Andreas Tillack

@kma Hello, this is usually caused by compiling for the wrong compute architecture (by default we compile for 52, 60, 61, and 70 and higher) - try compiling for your...

@parazue Two things are going on in the output you posted: the map file for the "A" atom type cannot be opened and the pMem_fgrids upload error. The first one...

@parazue Thank you for posting the potential buffer overflow. Although I have an idea what potentially could be the culprit (which could be part of code already modified in PR...

@parazue PR 139 is merged and released as v1.5. This very likely fixes your issue.

@miaohao0309 For A100 you will need to add `80` to the Cuda compilation targets. You can do this either by editing the Makefile or with your build command: `make DEVICE=CUDA...

@tomaszborowski Your card ran out of memory allocating the PRNG device memory. This is a Cuda version issue as it preallocates this memory and errs on the side of requesting...

@tomaszborowski By default, OpenCL AD-GPU is set to use the first device (counting starts at 1) on OpenCL platform 0 - if this fails (which the error message is about)...

@davidyanglee Nvidia keeps a list here: https://developer.nvidia.com/cuda-gpus#compute

Hi @Leo-T-Zang, while AD-GPU uses AD4's scoring function it uses an improved local search algorithm (Adadelta) by default and even Solis-Wets (SW, which is what AD4 uses) in AD-GPU should...

@L30nardoSV Those synchronization points (along with thread fences - which i did remove after thorough testing) where put in there during the later stages of bug hunting and since they...