diff --git a/.github/workflows/rocm-ci.yml b/.github/workflows/rocm-ci.yml
index c8d655330..de8653005 100644
--- a/.github/workflows/rocm-ci.yml
+++ b/.github/workflows/rocm-ci.yml
@@ -1,4 +1,4 @@
-# Copyright (c) 2024-2025, Advanced Micro Devices, Inc. All rights reserved.
+# Copyright (c) 2024-2026, Advanced Micro Devices, Inc. All rights reserved.
 #
 # See LICENSE for license information.
 
@@ -252,9 +252,10 @@ jobs:
           HIP_VISIBLE_DEVICES=2 ci/jax.sh > /workspace/jax_sgpu.log 2>&1 &
           jax_pid=$!; echo JAX test pid $!
           
-          ci/core.sh > /workspace/core_sgpu.log 2>&1 
-          core_rc=$?
+          HIP_VISIBLE_DEVICES=3 ci/core.sh > /workspace/core_sgpu.log 2>&1 &
+          core_pid=$!; echo Core test pid $!
           
+          wait $core_pid; core_rc=$?
           wait $jax_pid; jax_rc=$?
           wait $torch_pid; torch_rc=$?
           
diff --git a/ci/core.sh b/ci/core.sh
index 24ef245d2..e940b12ff 100755
--- a/ci/core.sh
+++ b/ci/core.sh
@@ -1,5 +1,5 @@
 #!/bin/sh
-# Copyright (c) 2024-2025, Advanced Micro Devices, Inc. All rights reserved.
+# Copyright (c) 2024-2026, Advanced Micro Devices, Inc. All rights reserved.
 #
 # See LICENSE for license information.
 
@@ -13,7 +13,7 @@ if [ -z "$TEST_SGPU" ]; then
     exit 0
 fi
 
-n_parallel_jobs=8
+n_parallel_jobs=4
 
 configure_omp_threads $n_parallel_jobs
 
diff --git a/ci/pytorch.sh b/ci/pytorch.sh
index be150485f..54d81aa14 100755
--- a/ci/pytorch.sh
+++ b/ci/pytorch.sh
@@ -56,7 +56,6 @@ run_test_config(){
     run_default_fa 1 test_fused_router.py
     run_default_fa 1 test_fusible_ops.py
     run_default_fa 1 test_gemm_autotune.py
-    run_default_fa 1 test_gemm_sm_count.py
     run 1 test_gqa.py
     run 1 test_jit.py
     run_default_fa 1 test_multi_tensor.py
@@ -88,6 +87,9 @@ run_test_config_mgpu(){
     echo ==== Run mGPU with Fused attention backend: $_fus_attn ====
     configure_omp_threads 8
     run_default_fa 1 test_fused_optimizer.py
+    #this test is not really mGPU but time sensitive so run it here because sGPU tests
+    #run in parallel on CI and it affects timing
+    run_default_fa 1 test_gemm_sm_count.py
     run_default_fa 3 test_sanity_import.py
     run_default_fa 2 distributed/test_fusible_ops.py
     run_default_fa 2 distributed/test_numerics.py