diff --git a/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/1p2d-dep2-tep4-8k1k.yaml b/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/1p1d-tp1-tp4-8k1k.yaml
similarity index 78%
rename from benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/1p2d-dep2-tep4-8k1k.yaml
rename to benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/1p1d-tp1-tp4-8k1k.yaml
index 5be198f113..ad0d248260 100644
--- a/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/1p2d-dep2-tep4-8k1k.yaml
+++ b/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/1p1d-tp1-tp4-8k1k.yaml
@@ -1,47 +1,37 @@
-name: "minimax-m3-vllm-disagg-b300-1p2d-fp4-dep2-tep4-8k1k"
-
+name: "minimax-m3-vllm-disagg-b300-1p1d-tp1-tp4-fp4-8k1k"
 model:
   path: "nvidia/MiniMax-M3-NVFP4"
-  container: "vllm/vllm-openai:vllm-minimax-m3-perf-x86_64-13.0.1-8b00f41"
+  container: "vllm/vllm-openai:nightly-2dfaae752b4db0d43cfc0715c780e33be030d0f1"
   precision: "fp4"
-
 resources:
   gpu_type: "b300"
   gpus_per_node: 8
   prefill_nodes: 1
-  decode_nodes: 1
+  decode_nodes: 0
   prefill_workers: 1
-  decode_workers: 2
-  gpus_per_prefill: 2
+  decode_workers: 1
+  gpus_per_prefill: 1
   gpus_per_decode: 4
-
 dynamo:
   install: true
-  version: 1.3.0.dev20260614
-
+  version: 1.3.0.dev20260713
 frontend:
   type: dynamo
   enable_multiple_frontends: false
-
 backend:
   type: vllm
-  connector: null
-
+  connector:
   prefill_environment:
     VLLM_FLOAT32_MATMUL_PRECISION: high
     UCX_TLS: "cuda_copy,rc"
-
   decode_environment:
     VLLM_FLOAT32_MATMUL_PRECISION: high
     UCX_TLS: "cuda_copy,rc"
-
   vllm_config:
     prefill:
       no-enable-flashinfer-autotune: true
       tensor-parallel-size: 1
-      data-parallel-size: 2
-      data-parallel-rpc-port: 13345
-      enable-expert-parallel: true
+      enable-expert-parallel: false
       trust-remote-code: true
       no-enable-prefix-caching: true
       kv-transfer-config: '{"kv_connector": "NixlConnector", "kv_role": "kv_both"}'
@@ -52,13 +42,12 @@ backend:
       max-model-len: 9472
       language-model-only: true
       stream-interval: 32
-      max-cudagraph-capture-size: 2048
-      max-num-batched-tokens: 16384
-
+      enforce-eager: true
+      max-num-batched-tokens: 8192
     decode:
       no-enable-flashinfer-autotune: true
       tensor-parallel-size: 4
-      enable-expert-parallel: true
+      enable-expert-parallel: false
       trust-remote-code: true
       no-enable-prefix-caching: true
       kv-transfer-config: '{"kv_connector": "NixlConnector", "kv_role": "kv_both"}'
@@ -72,14 +61,13 @@ backend:
       max-num-seqs: 512
       max-num-batched-tokens: 16384
       max-cudagraph-capture-size: 4096
-
+  allow_prefill_decode_colocation: true
 health_check:
   max_attempts: 360
   interval_seconds: 10
-
 benchmark:
   type: "sa-bench"
   isl: 8192
   osl: 1024
-  concurrencies: "16"
+  concurrencies: "8x16x24x32x48x64"
   req_rate: "inf"
diff --git a/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/1p4d-dep2-tep8-8k1k.yaml b/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/1p1d-tp1-tp8-8k1k.yaml
similarity index 75%
rename from benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/1p4d-dep2-tep8-8k1k.yaml
rename to benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/1p1d-tp1-tp8-8k1k.yaml
index 2154742821..74e87048ca 100644
--- a/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/1p4d-dep2-tep8-8k1k.yaml
+++ b/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/1p1d-tp1-tp8-8k1k.yaml
@@ -1,47 +1,38 @@
-name: "minimax-m3-vllm-disagg-b300-1p4d-fp4-dep2-tep8-8k1k"
-
+name: "minimax-m3-vllm-disagg-b300-1p1d-tp1-tp8-fp4-8k1k"
 model:
   path: "nvidia/MiniMax-M3-NVFP4"
-  container: "vllm/vllm-openai:vllm-minimax-m3-perf-x86_64-13.0.1-8b00f41"
+  container: "vllm/vllm-openai:nightly-2dfaae752b4db0d43cfc0715c780e33be030d0f1"
   precision: "fp4"
-
 resources:
   gpu_type: "b300"
   gpus_per_node: 8
   prefill_nodes: 1
-  decode_nodes: 4
+  decode_nodes: 1
   prefill_workers: 1
-  decode_workers: 4
-  gpus_per_prefill: 2
+  decode_workers: 1
+  gpus_per_prefill: 1
   gpus_per_decode: 8
-
 dynamo:
   install: true
-  version: 1.3.0.dev20260614
-
+  version: 1.3.0.dev20260713
 frontend:
   type: dynamo
   enable_multiple_frontends: false
-
 backend:
   type: vllm
-  connector: null
-
+  connector:
   prefill_environment:
     VLLM_FLOAT32_MATMUL_PRECISION: high
     UCX_TLS: "cuda_copy,rc"
-
   decode_environment:
     VLLM_FLOAT32_MATMUL_PRECISION: high
     UCX_TLS: "cuda_copy,rc"
-
+    VLLM_FLASHINFER_ALLREDUCE_BACKEND: trtllm
   vllm_config:
     prefill:
       no-enable-flashinfer-autotune: true
       tensor-parallel-size: 1
-      data-parallel-size: 2
-      data-parallel-rpc-port: 13345
-      enable-expert-parallel: true
+      enable-expert-parallel: false
       trust-remote-code: true
       no-enable-prefix-caching: true
       kv-transfer-config: '{"kv_connector": "NixlConnector", "kv_role": "kv_both"}'
@@ -52,13 +43,12 @@ backend:
       max-model-len: 9472
       language-model-only: true
       stream-interval: 32
-      max-cudagraph-capture-size: 2048
-      max-num-batched-tokens: 16384
-
+      enforce-eager: true
+      max-num-batched-tokens: 8192
     decode:
       no-enable-flashinfer-autotune: true
       tensor-parallel-size: 8
-      enable-expert-parallel: true
+      enable-expert-parallel: false
       trust-remote-code: true
       no-enable-prefix-caching: true
       kv-transfer-config: '{"kv_connector": "NixlConnector", "kv_role": "kv_both"}'
@@ -69,17 +59,15 @@ backend:
       max-model-len: 9472
       language-model-only: true
       stream-interval: 32
-      max-num-seqs: 1024
+      max-num-seqs: 256
       max-num-batched-tokens: 16384
-      max-cudagraph-capture-size: 4096
-
+      max-cudagraph-capture-size: 256
 health_check:
   max_attempts: 360
   interval_seconds: 10
-
 benchmark:
   type: "sa-bench"
   isl: 8192
   osl: 1024
-  concurrencies: "4"
+  concurrencies: "1x2x4x8x16"
   req_rate: "inf"
diff --git a/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/1p4d-dep2-tep4-8k1k.yaml b/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/2p1d-tp1-tp4-8k1k.yaml
similarity index 76%
rename from benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/1p4d-dep2-tep4-8k1k.yaml
rename to benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/2p1d-tp1-tp4-8k1k.yaml
index 90d688f615..cb966b1e7c 100644
--- a/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/1p4d-dep2-tep4-8k1k.yaml
+++ b/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/2p1d-tp1-tp4-8k1k.yaml
@@ -1,47 +1,37 @@
-name: "minimax-m3-vllm-disagg-b300-1p4d-fp4-dep2-tep4-8k1k"
-
+name: "minimax-m3-vllm-disagg-b300-2p1d-tp1-tp4-fp4-8k1k"
 model:
   path: "nvidia/MiniMax-M3-NVFP4"
-  container: "vllm/vllm-openai:vllm-minimax-m3-perf-x86_64-13.0.1-8b00f41"
+  container: "vllm/vllm-openai:nightly-2dfaae752b4db0d43cfc0715c780e33be030d0f1"
   precision: "fp4"
-
 resources:
   gpu_type: "b300"
   gpus_per_node: 8
   prefill_nodes: 1
-  decode_nodes: 2
-  prefill_workers: 1
-  decode_workers: 4
-  gpus_per_prefill: 2
+  decode_nodes: 0
+  prefill_workers: 2
+  decode_workers: 1
+  gpus_per_prefill: 1
   gpus_per_decode: 4
-
 dynamo:
   install: true
-  version: 1.3.0.dev20260614
-
+  version: 1.3.0.dev20260713
 frontend:
   type: dynamo
   enable_multiple_frontends: false
-
 backend:
   type: vllm
-  connector: null
-
+  connector:
   prefill_environment:
     VLLM_FLOAT32_MATMUL_PRECISION: high
     UCX_TLS: "cuda_copy,rc"
-
   decode_environment:
     VLLM_FLOAT32_MATMUL_PRECISION: high
     UCX_TLS: "cuda_copy,rc"
-
   vllm_config:
     prefill:
       no-enable-flashinfer-autotune: true
       tensor-parallel-size: 1
-      data-parallel-size: 2
-      data-parallel-rpc-port: 13345
-      enable-expert-parallel: true
+      enable-expert-parallel: false
       trust-remote-code: true
       no-enable-prefix-caching: true
       kv-transfer-config: '{"kv_connector": "NixlConnector", "kv_role": "kv_both"}'
@@ -52,13 +42,12 @@ backend:
       max-model-len: 9472
       language-model-only: true
       stream-interval: 32
-      max-cudagraph-capture-size: 2048
-      max-num-batched-tokens: 16384
-
+      enforce-eager: true
+      max-num-batched-tokens: 8192
     decode:
       no-enable-flashinfer-autotune: true
       tensor-parallel-size: 4
-      enable-expert-parallel: true
+      enable-expert-parallel: false
       trust-remote-code: true
       no-enable-prefix-caching: true
       kv-transfer-config: '{"kv_connector": "NixlConnector", "kv_role": "kv_both"}'
@@ -72,14 +61,13 @@ backend:
       max-num-seqs: 512
       max-num-batched-tokens: 16384
       max-cudagraph-capture-size: 4096
-
+  allow_prefill_decode_colocation: true
 health_check:
   max_attempts: 360
   interval_seconds: 10
-
 benchmark:
   type: "sa-bench"
   isl: 8192
   osl: 1024
-  concurrencies: "16x32x64x128"
+  concurrencies: "128"
   req_rate: "inf"
diff --git a/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/1p1d-dep2-tp4-8k1k.yaml b/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/3p1d-tp1-tp4-8k1k.yaml
similarity index 73%
rename from benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/1p1d-dep2-tp4-8k1k.yaml
rename to benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/3p1d-tp1-tp4-8k1k.yaml
index be2683d0ca..3c6917dfa9 100644
--- a/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/1p1d-dep2-tp4-8k1k.yaml
+++ b/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/3p1d-tp1-tp4-8k1k.yaml
@@ -1,48 +1,38 @@
-name: "minimax-m3-vllm-disagg-b300-1p1d-fp4-dep2-tp4-8k1k"
-
+name: "minimax-m3-vllm-disagg-b300-3p1d-tp1-tp4-fp4-8k1k"
 model:
   path: "nvidia/MiniMax-M3-NVFP4"
-  container: "vllm/vllm-openai:vllm-minimax-m3-perf-x86_64-13.0.1-8b00f41"
+  container: "vllm/vllm-openai:nightly-2dfaae752b4db0d43cfc0715c780e33be030d0f1"
   precision: "fp4"
-
 resources:
   gpu_type: "b300"
   gpus_per_node: 8
   prefill_nodes: 1
-  decode_nodes: 1
-  prefill_workers: 1
+  decode_nodes: 0
+  prefill_workers: 3
   decode_workers: 1
-  gpus_per_prefill: 2
+  gpus_per_prefill: 1
   gpus_per_decode: 4
-
 dynamo:
   install: true
-  version: 1.3.0.dev20260614
-
+  version: 1.3.0.dev20260713
 frontend:
   type: dynamo
   enable_multiple_frontends: false
-
 backend:
   type: vllm
-  connector: null
+  connector:
   allow_prefill_decode_colocation: true
-
   prefill_environment:
     VLLM_FLOAT32_MATMUL_PRECISION: high
-    UCX_TLS: "cuda_ipc,cuda_copy,rc"
-
+    UCX_TLS: "cuda_copy,rc"
   decode_environment:
     VLLM_FLOAT32_MATMUL_PRECISION: high
-    UCX_TLS: "cuda_ipc,cuda_copy,rc"
-
+    UCX_TLS: "cuda_copy,rc"
   vllm_config:
     prefill:
       no-enable-flashinfer-autotune: true
       tensor-parallel-size: 1
-      data-parallel-size: 2
-      data-parallel-rpc-port: 13345
-      enable-expert-parallel: true
+      enable-expert-parallel: false
       trust-remote-code: true
       no-enable-prefix-caching: true
       kv-transfer-config: '{"kv_connector": "NixlConnector", "kv_role": "kv_both"}'
@@ -53,9 +43,8 @@ backend:
       max-model-len: 9472
       language-model-only: true
       stream-interval: 32
-      max-cudagraph-capture-size: 2048
-      max-num-batched-tokens: 16384
-
+      enforce-eager: true
+      max-num-batched-tokens: 8192
     decode:
       no-enable-flashinfer-autotune: true
       tensor-parallel-size: 4
@@ -70,17 +59,15 @@ backend:
       max-model-len: 9472
       language-model-only: true
       stream-interval: 32
-      max-num-seqs: 1024
+      max-num-seqs: 512
       max-num-batched-tokens: 16384
-      max-cudagraph-capture-size: 2048
-
+      max-cudagraph-capture-size: 4096
 health_check:
   max_attempts: 360
   interval_seconds: 10
-
 benchmark:
   type: "sa-bench"
   isl: 8192
   osl: 1024
-  concurrencies: "1x4x8x16"
+  concurrencies: "256"
   req_rate: "inf"
diff --git a/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/4p2d-dep2-dep8-8k1k.yaml b/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/4p2d-dep2-dep8-8k1k.yaml
deleted file mode 100644
index 73473aac94..0000000000
--- a/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/4p2d-dep2-dep8-8k1k.yaml
+++ /dev/null
@@ -1,87 +0,0 @@
-name: "minimax-m3-vllm-disagg-b300-4p2d-fp4-dep2-dep8-8k1k"
-
-model:
-  path: "nvidia/MiniMax-M3-NVFP4"
-  container: "vllm/vllm-openai:vllm-minimax-m3-perf-x86_64-13.0.1-8b00f41"
-  precision: "fp4"
-
-resources:
-  gpu_type: "b300"
-  gpus_per_node: 8
-  prefill_nodes: 1
-  decode_nodes: 2
-  prefill_workers: 4
-  decode_workers: 2
-  gpus_per_prefill: 2
-  gpus_per_decode: 8
-
-dynamo:
-  install: true
-  version: 1.3.0.dev20260614
-
-frontend:
-  type: dynamo
-  enable_multiple_frontends: false
-
-backend:
-  type: vllm
-  connector: null
-
-  prefill_environment:
-    VLLM_FLOAT32_MATMUL_PRECISION: high
-    UCX_TLS: "cuda_copy,rc"
-
-  decode_environment:
-    VLLM_FLOAT32_MATMUL_PRECISION: high
-    UCX_TLS: "cuda_copy,rc"
-
-  vllm_config:
-    prefill:
-      no-enable-flashinfer-autotune: true
-      tensor-parallel-size: 1
-      data-parallel-size: 2
-      data-parallel-rpc-port: 13345
-      enable-expert-parallel: true
-      trust-remote-code: true
-      no-enable-prefix-caching: true
-      kv-transfer-config: '{"kv_connector": "NixlConnector", "kv_role": "kv_both"}'
-      attention-config: '{"backend": "FLASHINFER", "use_trtllm_attention": true, "indexer_kv_dtype": "fp8"}'
-      kv-cache-dtype: fp8
-      block-size: 128
-      gpu-memory-utilization: 0.90
-      max-model-len: 9472
-      language-model-only: true
-      stream-interval: 32
-      max-cudagraph-capture-size: 2048
-      max-num-batched-tokens: 16384
-
-    decode:
-      no-enable-flashinfer-autotune: true
-      tensor-parallel-size: 1
-      data-parallel-size: 8
-      data-parallel-rpc-port: 13345
-      enable-expert-parallel: true
-      trust-remote-code: true
-      no-enable-prefix-caching: true
-      kv-transfer-config: '{"kv_connector": "NixlConnector", "kv_role": "kv_both"}'
-      attention-config: '{"backend": "FLASHINFER", "use_trtllm_attention": true, "indexer_kv_dtype": "fp8"}'
-      kv-cache-dtype: fp8
-      block-size: 128
-      gpu-memory-utilization: 0.90
-      max-model-len: 9472
-      language-model-only: true
-      stream-interval: 32
-      max-num-seqs: 1024  # Per DP rank: 2 workers x DP8 = 16 ranks.
-      max-num-batched-tokens: 16384
-      max-cudagraph-capture-size: 4096
-
-health_check:
-  max_attempts: 360
-  interval_seconds: 10
-
-benchmark:
-  type: "sa-bench"
-  isl: 8192
-  osl: 1024
-  concurrencies: "4096"
-  req_rate: "inf"
diff --git a/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/2p2d-dep2-dep8-8k1k.yaml b/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/5p2d-tp1-dep8-8k1k.yaml
similarity index 80%
rename from benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/2p2d-dep2-dep8-8k1k.yaml
rename to benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/5p2d-tp1-dep8-8k1k.yaml
index c49fd1ccbf..2374f1b421 100644
--- a/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/2p2d-dep2-dep8-8k1k.yaml
+++ b/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/5p2d-tp1-dep8-8k1k.yaml
@@ -1,47 +1,37 @@
-name: "minimax-m3-vllm-disagg-b300-2p2d-fp4-dep2-dep8-8k1k"
-
+name: "minimax-m3-vllm-disagg-b300-5p2d-tp1-dep8-fp4-8k1k"
 model:
   path: "nvidia/MiniMax-M3-NVFP4"
-  container: "vllm/vllm-openai:vllm-minimax-m3-perf-x86_64-13.0.1-8b00f41"
+  container: "vllm/vllm-openai:nightly-2dfaae752b4db0d43cfc0715c780e33be030d0f1"
   precision: "fp4"
-
 resources:
   gpu_type: "b300"
   gpus_per_node: 8
   prefill_nodes: 1
   decode_nodes: 2
-  prefill_workers: 2
+  prefill_workers: 5
   decode_workers: 2
-  gpus_per_prefill: 2
+  gpus_per_prefill: 1
   gpus_per_decode: 8
-
 dynamo:
   install: true
-  version: 1.3.0.dev20260614
-
+  version: 1.3.0.dev20260713
 frontend:
   type: dynamo
   enable_multiple_frontends: false
-
 backend:
   type: vllm
-  connector: null
-
+  connector:
   prefill_environment:
     VLLM_FLOAT32_MATMUL_PRECISION: high
     UCX_TLS: "cuda_copy,rc"
-
   decode_environment:
     VLLM_FLOAT32_MATMUL_PRECISION: high
     UCX_TLS: "cuda_copy,rc"
-
   vllm_config:
     prefill:
       no-enable-flashinfer-autotune: true
       tensor-parallel-size: 1
-      data-parallel-size: 2
-      data-parallel-rpc-port: 13345
-      enable-expert-parallel: true
+      enable-expert-parallel: false
       trust-remote-code: true
       no-enable-prefix-caching: true
       kv-transfer-config: '{"kv_connector": "NixlConnector", "kv_role": "kv_both"}'
@@ -52,9 +42,8 @@ backend:
       max-model-len: 9472
       language-model-only: true
       stream-interval: 32
-      max-cudagraph-capture-size: 2048
-      max-num-batched-tokens: 16384
-
+      enforce-eager: true
+      max-num-batched-tokens: 8192
     decode:
       no-enable-flashinfer-autotune: true
       tensor-parallel-size: 1
@@ -74,14 +63,12 @@ backend:
       max-num-seqs: 1024
       max-num-batched-tokens: 16384
       max-cudagraph-capture-size: 4096
-
 health_check:
   max_attempts: 360
   interval_seconds: 10
-
 benchmark:
   type: "sa-bench"
   isl: 8192
   osl: 1024
-  concurrencies: "256x512"
+  concurrencies: "512"
   req_rate: "inf"
diff --git a/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/2p2d-dep2-tep8-8k1k.yaml b/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/8p2d-tp1-dep8-8k1k.yaml
similarity index 80%
rename from benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/2p2d-dep2-tep8-8k1k.yaml
rename to benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/8p2d-tp1-dep8-8k1k.yaml
index 1b8dfd627f..88c756e071 100644
--- a/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/2p2d-dep2-tep8-8k1k.yaml
+++ b/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3/b300-fp4/8k1k/8p2d-tp1-dep8-8k1k.yaml
@@ -1,47 +1,37 @@
-name: "minimax-m3-vllm-disagg-b300-2p2d-fp4-dep2-tep8-8k1k"
-
+name: "minimax-m3-vllm-disagg-b300-8p2d-tp1-dep8-fp4-8k1k"
 model:
   path: "nvidia/MiniMax-M3-NVFP4"
-  container: "vllm/vllm-openai:vllm-minimax-m3-perf-x86_64-13.0.1-8b00f41"
+  container: "vllm/vllm-openai:nightly-2dfaae752b4db0d43cfc0715c780e33be030d0f1"
   precision: "fp4"
-
 resources:
   gpu_type: "b300"
   gpus_per_node: 8
   prefill_nodes: 1
   decode_nodes: 2
-  prefill_workers: 2
+  prefill_workers: 8
   decode_workers: 2
-  gpus_per_prefill: 2
+  gpus_per_prefill: 1
   gpus_per_decode: 8
-
 dynamo:
   install: true
-  version: 1.3.0.dev20260614
-
+  version: 1.3.0.dev20260713
 frontend:
   type: dynamo
   enable_multiple_frontends: false
-
 backend:
   type: vllm
-  connector: null
-
+  connector:
   prefill_environment:
     VLLM_FLOAT32_MATMUL_PRECISION: high
     UCX_TLS: "cuda_copy,rc"
-
   decode_environment:
     VLLM_FLOAT32_MATMUL_PRECISION: high
     UCX_TLS: "cuda_copy,rc"
-
   vllm_config:
     prefill:
       no-enable-flashinfer-autotune: true
       tensor-parallel-size: 1
-      data-parallel-size: 2
-      data-parallel-rpc-port: 13345
-      enable-expert-parallel: true
+      enable-expert-parallel: false
       trust-remote-code: true
       no-enable-prefix-caching: true
       kv-transfer-config: '{"kv_connector": "NixlConnector", "kv_role": "kv_both"}'
@@ -52,12 +42,13 @@ backend:
       max-model-len: 9472
       language-model-only: true
       stream-interval: 32
-      max-cudagraph-capture-size: 2048
-      max-num-batched-tokens: 16384
-
+      enforce-eager: true
+      max-num-batched-tokens: 8192
     decode:
       no-enable-flashinfer-autotune: true
-      tensor-parallel-size: 8
+      tensor-parallel-size: 1
+      data-parallel-size: 8
+      data-parallel-rpc-port: 13345
       enable-expert-parallel: true
       trust-remote-code: true
       no-enable-prefix-caching: true
@@ -72,14 +63,12 @@ backend:
       max-num-seqs: 1024
       max-num-batched-tokens: 16384
       max-cudagraph-capture-size: 4096
-
 health_check:
   max_attempts: 360
   interval_seconds: 10
-
 benchmark:
   type: "sa-bench"
   isl: 8192
   osl: 1024
-  concurrencies: "16"
+  concurrencies: "768x1024"
   req_rate: "inf"
diff --git a/configs/nvidia-master.yaml b/configs/nvidia-master.yaml
index 73327351e1..eda73b21ba 100644
--- a/configs/nvidia-master.yaml
+++ b/configs/nvidia-master.yaml
@@ -13054,112 +13054,133 @@ minimaxm3-fp4-b300-dynamo-vllm:
           tp: 8
           ep: 8
           dp-attn: false
+# MiniMax-M3 NVFP4 B300 8k1k disagg sweep with TP1 prefill. This is separate
+# from the legacy entry above so updating the 8k1k frontier does not rerun 1k1k.
+minimaxm3-fp4-b300-dynamo-vllm-8k1k-tp1:
+  image: vllm/vllm-openai:nightly-2dfaae752b4db0d43cfc0715c780e33be030d0f1
+  model: nvidia/MiniMax-M3-NVFP4
+  model-prefix: minimaxm3
+  runner: b300
+  precision: fp4
+  framework: dynamo-vllm
+  router: { name: dynamo-router, version: "1.3.0.dev20260713" }
+  kv-p2p-transfer: nixl
+  multinode: true
+  disagg: true
+  scenarios:
+    fixed-seq-len:
     - isl: 8192
       osl: 1024
       search-space:
-      - conc-list: [256, 512]
+      - conc-list: [8, 16, 24, 32, 48, 64]
         prefill:
-          num-worker: 2
-          tp: 2
-          ep: 2
-          dp-attn: true
+          num-worker: 1
+          tp: 1
+          ep: 1
+          dp-attn: false
           additional-settings:
-          - "CONFIG_FILE=recipes/vllm/minimax-m3/b300-fp4/8k1k/2p2d-dep2-dep8-8k1k.yaml"
+          - "CONFIG_FILE=recipes/vllm/minimax-m3/b300-fp4/8k1k/1p1d-tp1-tp4-8k1k.yaml"
         decode:
-          num-worker: 2
-          tp: 8
-          ep: 8
-          dp-attn: true
-      - conc-list: [16]
+          num-worker: 1
+          tp: 4
+          ep: 1
+          dp-attn: false
+      - conc-list: [1, 2, 4, 8, 16]
         prefill:
-          num-worker: 2
-          tp: 2
-          ep: 2
-          dp-attn: true
+          num-worker: 1
+          tp: 1
+          ep: 1
+          dp-attn: false
           additional-settings:
-          - "CONFIG_FILE=recipes/vllm/minimax-m3/b300-fp4/8k1k/2p2d-dep2-tep8-8k1k.yaml"
+          - "CONFIG_FILE=recipes/vllm/minimax-m3/b300-fp4/8k1k/1p1d-tp1-tp8-8k1k.yaml"
         decode:
-          num-worker: 2
+          num-worker: 1
           tp: 8
-          ep: 8
+          ep: 1
           dp-attn: false
-      - conc-list: [4096]
+      - conc-list: [128]
         prefill:
-          num-worker: 4
-          tp: 2
-          ep: 2
-          dp-attn: true
-          additional-settings:
-          - "CONFIG_FILE=recipes/vllm/minimax-m3/b300-fp4/8k1k/4p2d-dep2-dep8-8k1k.yaml"
-        decode:
           num-worker: 2
-          tp: 8
-          ep: 8
-          dp-attn: true
-      - conc-list: [1, 4, 8, 16]
-        prefill:
-          num-worker: 1
-          tp: 2
-          ep: 2
-          dp-attn: true
+          tp: 1
+          ep: 1
+          dp-attn: false
           additional-settings:
-          - "CONFIG_FILE=recipes/vllm/minimax-m3/b300-fp4/8k1k/1p1d-dep2-tp4-8k1k.yaml"
+          - "CONFIG_FILE=recipes/vllm/minimax-m3/b300-fp4/8k1k/2p1d-tp1-tp4-8k1k.yaml"
         decode:
           num-worker: 1
           tp: 4
           ep: 1
           dp-attn: false
-      - conc-list: [4096]
+      - conc-list: [256]
         prefill:
-          num-worker: 4
-          tp: 2
-          ep: 2
-          dp-attn: true
+          num-worker: 3
+          tp: 1
+          ep: 1
+          dp-attn: false
           additional-settings:
-          - "CONFIG_FILE=recipes/vllm/minimax-m3/b300-fp4/8k1k/4p2d-dep2-tep4-8k1k.yaml"
+          - "CONFIG_FILE=recipes/vllm/minimax-m3/b300-fp4/8k1k/3p1d-tp1-tp4-8k1k.yaml"
         decode:
-          num-worker: 2
+          num-worker: 1
           tp: 4
-          ep: 4
+          ep: 1
           dp-attn: false
-      - conc-list: [16, 32, 64, 128]
+      - conc-list: [512]
         prefill:
-          num-worker: 1
-          tp: 2
-          ep: 2
-          dp-attn: true
+          num-worker: 5
+          tp: 1
+          ep: 1
+          dp-attn: false
           additional-settings:
-          - "CONFIG_FILE=recipes/vllm/minimax-m3/b300-fp4/8k1k/1p4d-dep2-tep4-8k1k.yaml"
+          - "CONFIG_FILE=recipes/vllm/minimax-m3/b300-fp4/8k1k/5p2d-tp1-dep8-8k1k.yaml"
         decode:
-          num-worker: 4
-          tp: 4
-          ep: 4
-          dp-attn: false
-      - conc-list: [16]
-        prefill:
-          num-worker: 1
-          tp: 2
-          ep: 2
+          num-worker: 2
+          tp: 8
+          ep: 8
           dp-attn: true
+      - conc-list: [768, 1024]
+        prefill:
+          num-worker: 8
+          tp: 1
+          ep: 1
+          dp-attn: false
           additional-settings:
-          - "CONFIG_FILE=recipes/vllm/minimax-m3/b300-fp4/8k1k/1p2d-dep2-tep4-8k1k.yaml"
+          - "CONFIG_FILE=recipes/vllm/minimax-m3/b300-fp4/8k1k/8p2d-tp1-dep8-8k1k.yaml"
         decode:
           num-worker: 2
-          tp: 4
-          ep: 4
-          dp-attn: false
-      - conc-list: [4]
+          tp: 8
+          ep: 8
+          dp-attn: true
+
+# Preserve the previously proven max-throughput point exactly, while keeping
+# it separate from the legacy mixed 1k1k/8k1k entry so 1k1k is not rerun.
+minimaxm3-fp4-b300-dynamo-vllm-8k1k-legacy-max-tput:
+  image: vllm/vllm-openai:vllm-minimax-m3-perf-x86_64-13.0.1-8b00f41
+  model: nvidia/MiniMax-M3-NVFP4
+  model-prefix: minimaxm3
+  runner: b300
+  precision: fp4
+  framework: dynamo-vllm
+  router: { name: dynamo-router, version: "1.3.0.dev20260614" }
+  kv-p2p-transfer: nixl
+  multinode: true
+  disagg: true
+  scenarios:
+    fixed-seq-len:
+    - isl: 8192
+      osl: 1024
+      search-space:
+      - conc-list: [4096]
         prefill:
-          num-worker: 1
+          num-worker: 4
           tp: 2
           ep: 2
           dp-attn: true
           additional-settings:
-          - "CONFIG_FILE=recipes/vllm/minimax-m3/b300-fp4/8k1k/1p4d-dep2-tep8-8k1k.yaml"
+          - "CONFIG_FILE=recipes/vllm/minimax-m3/b300-fp4/8k1k/4p2d-dep2-tep4-8k1k.yaml"
         decode:
-          num-worker: 4
-          tp: 8
-          ep: 8
+          num-worker: 2
+          tp: 4
+          ep: 4
           dp-attn: false
 
 # MiniMax-M3 GB300 disagg sweep — refreshed recipe set (no Marlin variants).
diff --git a/perf-changelog.yaml b/perf-changelog.yaml
index cd0320fa99..24edc89e4c 100644
--- a/perf-changelog.yaml
+++ b/perf-changelog.yaml
@@ -4788,3 +4788,13 @@
     - "Add EAGLE3 speculative-decoding arm for the Kimi K2.6 NVFP4 B300 AgentX recipe (draft lightseekorg/kimi-k2.6-eagle3-mla, TOKENSPEED_MLA attention backend with TRT-LLM ragged MLA kernel)."
     - "TP8/TP4 GPU-only KV points plus a TP4 native CPU-offload ladder via SimpleCPUOffloadConnector with lazy_offload off; TP4/DCP4 high-concurrency points (conc 32/64) using num_speculative_tokens=3 and synthetic_acceptance_length=2.88."
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2222
+  
+- config-keys:
+    - minimaxm3-fp4-b300-dynamo-vllm-8k1k-tp1
+    - minimaxm3-fp4-b300-dynamo-vllm-8k1k-legacy-max-tput
+  description:
+    - "Replace the MiniMax-M3 NVFP4 B300 8k1k Dynamo-vLLM disaggregated DEP2-prefill sweep with TP1-prefill recipes, while preserving the exact prior max-throughput point; retain the existing 1k1k sweep unchanged."
+    - "Use vllm/vllm-openai:nightly-2dfaae752b4db0d43cfc0715c780e33be030d0f1 with Dynamo 1.3.0.dev20260713, and pin NVIDIA/srt-slurm at c1fb6989fc5aca803b4ca0f2d17d8be85fad9732."
+    - "Sweep TP4, TP8, and DEP8 decode topologies from concurrency 1 through 1024, with one-node prefill/decode colocation where the worker GPUs fit."
+    - "Retain the proven 4P2D DEP2-prefill/TEP4-decode concurrency-4096 point unchanged on vllm/vllm-openai:vllm-minimax-m3-perf-x86_64-13.0.1-8b00f41 and Dynamo 1.3.0.dev20260614."
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2184
diff --git a/runners/launch_b300-nv.sh b/runners/launch_b300-nv.sh
index 6810ee5d85..be9ff2410e 100644
--- a/runners/launch_b300-nv.sh
+++ b/runners/launch_b300-nv.sh
@@ -76,6 +76,12 @@ elif [[ $FRAMEWORK == "dynamo-vllm" && $MODEL_PREFIX == "dsv4" ]]; then
     git checkout aflowers/vllm-gb200-v0.20.0
     mkdir -p recipes/vllm/deepseek-v4
     cp -rT "$GITHUB_WORKSPACE/benchmarks/multi_node/srt-slurm-recipes/vllm/deepseek-v4" recipes/vllm/deepseek-v4
+elif [[ $FRAMEWORK == "dynamo-vllm" && $MODEL_PREFIX == "minimaxm3" && $PRECISION == "fp4" && "$CONFIG_FILE" == recipes/vllm/minimax-m3/b300-fp4/8k1k/*-tp1-*.yaml ]]; then
+    git clone https://github.com/NVIDIA/srt-slurm.git "$SRT_REPO_DIR"
+    cd "$SRT_REPO_DIR" || exit 1
+    git checkout c1fb6989fc5aca803b4ca0f2d17d8be85fad9732
+    mkdir -p recipes/vllm/minimax-m3
+    cp -rT "$GITHUB_WORKSPACE/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m3" recipes/vllm/minimax-m3
 elif [[ $FRAMEWORK == "dynamo-vllm" && $MODEL_PREFIX == "minimaxm3" && ( $PRECISION == "fp4" || $PRECISION == "fp8" ) ]]; then
     git clone https://github.com/NVIDIA/srt-slurm.git "$SRT_REPO_DIR"
     cd "$SRT_REPO_DIR" || exit 1
@@ -181,6 +187,14 @@ SRTCTL_APPLY_ARGS=(
     -f "$CONFIG_FILE"
     --tags "b300,${MODEL_PREFIX},${PRECISION},${ISL}x${OSL},infmax-$(date +%Y%m%d)"
 )
+# The TP1 8k1k recipes use a newer srt-slurm revision whose preflight checks
+# model.path on this GHA login host. MiniMax-M3 NVFP4 is intentionally staged
+# under compute-node-local /scratch (as in the original B300 submission), so
+# the login host cannot stat it even though workers can. Keep this bypass
+# scoped to the new recipe set; runtime model loading still validates the path.
+if [[ $FRAMEWORK == "dynamo-vllm" && $MODEL_PREFIX == "minimaxm3" && $PRECISION == "fp4" && "$CONFIG_FILE" == recipes/vllm/minimax-m3/b300-fp4/8k1k/*-tp1-*.yaml ]]; then
+    SRTCTL_APPLY_ARGS+=(--no-preflight)
+fi
 if [[ -n "$SRTCTL_SETUP_SCRIPT" ]]; then
     SRTCTL_APPLY_ARGS+=(--setup-script "$SRTCTL_SETUP_SCRIPT")
 fi