Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
To keep a scheduled Python job from launching more native threads than intended, set the thread-limit variables that match its loaded numerical libraries before Python starts, then inspect the libraries and limits from the job’s actual environment. A Python worker count such as n_jobs does not necessarily include the native threads used by BLAS or OpenMP.
Why can a Python job use more CPU cores than n_jobs?
NumPy can call a multithreaded BLAS library for linear algebra. Those native threads come from the linked library, not from Python’s array calls themselves. If an outer layer also starts multiple jobs or processes, each one may create its own native thread pool.
Scikit-learn illustrates the arithmetic with a hypothetical setup: 8 CPUs × 8 joblib jobs × 8 OpenMP threads per job can produce 64 threads. This is an example of oversubscription, not a benchmark or a prediction of every workload. As scikit-learn puts it, “Over-subscription happens when a program is running too many threads at the same time.” Scikit-learn: Parallelism, resource management, and configuration.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Repair Windows errors before they cause bigger problems3Scan for outdated or missing drivers - takes under a minuteOversubscription is more likely when the total concurrent processes and native threads exceed the CPUs allocated to the task. The right balance depends on whether you want parallelism primarily across processes or inside numerical operations; the goal is useful concurrency without piling up runnable work.
#1 Best Overall
How do I limit OpenBLAS threads in Python?
Set the relevant variable in the job’s launch environment before starting the Python interpreter. The variable must match the implementation actually loaded by the process:
OPENBLAS_NUM_THREADS=1for a process using OpenBLAS.MKL_NUM_THREADS=1for a process using MKL.BLIS_NUM_THREADS=1for a process using BLIS.OMP_NUM_THREADS=1for OpenMP-controlled code, with care: it can also affect some BLAS implementations.
These are examples, not a universal set of interchangeable switches. Setting one variable does not prove that every native thread pool is limited. Scikit-learn’s configuration guide describes these controls and their interaction with parallelism: scikit-learn parallelism documentation.
For example, a shell-based scheduler launch could prefix the command with the variable for the confirmed backend: OPENBLAS_NUM_THREADS=1 python job.py. Adapt the environment assignment to the scheduler you use; shell syntax is not guaranteed to apply unchanged in every scheduler or service.
Free tools Windows power users keep installed
One-click scans. No signup required.
How should I set thread limits for joblib workers?
If joblib’s loky backend is creating worker processes, joblib can limit supported third-party native thread pools in those child processes. Its parallel_config(inner_max_num_threads=...) setting provides an explicit per-worker cap:
from joblib import parallel_config
with parallel_config(backend="loky", inner_max_num_threads=1):
# Run joblib work here
...Choose a cap that fits the CPUs allocated to the task and the number of simultaneous workers. Avoid combining a high outer job count with unrestricted native parallelism unless that concurrency is intentional. Joblib’s controls are specific to the workers it manages; they do not automatically configure separate processes launched by a scheduler or unrelated multiprocessing code. See Joblib: Parallel computing and scikit-learn’s parallelism guidance.
How can I check which BLAS library NumPy is using?
Inspect the loaded pools from the same Python executable, package installation, environment, and launch configuration as the scheduled job. One documented route is:
python -m threadpoolctl -i numpy scipyReview the reported libraries and their thread limits. This helps establish whether the process loaded OpenBLAS, MKL, another supported backend, or more than one native pool, and whether the intended control appears to have taken effect. Scikit-learn documents this inspection command and shows checking a loaded library’s response to OMP_NUM_THREADS: scikit-learn parallelism documentation.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
NumPy’s linear algebra may use an external BLAS backend, so a package name alone does not establish which native library the running process loaded. For background on NumPy’s global state and linked libraries, see NumPy: Global state.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.How do I choose a thread limit for the scheduled job?
Start with the resources assigned to this task, not the machine’s full CPU count. Account for how many processes can run simultaneously and how many native threads each process may create. Then decide which layer should own most of the parallel work: the outer job/process layer or the BLAS/OpenMP layer.
- Identify the job’s CPU allocation and maximum simultaneous processes.
- Inspect the libraries loaded by the deployed process.
- Set limits for the libraries and worker mechanism actually in use, before launch where environment variables are required.
- Run the inspection again in the scheduled environment and compare the reported pools with the allocation and process count.
- Benchmark representative work after confirming the settings; a lower thread count is not automatically faster.
If the scheduled process behaves differently from an interactive run, first check whether it uses a different Python executable, package environment, launch configuration, or inherited thread variable. A scheduler’s CPU allocation does not by itself configure every numerical library.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
