diff --git a/neuron/neuron/submission_tester/testing.py b/neuron/neuron/submission_tester/testing.py
index a4f4a50d..50413c2b 100644
--- a/neuron/neuron/submission_tester/testing.py
+++ b/neuron/neuron/submission_tester/testing.py
@@ -57,6 +57,7 @@ def generate_baseline(
 ) -> BaselineBenchmark:
     outputs: list[GenerationOutput] = []
 
+    start_vram = CURRENT_CONTEST.get_vram_used()
     with InferenceSandbox(
         repository_info=CURRENT_CONTEST.baseline_repository,
         baseline=True,
@@ -82,7 +83,7 @@ def generate_baseline(
             outputs.append(output)
 
     generation_time = mean(output.generation_time for output in outputs)
-    vram_used = max(output.vram_used for output in outputs)
+    vram_used = max(output.vram_used for output in outputs) - start_vram
     watts_used = max(output.watts_used for output in outputs)
 
     return BaselineBenchmark(
@@ -111,6 +112,7 @@ def compare_checkpoints(
 
     outputs: list[GenerationOutput] = []
 
+    start_vram = CURRENT_CONTEST.get_vram_used()
     with InferenceSandbox(
         repository_info=submission,
         baseline=False,
@@ -144,7 +146,7 @@ def compare_checkpoints(
             raise InvalidSubmissionError(f"Failed to run inference") from e
 
     average_time = sum(output.generation_time for output in outputs) / len(outputs)
-    vram_used = max(output.vram_used for output in outputs)
+    vram_used = max(output.vram_used for output in outputs) - start_vram
     watts_used = max(output.watts_used for output in outputs)
 
     with CURRENT_CONTEST.output_comparator() as output_comparator:
diff --git a/neuron/neuron/submission_tester/vram_monitor.py b/neuron/neuron/submission_tester/vram_monitor.py
index 988a454d..788c4671 100644
--- a/neuron/neuron/submission_tester/vram_monitor.py
+++ b/neuron/neuron/submission_tester/vram_monitor.py
@@ -1,32 +1,27 @@
-import threading
-import time
+from threading import Thread, Event
+from time import sleep
 
 from .. import Contest
 
-POLL_RATE_SECONDS = 0.1
-
+SAMPLE_RATE_MS = 10
 
 class VRamMonitor:
-    _thread: threading.Thread
     _contest: Contest
+    _thread: Thread
+    _stop_flag: Event
     _vram_usage: int = 0
-    _stop_flag: threading.Event
 
     def __init__(self, contest: Contest):
         self._contest = contest
-        self._stop_flag = threading.Event()
+        self._stop_flag = Event()
 
-        self._thread = threading.Thread(target=self.monitor)
+        self._thread = Thread(target=self._monitor)
         self._thread.start()
 
-    def monitor(self):
+    def _monitor(self):
         while not self._stop_flag.is_set():
-            vram = self._contest.get_vram_used()
-
-            if self._vram_usage < vram:
-                self._vram_usage = vram
-
-            time.sleep(POLL_RATE_SECONDS)
+            self._vram_usage = max(self._vram_usage, self._contest.get_vram_used())
+            sleep(SAMPLE_RATE_MS / 1000)
 
     def complete(self) -> int:
         self._stop_flag.set()
diff --git a/validator/submission_tester/api.py b/validator/submission_tester/api.py
index 4756fa92..e7acec92 100644
--- a/validator/submission_tester/api.py
+++ b/validator/submission_tester/api.py
@@ -53,7 +53,7 @@ def flush(self):
 sys.stdout = LogsIO(sys.stdout, "out")
 sys.stderr = LogsIO(sys.stderr, "err")
 logging.basicConfig(
-    level=logging.DEBUG,
+    level=logging.INFO,
     format="%(asctime)s - %(levelname)s - %(filename)s: %(message)s",
     datefmt="%Y-%m-%d %H:%M:%S",
     stream=sys.stdout