diff --git a/.github/workflows/triton-test.yaml b/.github/workflows/triton-test.yaml
index 9afe70c019..92bf3edbc7 100644
--- a/.github/workflows/triton-test.yaml
+++ b/.github/workflows/triton-test.yaml
@@ -13,7 +13,7 @@ concurrency:
 
 jobs:
   triton:
-    runs-on: aiter-1gpu-runner
+    runs-on: aiter-mi355-1gpu
     env:
       DOCKER_IMAGE: "rocm/pytorch:latest"
       TRITON_TEST: "op_tests/triton_tests/"
@@ -35,7 +35,7 @@ jobs:
           docker run -dt \
           --device=/dev/dri \
           --device=/dev/kfd \
-          --shm-size=16G \
+          --shm-size=32G \
           --group-add $(getent group render | cut -d: -f3) \
           --group-add $(getent group video | cut -d: -f3) \
           -v "${{ github.workspace }}:/workspace" \
diff --git a/op_tests/triton_tests/test_batched_gemm_a8w8.py b/op_tests/triton_tests/test_batched_gemm_a8w8.py
index 292e705da7..85d9bae1b5 100644
--- a/op_tests/triton_tests/test_batched_gemm_a8w8.py
+++ b/op_tests/triton_tests/test_batched_gemm_a8w8.py
@@ -101,7 +101,7 @@ def get_x_vals():
         (256, 8192, 1024),
         (320, 8192, 1024),
         (512, 8192, 1024),
-        (1024, 8192, 1024),
+        # (1024, 8192, 1024),
         (2048, 8192, 1024),
         (4096, 8192, 1024),
         (8192, 8192, 1024),