diff --git a/.github/configs/amd-master.yaml b/.github/configs/amd-master.yaml
index c362604f1..00c9f8bb2 100644
--- a/.github/configs/amd-master.yaml
+++ b/.github/configs/amd-master.yaml
@@ -140,7 +140,7 @@ dsr1-fp8-mi355x-sglang:
     - { tp: 8, conc-start: 4, conc-end: 64 }
 
 qwen3.5-bf16-mi355x-sglang:
-  image: rocm/sgl-dev:v0.5.8.post1-rocm720-mi35x-20260215
+  image: lmsysorg/sglang-rocm:v0.5.9-rocm720-mi35x-20260325
   model: Qwen/Qwen3.5-397B-A17B
   model-prefix: qwen3.5
   runner: mi355x
@@ -228,7 +228,7 @@ qwen3.5-fp8-mi325x-sglang:
     - { tp: 8, conc-start: 4, conc-end: 64 }
 
 qwen3.5-fp8-mi355x-sglang:
-  image: rocm/sgl-dev:v0.5.8.post1-rocm720-mi35x-20260218
+  image: lmsysorg/sglang-rocm:v0.5.9-rocm720-mi35x-20260325
   model: Qwen/Qwen3.5-397B-A17B-FP8
   model-prefix: qwen3.5
   runner: mi355x
diff --git a/benchmarks/single_node/qwen3.5_bf16_mi355x.sh b/benchmarks/single_node/qwen3.5_bf16_mi355x.sh
index f77390707..fb5550d0f 100755
--- a/benchmarks/single_node/qwen3.5_bf16_mi355x.sh
+++ b/benchmarks/single_node/qwen3.5_bf16_mi355x.sh
@@ -19,6 +19,8 @@ hf download "$MODEL"
 
 SERVER_LOG=/workspace/server.log
 PORT=${PORT:-8888}
+CONTEXT_LENGTH=$((ISL + OSL + 20))
+MAX_PREFILL_TOKENS=32768
 
 # Start GPU monitoring (power, temperature, clocks every second)
 start_gpu_monitor
@@ -30,6 +32,13 @@ python3 -m sglang.launch_server \
     --port $PORT \
     --tensor-parallel-size $TP \
     --trust-remote-code \
+    --tokenizer-worker-num 6 \
+    --enable-aiter-allreduce-fusion \
+    --cuda-graph-max-bs $CONC \
+    --context-length $CONTEXT_LENGTH \
+    --disable-radix-cache \
+    --max-prefill-tokens $MAX_PREFILL_TOKENS \
+    --scheduler-recv-interval 30 \
     --mem-fraction-static 0.8 > $SERVER_LOG 2>&1 &
 
 SERVER_PID=$!
diff --git a/benchmarks/single_node/qwen3.5_fp8_mi355x.sh b/benchmarks/single_node/qwen3.5_fp8_mi355x.sh
index f77390707..fb5550d0f 100644
--- a/benchmarks/single_node/qwen3.5_fp8_mi355x.sh
+++ b/benchmarks/single_node/qwen3.5_fp8_mi355x.sh
@@ -19,6 +19,8 @@ hf download "$MODEL"
 
 SERVER_LOG=/workspace/server.log
 PORT=${PORT:-8888}
+CONTEXT_LENGTH=$((ISL + OSL + 20))
+MAX_PREFILL_TOKENS=32768
 
 # Start GPU monitoring (power, temperature, clocks every second)
 start_gpu_monitor
@@ -30,6 +32,13 @@ python3 -m sglang.launch_server \
     --port $PORT \
     --tensor-parallel-size $TP \
     --trust-remote-code \
+    --tokenizer-worker-num 6 \
+    --enable-aiter-allreduce-fusion \
+    --cuda-graph-max-bs $CONC \
+    --context-length $CONTEXT_LENGTH \
+    --disable-radix-cache \
+    --max-prefill-tokens $MAX_PREFILL_TOKENS \
+    --scheduler-recv-interval 30 \
     --mem-fraction-static 0.8 > $SERVER_LOG 2>&1 &
 
 SERVER_PID=$!
diff --git a/perf-changelog.yaml b/perf-changelog.yaml
index 03fb6e082..7a6e789ac 100644
--- a/perf-changelog.yaml
+++ b/perf-changelog.yaml
@@ -1067,4 +1067,11 @@
     - "dsr1-fp8-b200-sglang-mtp: v0.5.8-cu130-amd64 → v0.5.9-cu130"
     - "dsr1-fp8-h200-sglang: v0.5.9-cu129-amd64 → v0.5.9-cu130"
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/943
+  
+- config-keys:
+    - qwen3.5-fp8-mi355x-sglang
+    - qwen3.5-bf16-mi355x-sglang
+  description:
+    - "Update cli args of Qwen3.5 FP8 and BF16 SGLang benchmarks for MI355X to achieve better performance"
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/942