14. On the “Compute” tab, set “Use GPU acceleration? Yes”
since the program will run a lot faster on GPUs than on CPUs.
15. On the “Compute” tab, adjust “Which GPUs to use:” to your
computer (see Note 10). On our computer with four cards, we
used “0,1:2,3”, where the “:” sign is used to divide tasks
among different MPI workers and the “,” sign is used to divide
tasks within a single MPI worker. We used three MPI processes
(see step 16), i.e., one master and two workers. Therefore,
GPUs 0 and 1 jointly do the work of the first worker, and
GPUs 2 and 3 divide the work of the second worker.
16. On the “Running” tab, we set “Number of MPI procs: 3” to
divide the calculations over one master and two worker processes (see Note 11). Provided your computer has enough
RAM, you could increase the number of MPI processes (see
Note 12).
17. On the “Running” tab, we set “Number of threads: 12” to
make optimal use of the 24 (hyper-threaded) cores on our
computer (see Note 13).
18. On the “Running” tab, we set “Submit to queue? No”, as we
were performing our calculation on a local computer. You can
set this option to “Yes” to submit your job to a cluster queuing
system instead. See the RELION wiki (www3.mrc-lmb.cam.
ac.uk/relion) for details on how to set this up for your
system.
19. Finally, press the “Run!” button to execute the calculation. On
our computer, this job took approximately 14 h.
3.3 Analyze the Body
Reconstructions
1. The multi-body refinement program writes out a reconstructed
map that is filtered to its estimated resolution for each body as
“Multibody/job001/run_body00[1--4].mrc”.
These
maps can be directly visualized in a molecular volume viewer,
or in slices from the “Display!” pull-down menu on the GUI
(Fig. 2; also see Note 14).
2. Besides the filtered maps, the program also writes out unfiltered maps for both independently refined half-sets. On the
“I/O” tab of the “Post-processing” job-type, input the first
half-map of the first body:“One of the 2 unfiltered halfmaps: Multibody/job001/run_half1_body001_unfil.
mrc”. Also set “Solvent mask: Mask-and-Ref/CORE_mask.mrc” and “Calibrated pixel size (A) 1.699”.
3. On the “Sharpen” tab, one can provide a STAR file with the
modulation transfer function of the detector (see Note 15).
4. Leave all other options on the “Sharpen”, “Filter”, and
“Running” tabs to their defaults and execute the calculation by
pressing the “Run!” button (see Note 16).
152
Takanori Nakane and Sjors H. W. Scheres
since the program will run a lot faster on GPUs than on CPUs.
15. On the “Compute” tab, adjust “Which GPUs to use:” to your
computer (see Note 10). On our computer with four cards, we
used “0,1:2,3”, where the “:” sign is used to divide tasks
among different MPI workers and the “,” sign is used to divide
tasks within a single MPI worker. We used three MPI processes
(see step 16), i.e., one master and two workers. Therefore,
GPUs 0 and 1 jointly do the work of the first worker, and
GPUs 2 and 3 divide the work of the second worker.
16. On the “Running” tab, we set “Number of MPI procs: 3” to
divide the calculations over one master and two worker processes (see Note 11). Provided your computer has enough
RAM, you could increase the number of MPI processes (see
Note 12).
17. On the “Running” tab, we set “Number of threads: 12” to
make optimal use of the 24 (hyper-threaded) cores on our
computer (see Note 13).
18. On the “Running” tab, we set “Submit to queue? No”, as we
were performing our calculation on a local computer. You can
set this option to “Yes” to submit your job to a cluster queuing
system instead. See the RELION wiki (www3.mrc-lmb.cam.
ac.uk/relion) for details on how to set this up for your
system.
19. Finally, press the “Run!” button to execute the calculation. On
our computer, this job took approximately 14 h.
3.3 Analyze the Body
Reconstructions
1. The multi-body refinement program writes out a reconstructed
map that is filtered to its estimated resolution for each body as
“Multibody/job001/run_body00[1--4].mrc”.
These
maps can be directly visualized in a molecular volume viewer,
or in slices from the “Display!” pull-down menu on the GUI
(Fig. 2; also see Note 14).
2. Besides the filtered maps, the program also writes out unfiltered maps for both independently refined half-sets. On the
“I/O” tab of the “Post-processing” job-type, input the first
half-map of the first body:“One of the 2 unfiltered halfmaps: Multibody/job001/run_half1_body001_unfil.
mrc”. Also set “Solvent mask: Mask-and-Ref/CORE_mask.mrc” and “Calibrated pixel size (A) 1.699”.
3. On the “Sharpen” tab, one can provide a STAR file with the
modulation transfer function of the detector (see Note 15).
4. Leave all other options on the “Sharpen”, “Filter”, and
“Running” tabs to their defaults and execute the calculation by
pressing the “Run!” button (see Note 16).
152
Takanori Nakane and Sjors H. W. Scheres