diff --git a/.github/workflows/nightly-test-amd.yml b/.github/workflows/nightly-test-amd.yml
index 6a48573a1..1c1081ed8 100644
--- a/.github/workflows/nightly-test-amd.yml
+++ b/.github/workflows/nightly-test-amd.yml
@@ -260,7 +260,7 @@ jobs:
       - name: Nightly Perf Test (8-GPU Grok-1 + Grok-2)
         timeout-minutes: 60
         run: |
-          bash scripts/ci/amd_ci_exec.sh -w /sglang-checkout/test -e RCCL_MSCCL_ENABLE=0 -e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" python3 registered/test_grok_perf.py || TEST_EXIT_CODE=$?
+          bash scripts/ci/amd_ci_exec.sh -w /sglang-checkout/test -e RCCL_MSCCL_ENABLE=0 -e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" python3 registered/amd/test_grok_perf.py || TEST_EXIT_CODE=$?
           echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
           exit ${TEST_EXIT_CODE:-0}
 
@@ -285,7 +285,7 @@ jobs:
       - name: Nightly Perf Test (8-GPU DeepSeek-V3)
         timeout-minutes: 300
         run: |
-          bash scripts/ci/amd_ci_exec.sh -w /sglang-checkout/test -e SGLANG_USE_ROCM700A=1 -e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" python3 registered/test_deepseek_v3_perf.py || TEST_EXIT_CODE=$?
+          bash scripts/ci/amd_ci_exec.sh -w /sglang-checkout/test -e SGLANG_USE_ROCM700A=1 -e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" python3 registered/amd/test_deepseek_v3_perf.py || TEST_EXIT_CODE=$?
           echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
           exit ${TEST_EXIT_CODE:-0}
 
@@ -310,7 +310,7 @@ jobs:
       - name: Nightly Perf Test (8-GPU DeepSeek-V3.1)
         timeout-minutes: 300
         run: |
-          bash scripts/ci/amd_ci_exec.sh -w /sglang-checkout/test -e SGLANG_USE_ROCM700A=1 -e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" python3 registered/test_deepseek_v31_perf.py || TEST_EXIT_CODE=$?
+          bash scripts/ci/amd_ci_exec.sh -w /sglang-checkout/test -e SGLANG_USE_ROCM700A=1 -e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" python3 registered/amd/test_deepseek_v31_perf.py || TEST_EXIT_CODE=$?
           echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
           exit ${TEST_EXIT_CODE:-0}
 
diff --git a/test/registered/test_deepseek_v31_perf.py b/test/registered/amd/test_deepseek_v31_perf.py
similarity index 100%
rename from test/registered/test_deepseek_v31_perf.py
rename to test/registered/amd/test_deepseek_v31_perf.py
diff --git a/test/registered/test_deepseek_v3_perf.py b/test/registered/amd/test_deepseek_v3_perf.py
similarity index 100%
rename from test/registered/test_deepseek_v3_perf.py
rename to test/registered/amd/test_deepseek_v3_perf.py
diff --git a/test/registered/test_grok_perf.py b/test/registered/amd/test_grok_perf.py
similarity index 100%
rename from test/registered/test_grok_perf.py
rename to test/registered/amd/test_grok_perf.py
diff --git a/test/srt/nightly/test_gsm8k_eval_amd.py b/test/srt/nightly/test_gsm8k_eval_amd.py
index 19dacb432..5a94a7780 100644
--- a/test/srt/nightly/test_gsm8k_eval_amd.py
+++ b/test/srt/nightly/test_gsm8k_eval_amd.py
@@ -37,7 +37,7 @@ MODEL_SCORE_THRESHOLDS = {
     "Qwen/Qwen2.5-7B-Instruct": 0.85,
     # Qwen3 series
     "Qwen/Qwen3-30B-A3B-Thinking-2507": 0.84,  # MoE model verified on MI300X
-    "Qwen/Qwen3-8B": 0.80,
+    "Qwen/Qwen3-8B": 0.77,
     # Google Gemma
     "google/gemma-2-27b-it": 0.91,
     "google/gemma-2-9b-it": 0.72,