Modern graphics hardware offers peak performances close to 1 Tflop\/s, and NVIDIA’s CUDA provides a flexible and convenient programming interface to exploit these immense computing resources. We demonstrate the ability of GPUs to perform high-precision molecular dynamics simulations for nearly a million particles running stably over many days. Particular emphasis is put on the numerical long-time stability in terms of energy and momentum conservation. Floating point precision is a crucial issue here, and sufficient precision is maintained by double-single emulation of the floating point arithmetic. As a demanding test case, we have reproduced the slow dynamics of a binary Lennard-Jones mixture close to the glass transition. The improved numerical accuracy permits us to follow the relaxation dynamics of a large system over 4 non-trivial decades in time. Further, our data provide evidence for a negative power-law decay of the velocity autocorrelation function with exponent 5\/2 in the close vicinity of the transition. The performance of our MD simulation package on the GPU shows speedup factors of up to 80 compared to a single processor core, and a single GPU is found to compare with the LAMMPS package on 64 distributed cores.<\/p>\n","protected":false},"excerpt":{"rendered":"
Modern graphics hardware offers peak performances close to 1 Tflop\/s, and NVIDIA’s CUDA provides a flexible and convenient programming interface to exploit these immense computing resources. We demonstrate the ability of GPUs to perform high-precision molecular dynamics simulations for nearly a million particles running stably over many days. Particular emphasis is put on the numerical […]<\/p>\n","protected":false},"author":351,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_jetpack_memberships_contains_paid_content":false,"footnotes":"","jetpack_publicize_message":"","jetpack_publicize_feature_enabled":true,"jetpack_social_post_already_shared":false,"jetpack_social_options":{"image_generator_settings":{"template":"highway","default_image_id":0,"font":"","enabled":false},"version":2}},"categories":[89,3,12],"tags":[98,14,20,234,176,1783,358],"class_list":["post-1294","post","type-post","status-publish","format-standard","hentry","category-nvidia-cuda","category-paper","category-physics","tag-computational-physics","tag-cuda","tag-nvidia","tag-nvidia-geforce-gtx-280","tag-package","tag-physics","tag-soft-condensed-matter"],"views":2565,"jetpack_publicize_connections":[],"jetpack_featured_media_url":"","jetpack_sharing_enabled":true,"_links":{"self":[{"href":"https:\/\/hgpu.org\/index.php?rest_route=\/wp\/v2\/posts\/1294","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/hgpu.org\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/hgpu.org\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/hgpu.org\/index.php?rest_route=\/wp\/v2\/users\/351"}],"replies":[{"embeddable":true,"href":"https:\/\/hgpu.org\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=1294"}],"version-history":[{"count":0,"href":"https:\/\/hgpu.org\/index.php?rest_route=\/wp\/v2\/posts\/1294\/revisions"}],"wp:attachment":[{"href":"https:\/\/hgpu.org\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=1294"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/hgpu.org\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=1294"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/hgpu.org\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=1294"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}