diff --git a/.github/configs/amd-master.yaml b/.github/configs/amd-master.yaml
index 4596806b6..1eeed7bf1 100644
--- a/.github/configs/amd-master.yaml
+++ b/.github/configs/amd-master.yaml
@@ -113,7 +113,7 @@ dsr1-fp8-mi355x-sglang:
     - { tp: 8, conc-start: 4, conc-end: 64 }
 
 gptoss-fp4-mi300x-vllm:
-  image: rocm/7.0:rocm7.0_ubuntu_22.04_vllm_0.10.1_instinct_20250927_rc1
+  image: vllm/vllm-openai-rocm:v0.14.0
   model: openai/gpt-oss-120b
   model-prefix: gptoss
   runner: mi300x
@@ -144,7 +144,7 @@ gptoss-fp4-mi300x-vllm:
     - { tp: 8, conc-start: 4, conc-end: 16 }
 
 gptoss-fp4-mi325x-vllm:
-  image: rocm/7.0:rocm7.0_ubuntu_22.04_vllm_0.10.1_instinct_20250927_rc1
+  image: vllm/vllm-openai-rocm:v0.14.0
   model: openai/gpt-oss-120b
   model-prefix: gptoss
   runner: mi325x
diff --git a/benchmarks/gptoss_fp4_mi300x_docker.sh b/benchmarks/gptoss_fp4_mi300x_docker.sh
index 467a32a58..4a4fb5f89 100644
--- a/benchmarks/gptoss_fp4_mi300x_docker.sh
+++ b/benchmarks/gptoss_fp4_mi300x_docker.sh
@@ -24,9 +24,14 @@ if [[ "$version" == "" || $version -lt 177 ]]; then
   export HSA_NO_SCRATCH_RECLAIM=1
 fi
 
+# Set HIP_VISIBLE_DEVICES to match ROCR_VISIBLE_DEVICES for Ray compatibility in vLLM 0.14+
+if [ -n "$ROCR_VISIBLE_DEVICES" ]; then
+    export HIP_VISIBLE_DEVICES="$ROCR_VISIBLE_DEVICES"
+fi
+
 export VLLM_USE_AITER_UNIFIED_ATTENTION=1
 export VLLM_ROCM_USE_AITER_MHA=0
-export VLLM_ROCM_USE_AITER_TRITON_BF16_GEMM=0 
+export VLLM_ROCM_USE_AITER_TRITON_BF16_GEMM=0
 export VLLM_ROCM_QUICK_REDUCE_QUANTIZATION=INT4
 
 SERVER_LOG=$(mktemp /tmp/server-XXXXXX.log)
@@ -36,12 +41,10 @@ vllm serve $MODEL --port $PORT \
 --tensor-parallel-size=$TP \
 --gpu-memory-utilization 0.95 \
 --max-model-len $MAX_MODEL_LEN \
---max-seq-len-to-capture $MAX_MODEL_LEN \
 --compilation-config  '{"cudagraph_mode": "FULL_AND_PIECEWISE"}' \
 --block-size=64 \
 --no-enable-prefix-caching \
---disable-log-requests \
---async-scheduling > $SERVER_LOG 2>&1 &
+--disable-log-requests > $SERVER_LOG 2>&1 &
 
 SERVER_PID=$!
 
diff --git a/benchmarks/gptoss_fp4_mi300x_slurm.sh b/benchmarks/gptoss_fp4_mi300x_slurm.sh
index bc385c264..ac251d99c 100644
--- a/benchmarks/gptoss_fp4_mi300x_slurm.sh
+++ b/benchmarks/gptoss_fp4_mi300x_slurm.sh
@@ -30,6 +30,11 @@ if [[ "$version" == "" || $version -lt 177 ]]; then
   export HSA_NO_SCRATCH_RECLAIM=1
 fi
 
+# Set HIP_VISIBLE_DEVICES to match ROCR_VISIBLE_DEVICES for Ray compatibility in vLLM 0.14+
+if [ -n "$ROCR_VISIBLE_DEVICES" ]; then
+    export HIP_VISIBLE_DEVICES="$ROCR_VISIBLE_DEVICES"
+fi
+
 export VLLM_USE_AITER_UNIFIED_ATTENTION=1
 export VLLM_ROCM_USE_AITER_MHA=0
 export VLLM_ROCM_USE_AITER_TRITON_BF16_GEMM=0
@@ -42,12 +47,10 @@ vllm serve $MODEL --port $PORT \
 --tensor-parallel-size=$TP \
 --gpu-memory-utilization 0.95 \
 --max-model-len $MAX_MODEL_LEN \
---max-seq-len-to-capture $MAX_MODEL_LEN \
 --compilation-config  '{"cudagraph_mode": "FULL_AND_PIECEWISE"}' \
 --block-size=64 \
 --no-enable-prefix-caching \
 --disable-log-requests \
---async-scheduling \
 > $SERVER_LOG 2>&1 &
 
 SERVER_PID=$!
diff --git a/benchmarks/gptoss_fp4_mi325x_docker.sh b/benchmarks/gptoss_fp4_mi325x_docker.sh
index 054f6c377..ad22cda35 100644
--- a/benchmarks/gptoss_fp4_mi325x_docker.sh
+++ b/benchmarks/gptoss_fp4_mi325x_docker.sh
@@ -24,6 +24,11 @@ if [[ "$version" == "" || $version -lt 177 ]]; then
   export HSA_NO_SCRATCH_RECLAIM=1
 fi
 
+# Set HIP_VISIBLE_DEVICES to match ROCR_VISIBLE_DEVICES for Ray compatibility in vLLM 0.14+
+if [ -n "$ROCR_VISIBLE_DEVICES" ]; then
+    export HIP_VISIBLE_DEVICES="$ROCR_VISIBLE_DEVICES"
+fi
+
 export VLLM_USE_AITER_UNIFIED_ATTENTION=1
 export VLLM_ROCM_USE_AITER_MHA=0
 export VLLM_ROCM_USE_AITER_TRITON_BF16_GEMM=0
@@ -35,12 +40,10 @@ vllm serve $MODEL --port $PORT \
 --tensor-parallel-size=$TP \
 --gpu-memory-utilization 0.95 \
 --max-model-len $MAX_MODEL_LEN \
---max-seq-len-to-capture $MAX_MODEL_LEN \
 --compilation-config  '{"cudagraph_mode": "FULL_AND_PIECEWISE"}' \
 --block-size=64 \
 --no-enable-prefix-caching \
---disable-log-requests \
---async-scheduling > $SERVER_LOG 2>&1 &
+--disable-log-requests > $SERVER_LOG 2>&1 &
 
 SERVER_PID=$!
 
diff --git a/benchmarks/gptoss_fp4_mi325x_slurm.sh b/benchmarks/gptoss_fp4_mi325x_slurm.sh
index c0c9597c2..03d131285 100644
--- a/benchmarks/gptoss_fp4_mi325x_slurm.sh
+++ b/benchmarks/gptoss_fp4_mi325x_slurm.sh
@@ -30,6 +30,11 @@ if [[ "$version" == "" || $version -lt 177 ]]; then
   export HSA_NO_SCRATCH_RECLAIM=1
 fi
 
+# Set HIP_VISIBLE_DEVICES to match ROCR_VISIBLE_DEVICES for Ray compatibility in vLLM 0.14+
+if [ -n "$ROCR_VISIBLE_DEVICES" ]; then
+    export HIP_VISIBLE_DEVICES="$ROCR_VISIBLE_DEVICES"
+fi
+
 export VLLM_USE_AITER_UNIFIED_ATTENTION=1
 export VLLM_ROCM_USE_AITER_MHA=0
 export VLLM_ROCM_USE_AITER_TRITON_BF16_GEMM=0
@@ -39,12 +44,11 @@ vllm serve $MODEL --port $PORT \
 --tensor-parallel-size=$TP \
 --gpu-memory-utilization 0.95 \
 --max-model-len $MAX_MODEL_LEN \
---max-seq-len-to-capture $MAX_MODEL_LEN \
 --compilation-config  '{"cudagraph_mode": "FULL_AND_PIECEWISE"}' \
 --block-size=64 \
 --no-enable-prefix-caching \
 --disable-log-requests \
---async-scheduling > $SERVER_LOG 2>&1 &
+> $SERVER_LOG 2>&1 &
 
 SERVER_PID=$!
 
diff --git a/perf-changelog.yaml b/perf-changelog.yaml
index 5227dd9d2..71a44dabb 100644
--- a/perf-changelog.yaml
+++ b/perf-changelog.yaml
@@ -185,3 +185,13 @@
   description:
     - Add internal AMD ATOM inference engine for DeepSeek R1 FP8, FP4 and GPTOSS FP4 Mi355X
   pr-link: https://github.com/InferenceMAX/InferenceMAX/pull/419
+
+- config-keys:
+    - gptoss-fp4-mi300x-vllm
+    - gptoss-fp4-mi325x-vllm
+  description:
+    - "Update AMD MI300X and MI325X GPT-OSS 120B vLLM to use upstream ROCm image vllm/vllm-openai-rocm:v0.14.0"
+    - "Remove deprecated --async-scheduling flag (now enabled by default in vLLM v0.14.0)"
+    - "Remove deprecated --max-seq-len-to-capture flag"
+    - "Add HIP_VISIBLE_DEVICES env var for Ray compatibility in vLLM 0.14+"
+  pr-link: https://github.com/InferenceMAX/InferenceMAX/pull/496