Here are some benchmark numbers running on my two GPU workstations for our own references.

One is named Kite: Intel Xeon Silver 4116 @2.10 GHz (HT), 256 GB ram, 4x Nvidia GeForce RTX 2080 Ti (founder edition).

The other one is named Hawk: Intel Xeon ES-2620 v4 @2.10 GHz (HT), 128 GB ram, 2x Nvidia GeForce GTX 1080 Ti (game card edition, not founder edition) .

The benchmark dataset is EMD2660 suggested by RELION’s wiki, and the script we used is:

mpirun -n 3 `which relion_refine_mpi` --i EMD_2660.star 
--ref emd_2660.map:mrc --firstiter_cc --ini_high 60 
--ctf --ctf_corrected_ref --iter 25 --tau2_fudge 4 
--particle_diameter 360 --K 6 --flatten_solvent 
--zero_mask --oversampling 1 --healpix_order 2 
--offset_range 5 --offset_step 2 --sym C1 --norm 
--scale --random_seed 0 --o class3d_2 --pool 100  
--gpu --j 6 #--scratch_dir /ssd/relion

3 variables are: n (MPI), j (thread) and scratch_dir (used or not). Looks like 2080 Ti is running much faster than 1080 Ti (both are 11GB ram) when all variables are identical (n=3,j=2, ssd=n). Green and blue benchmarks come from Hawk and Kite, respectively.