Tuning CPU training performance

[hacksparr0w]

Hello,

I'm trying to utilize a 36-CPU node to train my DeepMD model, but the performance gains I'm getting compared to a 12-CPU node are quite poor, in my opinion.

I'm using the DeepMD-kit Docker image deepmodeling/deepmd-kit:2.1.1_cpu to train my model on both machines.

When running the training process on the 12-CPU node, I'm using the following command:

OMP_NUM_THREADS=12 TF_INTRA_OP_PARALLELISM_THREADS=24 TF_INTER_OP_PARALLELISM_THREADS=12 dp train input.json

Using this setup, I can get the following timings:

DEEPMD INFO    batch    3000 training time 47.94 s, testing time 0.04 s

When running the training process on the 36-CPU node, I'm using the the following command:

OMP_NUM_THREADS=2 TF_INTRA_OP_PARALLELISM_THREADS=36 TF_INTER_OP_PARALLELISM_THREADS=6 dp train input.json

The best training times I'm able to get with this setup are around 33.14 s.

I've followed this document while trying to optimize the OMP_NUM_THREADS, TF_INTRA_OP_PARALLELISM_THREADS and TF_INTER_OP_PARALLELISM_THREADS parameters, but have also done some empirical measurements to arrive at the exact numbers I'm currently using. I've also tried using horovod, but haven't had much luck with it (as far as I understand, horovod is meant to be used in multi-node setups).

I'd like to ask whether I have some more options to try when tuning the performance. I hoped for much lower training times on the 36-CPU node, but perhaps my expectations for this were inaccurate?

The following is the output of lscpu on the 36-CPU node:

Architecture:                    x86_64
CPU op-mode(s):                  32-bit, 64-bit
Byte Order:                      Little Endian
Address sizes:                   46 bits physical, 48 bits virtual
CPU(s):                          36
On-line CPU(s) list:             0-35
Thread(s) per core:              1
Core(s) per socket:              18
Socket(s):                       2
NUMA node(s):                    2
Vendor ID:                       GenuineIntel
CPU family:                      6
Model:                           85
Model name:                      Intel(R) Xeon(R) Gold 6240 CPU @ 2.60GHz
Stepping:                        7
Frequency boost:                 enabled
CPU MHz:                         999.367
CPU max MHz:                     2601.0000
CPU min MHz:                     1000.0000
BogoMIPS:                        5200.00
Virtualization:                  VT-x
L1d cache:                       1.1 MiB
L1i cache:                       1.1 MiB
L2 cache:                        36 MiB
L3 cache:                        49.5 MiB
NUMA node0 CPU(s):               0-17
NUMA node1 CPU(s):               18-35
Vulnerability Itlb multihit:     KVM: Vulnerable
Vulnerability L1tf:              Not affected
Vulnerability Mds:               Not affected
Vulnerability Meltdown:          Not affected
Vulnerability Spec store bypass: Vulnerable
Vulnerability Spectre v1:        Vulnerable: __user pointer sanitization and usercopy barriers only; no swapgs barriers
Vulnerability Spectre v2:        Vulnerable, IBPB: disabled, STIBP: disabled
Vulnerability Srbds:             Not affected
Vulnerability Tsx async abort:   Mitigation; TSX disabled
Flags:                           fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm constant_tsc art arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid dca sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm abm 3dnowprefetch cpuid_fault epb cat_l3 cdp_l3 invpcid_single intel_ppin ssbd mba ibrs ibpb stibp ibrs_enhanced tpr_shadow vnmi flexpriority ept vpid ept_ad fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid cqm mpx rdt_a avx512f avx512dq rdseed adx smap clflushopt clwb intel_pt avx512cd avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local dtherm ida arat pln pts hwp hwp_act_window hwp_epp hwp_pkg_req pku ospke avx512_vnni md_clear flush_l1d arch_capabilities

This is the output of lscpu on the 12-CPU node:

Architecture:            x86_64
  CPU op-mode(s):        32-bit, 64-bit
  Address sizes:         43 bits physical, 48 bits virtual
  Byte Order:            Little Endian
CPU(s):                  24
  On-line CPU(s) list:   0-23
Vendor ID:               AuthenticAMD
  Model name:            AMD Ryzen 9 3900X 12-Core Processor
    CPU family:          23
    Model:               113
    Thread(s) per core:  2
    Core(s) per socket:  12
    Socket(s):           1
    Stepping:            0
    Frequency boost:     enabled
    CPU(s) scaling MHz:  49%
    CPU max MHz:         4672.0698
    CPU min MHz:         2200.0000
    BogoMIPS:            7586.18
    Flags:               fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm constant_tsc rep_good
                          nopl nonstop_tsc cpuid extd_apicid aperfmperf rapl pni pclmulqdq monitor ssse3 fma cx16 sse4_1 sse4_2 movbe popcnt aes xsave avx f16c rdrand lahf_lm cmp_legacy svm ex
                         tapic cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ibs skinit wdt tce topoext perfctr_core perfctr_nb bpext perfctr_llc mwaitx cpb cat_l3 cdp_l3 hw_pstate ssbd 
                         mba ibpb stibp vmmcall fsgsbase bmi1 avx2 smep bmi2 cqm rdt_a rdseed adx smap clflushopt clwb sha_ni xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total
                          cqm_mbm_local clzero irperf xsaveerptr rdpru wbnoinvd arat npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold avic v_v
                         msave_vmload vgif v_spec_ctrl umip rdpid overflow_recov succor smca sev sev_es
Virtualization features: 
  Virtualization:        AMD-V
Caches (sum of all):     
  L1d:                   384 KiB (12 instances)
  L1i:                   384 KiB (12 instances)
  L2:                    6 MiB (12 instances)
  L3:                    64 MiB (4 instances)
NUMA:                    
  NUMA node(s):          1
  NUMA node0 CPU(s):     0-23
Vulnerabilities:         
  Itlb multihit:         Not affected
  L1tf:                  Not affected
  Mds:                   Not affected
  Meltdown:              Not affected
  Mmio stale data:       Not affected
  Retbleed:              Mitigation; untrained return thunk; SMT enabled with STIBP protection
  Spec store bypass:     Mitigation; Speculative Store Bypass disabled via prctl
  Spectre v1:            Mitigation; usercopy/swapgs barriers and __user pointer sanitization
  Spectre v2:            Mitigation; Retpolines, IBPB conditional, STIBP always-on, RSB filling, PBRSB-eIBRS Not affected
  Srbds:                 Not affected
  Tsx async abort:       Not affected

I'm also including the configuration of my model:

{
  "model": {
    "type_map": ["Si"],
    "descriptor": {
      "type": "se_e2_a",
      "sel": [70],
      "rcut_smth": 1.8,
      "rcut": 6.0,
      "neuron": [25, 50, 100],
      "resnet_dt": false,
      "seed": 1
    },
    "fitting_net": {
      "neuron": [100, 100, 100],
      "resnet_dt": true,
      "seed": 1
    }
  },
  "learning_rate": {
    "type": "exp",
    "decay_steps": 20000,
    "start_lr": 0.001,
    "stop_lr": 1e-8
  },
  "loss": {
    "type": "ener",
    "start_pref_e": 10,
    "limit_pref_e": 500,
    "start_pref_f": 1000,
    "limit_pref_f": 1,
    "start_pref_v": 0,
    "limit_pref_v": 0
  },
  "training": {
    "training_data": {
      "systems": [
        "../../dataset/si/training/a-2.25",
        "../../dataset/si/training/a-2.29",
        "../../dataset/si/training/a-2.35",
        "../../dataset/si/training/c-0.99",
        "../../dataset/si/training/c-1.01",
        "../../dataset/si/training/c-1.03"
      ],
      "batch_size": "auto:32"
    },
    "validation_data": {
      "systems": [
        "../../dataset/si/validation/a-2.25",
        "../../dataset/si/validation/a-2.29",
        "../../dataset/si/validation/a-2.35",
        "../../dataset/si/validation/c-0.99",
        "../../dataset/si/validation/c-1.01",
        "../../dataset/si/validation/c-1.03"
      ],
      "batch_size": "auto",
      "numb_btch": 1
    },
    "numb_steps": 2000000,
    "seed": 10,
    "disp_file": "l_curve.csv",
    "disp_freq": 1000,
    "save_freq": 1000
  }
}

Any comments would be highly appreciated.

[njzjz]

You can use enable_profiler to enable TensorFlow profiler and compare performance on two machines.