diff --git a/CLAUDE.md b/CLAUDE.md
index dcaa076..b6e8b0a 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -58,7 +58,7 @@ Unexpected `**kwargs` for any mode trigger a `UserWarning`.
 | ---------------------- | ------------------------------------------------------------------------------------------------------- |
 | `src/pdex/__init__.py` | `pdex()` entry point and full pipeline logic                                                            |
 | `src/pdex/_math.py`    | Numba JIT-compiled `fold_change()`, `percent_change()`, and `mwu()` wrappers; `pseudobulk()` dispatcher |
-| `src/pdex/_utils.py`   | `set_numba_threadpool()` — sets Numba thread count before JIT warmup; `_detect_is_log1p()` heuristic    |
+| `src/pdex/_utils.py`   | `set_numba_threadpool()` — sets Numba thread count before JIT warmup; `_available_cpus()` — affinity-aware CPU count (respects cgroup/SLURM limits); `_detect_is_log1p()` heuristic |
 
 ### Performance Design
 
diff --git a/pyproject.toml b/pyproject.toml
index 1d3849f..e83f869 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -1,6 +1,6 @@
 [project]
 name = "pdex"
-version = "0.2.3"
+version = "0.2.4"
 description = "Parallel differential expression for single-cell perturbation sequencing"
 readme = "README.md"
 authors = [{ name = "noam teyssier", email = "noam.teyssier@arcinstitute.org" }]
diff --git a/src/pdex/_utils.py b/src/pdex/_utils.py
index d18c3c7..dbb33e4 100644
--- a/src/pdex/_utils.py
+++ b/src/pdex/_utils.py
@@ -1,4 +1,5 @@
 import logging
+import multiprocessing as mp
 import os
 
 import numba
@@ -8,16 +9,28 @@
 log = logging.getLogger(__name__)
 
 
+def _available_cpus() -> int:
+    """Return the number of CPUs the current process is allowed to use.
+
+    Uses ``os.sched_getaffinity`` on Linux so SLURM/cgroup/taskset limits are
+    respected; falls back to ``multiprocessing.cpu_count`` on macOS/Windows where
+    that API is unavailable (those platforms typically run locally without cgroup
+    caps). This mirrors how Numba derives ``NUMBA_NUM_THREADS``, so the value never
+    exceeds Numba's cap — unlike ``os.cpu_count()``, which reports every physical
+    CPU even when affinity or a cgroup restricts the process to far fewer, causing
+    ``numba.set_num_threads`` to raise.
+    """
+    try:
+        return len(os.sched_getaffinity(0))
+    except AttributeError:
+        return mp.cpu_count()
+
+
 def set_numba_threadpool(threads: int = 0):
-    available_threads = os.cpu_count()
-    if available_threads is None:
-        available_threads = 1
+    available_threads = _available_cpus()
 
     if threads == 0:
-        if not available_threads:
-            threads = 1
-        else:
-            threads = available_threads
+        threads = available_threads
     else:
         threads = min(threads, available_threads)
 
diff --git a/tests/test_utils.py b/tests/test_utils.py
index b79d450..054feef 100644
--- a/tests/test_utils.py
+++ b/tests/test_utils.py
@@ -1,10 +1,8 @@
 """Tests for pdex._utils (set_numba_threadpool)."""
 
-import os
-
 import numba
 
-from pdex._utils import set_numba_threadpool
+from pdex._utils import _available_cpus, set_numba_threadpool
 
 
 class TestSetNumbaThreadpool:
@@ -12,10 +10,9 @@ def test_explicit_thread_count(self):
         set_numba_threadpool(4)
         assert numba.get_num_threads() == 4
 
-    def test_zero_uses_all_cpus(self):
+    def test_zero_uses_all_available_cpus(self):
         set_numba_threadpool(0)
-        expected = os.cpu_count() or 1
-        assert numba.get_num_threads() == expected
+        assert numba.get_num_threads() == _available_cpus()
 
     def test_single_thread(self):
         set_numba_threadpool(1)