diff --git a/.github/workflows/nvidia-smoke.yml b/.github/workflows/nvidia-smoke.yml
new file mode 100644
index 00000000..1cfdbec6
--- /dev/null
+++ b/.github/workflows/nvidia-smoke.yml
@@ -0,0 +1,50 @@
+name: gpu-smoke
+on:
+  push:
+    branches: [ nvidia-gpu-runners ]
+  workflow_dispatch: {}
+jobs:
+  gpu-test:
+    runs-on: [self-hosted, nvidia-docker-b200-8-x86-64]
+
+    steps:
+      - name: Checkout repo
+        uses: actions/checkout@v4
+
+      - name: Show GPU info
+        run: |
+          echo "===== nvidia-smi ====="
+          nvidia-smi || echo "nvidia-smi not available"
+          echo "======================"
+
+      - name: Run CUDA sanity test with PyTorch
+        run: |
+          python3 - << 'EOF'
+          import torch, time
+
+          print("PyTorch version:", torch.__version__)
+          print("CUDA available:", torch.cuda.is_available())
+          print("CUDA device count:", torch.cuda.device_count())
+
+          if not torch.cuda.is_available():
+              raise SystemExit("ERROR: CUDA not available on this runner ❌")
+
+          # list all visible GPUs
+          for i in range(torch.cuda.device_count()):
+              print(f"Device {i}: {torch.cuda.get_device_name(i)}")
+
+          # simple GPU compute test on cuda:0
+          device = torch.device("cuda:0")
+          a = torch.randn(4096, 4096, device=device)
+          b = torch.randn(4096, 4096, device=device)
+
+          torch.cuda.synchronize()
+          t0 = time.time()
+          c = a @ b
+          torch.cuda.synchronize()
+          t1 = time.time()
+
+          print("Matmul result shape:", tuple(c.shape))
+          print(f"Matmul took {t1 - t0:.3f} sec on GPU")
+          print("All good ✅")
+          EOF
diff --git a/src/kernelbot/cogs/admin_cog.py b/src/kernelbot/cogs/admin_cog.py
index 5fe8aeba..238364bb 100644
--- a/src/kernelbot/cogs/admin_cog.py
+++ b/src/kernelbot/cogs/admin_cog.py
@@ -276,7 +276,11 @@ async def leaderboard_create_impl(  # noqa: C901
             forum_thread = await forum_channel.create_thread(
                 name=leaderboard_name,
                 content=self._leaderboard_opening_message(
-                    leaderboard_name, date_value, definition.description
+                    leaderboard_name,
+                    date_value,
+                    definition.description[:1500]
+                    if len(definition.description) > 1500
+                    else definition.description,
                 ),
                 auto_archive_duration=10080,  # 7 days
             )