Frequency calculation speedup

Hi, I am doing frequency calculation using wB97xD/def2-svp for a 70 atom unrestricted system. The speed of the calculation around 2 days but no matter how many if I used 16 or 32 or 64 cores, with 150GB memory, it’s similar speed. I also tried to set CPSCF_NSEG, but did not help. What I see is the CPU usage go down to only one core running for hours. Do you guys know what the problem is and how I can improve the performance?

Here is a copy of my sample input file:
$rem
JOBTYPE freq
BASIS = def2-svp
METHOD = WB97X-D
SCF_CONVERGENCE = 8
MEM_TOTAL = 150000
MAX_SCF_CYCLES 250
SCF_ALGORITHM gdm
SCF_GUESS_mix = 5
XC_GRID = 3
SYM_IGNORE TRUE
SYMmetry FALSE
unrestricted true
$end

$molecule
0 1
S 0.1085158 -0.1009982 0.0451984
C 0.0159110 -0.0975664 1.7585852
C 1.2334648 0.0787876 2.4280373
H 2.1554405 0.1670083 1.8479685
C 1.2880050 0.1773946 3.8357305
C 2.5151492 0.3851675 4.5232414
H 3.4390551 0.4651048 3.9459877
C 2.5350206 0.4856165 5.8941591
H 3.4785768 0.6454772 6.4189819
C 1.3327747 0.3824924 6.6303721
H 1.3602090 0.4659237 7.7188104
C 0.1287538 0.1778285 5.9887147
H -0.7968170 0.0979279 6.5630356
C 0.0761860 0.0701455 4.5814068
C -1.1499133 -0.1571685 3.8787397
H -2.0705342 -0.2722399 4.4552018
C -1.1897410 -0.2420038 2.5192377
H -2.1282694 -0.4464410 2.0112102
B -1.6682624 0.4972249 -0.7938201
C -2.3027599 1.5431114 0.2821836
C -3.5707472 1.4289847 0.8516861
F -4.3795972 0.4164937 0.5453360
C -4.0612572 2.3240137 1.8035027
F -5.2680993 2.1574975 2.3219937
C -3.2686283 3.3856619 2.2225842
F -3.7157821 4.2355219 3.1293388
C -1.9960180 3.5438746 1.6794394
F -1.2290972 4.5494143 2.0695161
C -1.5501675 2.6308800 0.7327886
F -0.3200289 2.8239038 0.2513979
C -2.5888607 -0.8008931 -1.1518560
C -2.5322412 -2.0513585 -0.5421251
F -1.6436075 -2.3006234 0.4325386
C -3.3781862 -3.1084234 -0.8637997
F -3.2499605 -4.2804721 -0.2606277
C -4.3536783 -2.9283017 -1.8352236
F -5.1831801 -3.9097324 -2.1388901
C -4.4489631 -1.7010148 -2.4812790
F -5.3660388 -1.5172093 -3.4166358
C -3.5801764 -0.6735881 -2.1284708
F -3.7689264 0.4937592 -2.7429088
C -1.1109184 1.1976004 -2.1662544
C -0.2582993 0.5022779 -3.0225549
F 0.1782142 -0.7115232 -2.6750620
C 0.1713066 0.9813433 -4.2554715
F 0.9604507 0.2470449 -5.0257537
C -0.2577723 2.2345874 -4.6772847
F 0.1332801 2.7176061 -5.8423060
C -1.1061168 2.9733250 -3.8580369
F -1.5303500 4.1620511 -4.2557584
C -1.5205191 2.4481184 -2.6352728
F -2.3718026 3.1979440 -1.9430761
H -0.9408008 -1.8870876 -4.4083750
C -1.6381687 -2.7238333 -4.3980085
C -2.7329665 -2.7453154 -5.2447583
H -2.9060718 -1.9048204 -5.9203995
C -3.6220437 -3.8274207 -5.2374235
H -4.4892193 -3.8297419 -5.8999748
C -3.3950968 -4.9045633 -4.3714050
H -4.0884655 -5.7478172 -4.3603002
C -2.3134443 -4.9020979 -3.5080210
H -2.1512774 -5.7307083 -2.8165604
C -1.4048221 -3.8005738 -3.4877750
C -0.3381137 -3.7698850 -2.6107636
C 0.5433688 -3.6103827 -1.6641256
H 0.2098496 -3.1113033 -0.7365755
C 1.9940917 -4.0145927 -1.7082876
H 2.2525574 -4.4910134 -2.6633385
H 2.6353739 -3.1288705 -1.5755652
H 2.2278233 -4.7154325 -0.8915507
$end

What happens if you get rid of the XC_GRID keyword? We’ve seen some strange timings lately (in recent versions of Q-Chem) related to the quadrature grids. The SG-3 grid shouldn’t be required for the functional that you’ve selected, the default (SG-2) should be sufficient.

By the way, using segments for the CPSCF will slow things down not speed them up, because integrals need to be recomputed for each segment. Use this option only if you run into memory limitations.