diff --git a/benchmarks/single_node/qwen3.5_bf16_mi300x.sh b/benchmarks/single_node/qwen3.5_bf16_mi300x.sh
index aa74785fe..8aca9860a 100755
--- a/benchmarks/single_node/qwen3.5_bf16_mi300x.sh
+++ b/benchmarks/single_node/qwen3.5_bf16_mi300x.sh
@@ -36,7 +36,8 @@ python3 -m sglang.launch_server \
     --port $PORT \
     --tensor-parallel-size $TP \
     --trust-remote-code \
-    --mem-fraction-static 0.8 $EVAL_CONTEXT_ARGS > $SERVER_LOG 2>&1 &
+    --mem-fraction-static 0.8 \
+    --disable-radix-cache $EVAL_CONTEXT_ARGS > $SERVER_LOG 2>&1 &
 
 SERVER_PID=$!
 
diff --git a/benchmarks/single_node/qwen3.5_bf16_mi325x.sh b/benchmarks/single_node/qwen3.5_bf16_mi325x.sh
index aa74785fe..8aca9860a 100644
--- a/benchmarks/single_node/qwen3.5_bf16_mi325x.sh
+++ b/benchmarks/single_node/qwen3.5_bf16_mi325x.sh
@@ -36,7 +36,8 @@ python3 -m sglang.launch_server \
     --port $PORT \
     --tensor-parallel-size $TP \
     --trust-remote-code \
-    --mem-fraction-static 0.8 $EVAL_CONTEXT_ARGS > $SERVER_LOG 2>&1 &
+    --mem-fraction-static 0.8 \
+    --disable-radix-cache $EVAL_CONTEXT_ARGS > $SERVER_LOG 2>&1 &
 
 SERVER_PID=$!
 
diff --git a/benchmarks/single_node/qwen3.5_fp8_mi300x.sh b/benchmarks/single_node/qwen3.5_fp8_mi300x.sh
index 7bff57b61..00cc9cf91 100755
--- a/benchmarks/single_node/qwen3.5_fp8_mi300x.sh
+++ b/benchmarks/single_node/qwen3.5_fp8_mi300x.sh
@@ -37,7 +37,8 @@ python3 -m sglang.launch_server \
     --port $PORT \
     --tensor-parallel-size $TP \
     --trust-remote-code \
-    --mem-fraction-static 0.8 $EVAL_CONTEXT_ARGS > $SERVER_LOG 2>&1 &
+    --mem-fraction-static 0.8 \
+    --disable-radix-cache $EVAL_CONTEXT_ARGS > $SERVER_LOG 2>&1 &
 
 SERVER_PID=$!
 
diff --git a/benchmarks/single_node/qwen3.5_fp8_mi325x.sh b/benchmarks/single_node/qwen3.5_fp8_mi325x.sh
index 7bff57b61..00cc9cf91 100755
--- a/benchmarks/single_node/qwen3.5_fp8_mi325x.sh
+++ b/benchmarks/single_node/qwen3.5_fp8_mi325x.sh
@@ -37,7 +37,8 @@ python3 -m sglang.launch_server \
     --port $PORT \
     --tensor-parallel-size $TP \
     --trust-remote-code \
-    --mem-fraction-static 0.8 $EVAL_CONTEXT_ARGS > $SERVER_LOG 2>&1 &
+    --mem-fraction-static 0.8 \
+    --disable-radix-cache $EVAL_CONTEXT_ARGS > $SERVER_LOG 2>&1 &
 
 SERVER_PID=$!
 
diff --git a/perf-changelog.yaml b/perf-changelog.yaml
index 81fc5f100..b3720abc8 100644
--- a/perf-changelog.yaml
+++ b/perf-changelog.yaml
@@ -1189,3 +1189,12 @@
     - "Separate evals, change to 8k1k, fail loudly, 5-shot, top of curve & middle of curve"
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/911
   evals-only: true
+
+- config-keys:
+    - qwen3.5-bf16-mi300x-sglang
+    - qwen3.5-bf16-mi325x-sglang
+    - qwen3.5-fp8-mi300x-sglang
+    - qwen3.5-fp8-mi325x-sglang
+  description:
+    - "Add --disable-radix-cache to SGLang server launch command for qwen3.5 MI300X and MI325X benchmark scripts"
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/970