecture (-arch) [...]"
You need to upgrade your driver.
I get errors, and the output includes "CUDA_ERROR_UNKNOWN (999)"
You may have a notebook (or older) graphic card. They are not really suited to sustain hard-core computations (thanks anyway).
I get errors during the computation, and I am using Avast AV
You may need to whitelist the app. See here.
I get errors during the computation, and I am overclocking
Even memory errors as small as single-bits, which would be invisible during gaming use, will crash MD calculations. Hence, overclocking is generally a bad idea.
I read messages saying "Detected memory leaks"
Such messages are always present in Windows. They are not related to successful or not termination. If an error message is present, it's elsewhere in the output.