diff --git a/slurm-eff-tool.py b/slurm-eff-tool.py
index 7acaf70..729fd14 100755
--- a/slurm-eff-tool.py
+++ b/slurm-eff-tool.py
@@ -62,6 +62,7 @@ DEFAULT_COLUMNS = [
     "CPUs",
     "Nodes",
     "ReqMem",
+    "MemPerCPU",
     "ReqWalltime",
     "Count",
     "CPU_Efficiency",
@@ -76,6 +77,7 @@ ALIASES = {
     "c": "CPUs",
     "N": "Nodes",
     "m": "ReqMem",
+    "p": "MemPerCPU",
     "l": "ReqWalltime",
     "C": "Count",
     "e": "CPU_Efficiency",
@@ -88,6 +90,7 @@ NUMERIC_COLUMNS = {
     "CPUs",
     "Nodes",
     "ReqMem",
+    "MemPerCPU",
     "ReqWalltime",
     "Count",
     "CPU_Efficiency",
@@ -104,6 +107,7 @@ class JobRecord:
     cpus: int
     nodes: int
     reqmem_gb: float | None
+    mem_per_cpu_gb: float | None
     reqwall_hours: float | None
     reqmem_bytes_total: float | None
     elapsed_sec: int
@@ -121,6 +125,7 @@ class OutputRow:
     CPUs: int
     Nodes: int
     ReqMem: float | None
+    MemPerCPU: float | None
     ReqWalltime: float | None
     Count: int
     CPU_Efficiency: float | None
@@ -137,6 +142,7 @@ class OutputRow:
             "CPUs": self.CPUs,
             "Nodes": self.Nodes,
             "ReqMem": self.ReqMem,
+            "MemPerCPU": self.MemPerCPU,
             "ReqWalltime": self.ReqWalltime,
             "Count": self.Count,
             "CPU_Efficiency": self.CPU_Efficiency,
@@ -385,6 +391,7 @@ def build_job_records(rows: list[dict[str, str]], only_user: str | None = None)
             reqmem_gb = reqmem_total / (1024**3)
 
         reqwall_hours = timelimit_seconds / 3600.0 if timelimit_seconds > 0 else None
+        mem_per_cpu_gb = reqmem_gb / cpus if reqmem_gb is not None and cpus > 0 else None
 
         cpu_eff = pct(totalcpu, cputime_raw)
         mem_eff = pct(maxrss, reqmem_total)
@@ -398,6 +405,7 @@ def build_job_records(rows: list[dict[str, str]], only_user: str | None = None)
                 cpus=cpus,
                 nodes=nodes,
                 reqmem_gb=reqmem_gb,
+                mem_per_cpu_gb=mem_per_cpu_gb,
                 reqwall_hours=reqwall_hours,
                 reqmem_bytes_total=reqmem_total,
                 elapsed_sec=elapsed,
@@ -450,6 +458,7 @@ def make_single_row(rec: JobRecord) -> OutputRow:
         CPUs=rec.cpus,
         Nodes=rec.nodes,
         ReqMem=rec.reqmem_gb,
+        MemPerCPU=rec.mem_per_cpu_gb,
         ReqWalltime=rec.reqwall_hours,
         Count=1,
         CPU_Efficiency=rec.cpu_eff,
@@ -472,6 +481,7 @@ def make_aggregate_row(records: list[JobRecord], username: str, jobname: str) ->
         CPUs=first.cpus,
         Nodes=first.nodes,
         ReqMem=first.reqmem_gb,
+        MemPerCPU=first.mem_per_cpu_gb,
         ReqWalltime=first.reqwall_hours,
         Count=len(records),
         CPU_Efficiency=mean_or_none(cpu_vals),
@@ -510,18 +520,23 @@ def sort_rows(rows: list[OutputRow], spec: str | None) -> list[OutputRow]:
         asc = t.startswith("+")
         name = t[1:] if (desc or asc) else t
         col = resolve_column(name)
-        if col not in NUMERIC_COLUMNS:
-            die(f"--sort column must be numeric, got {col}")
         parsed.append((col, desc))
 
     sorted_rows = rows
     # Stable-sort from least significant to most significant.
     for col, desc in reversed(parsed):
-        sorted_rows = sorted(
-            sorted_rows,
-            key=lambda r: float("-inf") if getattr(r, col) is None else getattr(r, col),
-            reverse=desc,
-        )
+        if col in NUMERIC_COLUMNS:
+            sorted_rows = sorted(
+                sorted_rows,
+                key=lambda r: float("-inf") if getattr(r, col) is None else getattr(r, col),
+                reverse=desc,
+            )
+        else:
+            sorted_rows = sorted(
+                sorted_rows,
+                key=lambda r: "" if getattr(r, col) is None else str(getattr(r, col)),
+                reverse=desc,
+            )
     return sorted_rows
 
 
@@ -551,6 +566,8 @@ def format_reqwall_hours(value: float | None) -> str:
 def format_value(value: Any, column: str | None = None) -> str:
     if column == "ReqMem":
         return format_reqmem_gb(value)
+    if column == "MemPerCPU":
+        return format_reqmem_gb(value)
     if column == "ReqWalltime":
         return format_reqwall_hours(value)
 
@@ -609,10 +626,14 @@ def parse_args(argv: list[str]) -> argparse.Namespace:
         description="Display seff-style CPU, memory and walltime efficiency values from sacct data."
     )
 
-    p.add_argument("-S", "--start", help="sacct start time, passed to sacct -S")
-    p.add_argument("-E", "--end", help="sacct end time, passed to sacct -E")
+    p.add_argument("-S", "--start", help="sacct start time, passed to sacct -S",
+                   default="now - 24 hours")
+    p.add_argument("-E", "--end", help="sacct end time, passed to sacct -E",
+                   default="now")
     p.add_argument("-u", "--user", help="restrict to one user; passed as sacct -u unless reading from cache")
-    p.add_argument("--state", "--job-state", dest="state", help="sacct state filter, e.g. COMPLETED,FAILED,TIMEOUT")
+    p.add_argument("--state", "--job-state", dest="state",
+                   default="COMPLETED",
+                   help="sacct state filter, e.g. COMPLETED,FAILED,TIMEOUT")
 
     p.add_argument("-O", "--output-cache", help="write raw sacct output cache to this file")
     p.add_argument("-F", "--from-cache", help="read raw sacct output cache from this file instead of running sacct")
@@ -633,7 +654,7 @@ def parse_args(argv: list[str]) -> argparse.Namespace:
         "--format",
         help=(
             "Slurm-like output format using aliases: "
-            "u=username,c=CPUs,N=Nodes,m=ReqMem,l=ReqWalltime,C=Count,"
+            "u=username,c=CPUs,N=Nodes,m=ReqMem,p=MemPerCPU,l=ReqWalltime,C=Count,"
             "e=CPU_Efficiency,M=Memory_Efficiency,t=Time_Efficiency,j=jobname"
         ),
     )