Closed yhgon closed 2 years ago
another simulation result show below.
#"Progress (%)","Step","Potential Energy (kJ/mole)","Kinetic Energy (kJ/mole)","Total Energy (kJ/mole)","Temperature (K)","Speed (ns/day)","Elapsed Time (s)","Time Remaining"
0.1%,100,-121370.49897545716,49133.63804504863,-72236.86093040853,193.83086050003715,0,0.00028634071350097656,--
0.1%,200,-118733.46772545716,55663.14637669065,-63070.321348766505,219.58959258097806,144,0.12048935890197754,4:00
0.1%,300,-116580.84272545716,60875.83453027613,-55705.00819518103,240.1535050150919,145,0.23886895179748535,3:58
0.2%,400,-113773.78022545716,63534.93906139524,-50238.841164061916,250.6435997838489,146,0.3542344570159912,3:55
0.2%,500,-112449.37397545716,66616.5959524457,-45832.77802301146,262.80065207479714,147,0.4690840244293213,3:53
0.3%,600,-111380.03022545716,69195.02884695033,-42185.00137850683,272.9724994398387,148,0.5849928855895996,3:53
0.3%,700,-109970.49897545716,71040.78516078438,-38929.71381467278,280.2539576997748,148,0.6993200778961182,3:52
0.4%,800,-109522.93647545716,72284.3991101383,-37238.53736531886,285.1599807732568,149,0.8135404586791992,3:51
0.5%,900,-108419.09272545716,72012.97321176575,-36406.1195136914,284.08921301542637,149,0.9291989803314209,3:51
0.5%,1000,-108747.96772545716,73945.91190383676,-34802.05582162039,291.71460337700364,119,1.3119151592254639,4:50
0.6%,1100,-108387.65522545716,74383.8260657452,-34003.82915971195,293.4421627885425,121,1.4302194118499756,4:44
0.6%,1200,-108362.46772545716,75495.17901755986,-32867.2887078973,297.8264198381016,123,1.5439362525939941,4:39
0.7%,1300,-107781.31147545716,75015.68742250348,-32765.624052953674,295.93483858806064,125,1.6578710079193115,4:34
0.7%,1400,-108697.03022545716,76548.92161350226,-32148.1086119549,301.98340560678804,127,1.772719383239746,4:30
it seems that inital value is not memset(0)
and only X and Y part is updated in the value XXXXX.YYYYYZZZZZZZ
5 digists of Y is related with single precision.
-121370.49897545716,
-118733.46772545716,
That's because single precision only has about eight significant digits. The energy gets converted to double precision when it's returned, but any additional digits beyond that are meaningless.
I found odd report what Potential Energy mantisa is same during whole iteration.
X..806077383
as below :test code is same as openMM tutorial except I use CUDA single precision.