diff --git a/benchmarks/multi_node/srt-slurm-recipes/sglang/dsr1/b200-fp4/1k1k/disagg/mtp/1k1k_mtp_lowlat_0.yaml b/benchmarks/multi_node/srt-slurm-recipes/sglang/dsr1/b200-fp4/1k1k/disagg/mtp/1k1k_mtp_lowlat_0.yaml
new file mode 100644
index 0000000000..11f56972ee
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/sglang/dsr1/b200-fp4/1k1k/disagg/mtp/1k1k_mtp_lowlat_0.yaml
@@ -0,0 +1,123 @@
+# Derived from the srt-slurm b200-fp4 1k1k recipe (recipes/b200-fp4/1k1k.yaml
+# base + zip_override_mtp_lowlat[0]): 1p5d low-latency (dep4 prefill / tep8 decode, 5 decode nodes).
+# One flat YAML per concrete topology, matching the 8k1k local recipe layout
+# (sglang/<model>/<hw>-<precision>/<seq>/disagg/<variant>/...).
+
+name: b200-fp4-mtp-low-latency-dep4-1p-tep8-5d
+model:
+  path: dsr1
+  container: dynamo-sglang
+  precision: fp4
+
+dynamo:
+  hash: "5b4bc1dd70965017a737c71b19db5a0aeaa88727"
+  install: true
+
+resources:
+  gpu_type: b200
+  prefill_nodes: 1
+  prefill_workers: 1
+  gpus_per_prefill: 4
+  decode_nodes: 5
+  decode_workers: 5
+  gpus_per_node: 8
+backend:
+  prefill_environment:
+    TORCH_DISTRIBUTED_DEFAULT_TIMEOUT: '1800'
+    PYTHONUNBUFFERED: '1'
+    DYN_SKIP_SGLANG_LOG_FORMATTING: '1'
+    SGLANG_ENABLE_JIT_DEEPGEMM: 'false'
+    SGLANG_DISAGGREGATION_HEARTBEAT_MAX_FAILURE: '100000'
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: '100000'
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: '100000'
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: 'True'
+    SGLANG_USE_MESSAGE_QUEUE_BROADCASTER: '0'
+    SGLANG_DISABLE_TP_MEMORY_INBALANCE_CHECK: '1'
+    MC_FORCE_MNNVL: '1'
+    NCCL_MNNVL_ENABLE: '1'
+    NCCL_CUMEM_ENABLE: '1'
+    SGLANG_PER_TOKEN_GROUP_QUANT_8BIT_V2: '1'
+    DYN_REQUEST_PLANE: nats
+    SGLANG_ENABLE_SPEC_V2: '1'
+  decode_environment:
+    TORCH_DISTRIBUTED_DEFAULT_TIMEOUT: '1800'
+    PYTHONUNBUFFERED: '1'
+    DYN_SKIP_SGLANG_LOG_FORMATTING: '1'
+    SGLANG_ENABLE_JIT_DEEPGEMM: 'false'
+    SGLANG_DISAGGREGATION_HEARTBEAT_MAX_FAILURE: '100000'
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: '100000'
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: '100000'
+    SGLANG_DECODE_BOOTSTRAP_TIMEOUT: '1000'
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: 'True'
+    SGLANG_USE_MESSAGE_QUEUE_BROADCASTER: '0'
+    SGLANG_DISABLE_TP_MEMORY_INBALANCE_CHECK: '1'
+    MC_FORCE_MNNVL: '1'
+    NCCL_MNNVL_ENABLE: '1'
+    NCCL_CUMEM_ENABLE: '1'
+    SGLANG_PER_TOKEN_GROUP_QUANT_8BIT_V2: '1'
+    DYN_REQUEST_PLANE: nats
+    SGLANG_ENABLE_SPEC_V2: '1'
+  sglang_config:
+    prefill:
+      served-model-name: deepseek-ai/DeepSeek-R1
+      trust-remote-code: true
+      quantization: modelopt_fp4
+      disaggregation-mode: prefill
+      disaggregation-transfer-backend: nixl
+      mem-fraction-static: 0.85
+      max-prefill-tokens: 32768
+      chunked-prefill-size: 32768
+      context-length: 2200
+      max-running-requests: 512
+      disable-cuda-graph: true
+      tensor-parallel-size: 4
+      data-parallel-size: 4
+      expert-parallel-size: 4
+      enable-dp-attention: true
+      enable-dp-lm-head: true
+      attention-backend: trtllm_mla
+      kv-cache-dtype: fp8_e4m3
+      moe-runner-backend: flashinfer_trtllm
+      moe-dense-tp-size: 1
+      stream-interval: 30
+      watchdog-timeout: 1000000
+      enable-flashinfer-allreduce-fusion: true
+      disable-radix-cache: true
+      fp4-gemm-backend: flashinfer_trtllm
+    decode:
+      served-model-name: deepseek-ai/DeepSeek-R1
+      trust-remote-code: true
+      quantization: modelopt_fp4
+      disaggregation-mode: decode
+      disaggregation-transfer-backend: nixl
+      mem-fraction-static: 0.85
+      max-prefill-tokens: 32768
+      chunked-prefill-size: 32768
+      context-length: 2200
+      max-running-requests: 512
+      cuda-graph-max-bs: 512
+      tensor-parallel-size: 8
+      data-parallel-size: 1
+      expert-parallel-size: 8
+      attention-backend: trtllm_mla
+      kv-cache-dtype: fp8_e4m3
+      moe-runner-backend: flashinfer_trtllm
+      stream-interval: 30
+      watchdog-timeout: 1000000
+      enable-flashinfer-allreduce-fusion: true
+      disable-radix-cache: true
+      fp4-gemm-backend: flashinfer_trtllm
+      speculative-algorithm: EAGLE
+      speculative-num-steps: 2
+      speculative-eagle-topk: 1
+      speculative-num-draft-tokens: 3
+health_check:
+  max_attempts: 360
+  interval_seconds: 10
+benchmark:
+  type: sa-bench
+  isl: 1024
+  osl: 1024
+  req_rate: inf
+  concurrencies: 16x512
+  use_chat_template: true
diff --git a/benchmarks/multi_node/srt-slurm-recipes/sglang/dsr1/b200-fp4/1k1k/disagg/mtp/1k1k_mtp_lowlat_1.yaml b/benchmarks/multi_node/srt-slurm-recipes/sglang/dsr1/b200-fp4/1k1k/disagg/mtp/1k1k_mtp_lowlat_1.yaml
new file mode 100644
index 0000000000..acbec6f11a
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/sglang/dsr1/b200-fp4/1k1k/disagg/mtp/1k1k_mtp_lowlat_1.yaml
@@ -0,0 +1,123 @@
+# Derived from the srt-slurm b200-fp4 1k1k recipe (recipes/b200-fp4/1k1k.yaml
+# base + zip_override_mtp_lowlat[1]): 1p6d low-latency (dep4 prefill / tep8 decode, 6 decode nodes).
+# One flat YAML per concrete topology, matching the 8k1k local recipe layout
+# (sglang/<model>/<hw>-<precision>/<seq>/disagg/<variant>/...).
+
+name: b200-fp4-mtp-low-latency-dep4-1p-tep8-6d
+model:
+  path: dsr1
+  container: dynamo-sglang
+  precision: fp4
+
+dynamo:
+  hash: "5b4bc1dd70965017a737c71b19db5a0aeaa88727"
+  install: true
+
+resources:
+  gpu_type: b200
+  prefill_nodes: 1
+  prefill_workers: 1
+  gpus_per_prefill: 4
+  decode_nodes: 6
+  decode_workers: 6
+  gpus_per_node: 8
+backend:
+  prefill_environment:
+    TORCH_DISTRIBUTED_DEFAULT_TIMEOUT: '1800'
+    PYTHONUNBUFFERED: '1'
+    DYN_SKIP_SGLANG_LOG_FORMATTING: '1'
+    SGLANG_ENABLE_JIT_DEEPGEMM: 'false'
+    SGLANG_DISAGGREGATION_HEARTBEAT_MAX_FAILURE: '100000'
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: '100000'
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: '100000'
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: 'True'
+    SGLANG_USE_MESSAGE_QUEUE_BROADCASTER: '0'
+    SGLANG_DISABLE_TP_MEMORY_INBALANCE_CHECK: '1'
+    MC_FORCE_MNNVL: '1'
+    NCCL_MNNVL_ENABLE: '1'
+    NCCL_CUMEM_ENABLE: '1'
+    SGLANG_PER_TOKEN_GROUP_QUANT_8BIT_V2: '1'
+    DYN_REQUEST_PLANE: nats
+    SGLANG_ENABLE_SPEC_V2: '1'
+  decode_environment:
+    TORCH_DISTRIBUTED_DEFAULT_TIMEOUT: '1800'
+    PYTHONUNBUFFERED: '1'
+    DYN_SKIP_SGLANG_LOG_FORMATTING: '1'
+    SGLANG_ENABLE_JIT_DEEPGEMM: 'false'
+    SGLANG_DISAGGREGATION_HEARTBEAT_MAX_FAILURE: '100000'
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: '100000'
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: '100000'
+    SGLANG_DECODE_BOOTSTRAP_TIMEOUT: '1000'
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: 'True'
+    SGLANG_USE_MESSAGE_QUEUE_BROADCASTER: '0'
+    SGLANG_DISABLE_TP_MEMORY_INBALANCE_CHECK: '1'
+    MC_FORCE_MNNVL: '1'
+    NCCL_MNNVL_ENABLE: '1'
+    NCCL_CUMEM_ENABLE: '1'
+    SGLANG_PER_TOKEN_GROUP_QUANT_8BIT_V2: '1'
+    DYN_REQUEST_PLANE: nats
+    SGLANG_ENABLE_SPEC_V2: '1'
+  sglang_config:
+    prefill:
+      served-model-name: deepseek-ai/DeepSeek-R1
+      trust-remote-code: true
+      quantization: modelopt_fp4
+      disaggregation-mode: prefill
+      disaggregation-transfer-backend: nixl
+      mem-fraction-static: 0.85
+      max-prefill-tokens: 32768
+      chunked-prefill-size: 32768
+      context-length: 2200
+      max-running-requests: 512
+      disable-cuda-graph: true
+      tensor-parallel-size: 4
+      data-parallel-size: 4
+      expert-parallel-size: 4
+      enable-dp-attention: true
+      enable-dp-lm-head: true
+      attention-backend: trtllm_mla
+      kv-cache-dtype: fp8_e4m3
+      moe-runner-backend: flashinfer_trtllm
+      moe-dense-tp-size: 1
+      stream-interval: 30
+      watchdog-timeout: 1000000
+      enable-flashinfer-allreduce-fusion: true
+      disable-radix-cache: true
+      fp4-gemm-backend: flashinfer_trtllm
+    decode:
+      served-model-name: deepseek-ai/DeepSeek-R1
+      trust-remote-code: true
+      quantization: modelopt_fp4
+      disaggregation-mode: decode
+      disaggregation-transfer-backend: nixl
+      mem-fraction-static: 0.85
+      max-prefill-tokens: 32768
+      chunked-prefill-size: 32768
+      context-length: 2200
+      max-running-requests: 512
+      cuda-graph-max-bs: 512
+      tensor-parallel-size: 8
+      data-parallel-size: 1
+      expert-parallel-size: 8
+      attention-backend: trtllm_mla
+      kv-cache-dtype: fp8_e4m3
+      moe-runner-backend: flashinfer_trtllm
+      stream-interval: 30
+      watchdog-timeout: 1000000
+      enable-flashinfer-allreduce-fusion: true
+      disable-radix-cache: true
+      fp4-gemm-backend: flashinfer_trtllm
+      speculative-algorithm: EAGLE
+      speculative-num-steps: 2
+      speculative-eagle-topk: 1
+      speculative-num-draft-tokens: 3
+health_check:
+  max_attempts: 360
+  interval_seconds: 10
+benchmark:
+  type: sa-bench
+  isl: 1024
+  osl: 1024
+  req_rate: inf
+  concurrencies: 32x64x256x512
+  use_chat_template: true
diff --git a/benchmarks/multi_node/srt-slurm-recipes/sglang/dsr1/b200-fp4/1k1k/disagg/mtp/1k1k_mtp_maxtpt_0.yaml b/benchmarks/multi_node/srt-slurm-recipes/sglang/dsr1/b200-fp4/1k1k/disagg/mtp/1k1k_mtp_maxtpt_0.yaml
new file mode 100644
index 0000000000..5f67dfb50f
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/sglang/dsr1/b200-fp4/1k1k/disagg/mtp/1k1k_mtp_maxtpt_0.yaml
@@ -0,0 +1,128 @@
+# Derived from the srt-slurm b200-fp4 1k1k recipe (recipes/b200-fp4/1k1k.yaml
+# base + zip_override_mtp_maxtpt[0]): 1p1d max-throughput (dep4 prefill / dep8 decode, mem-fraction 0.75).
+# One flat YAML per concrete topology, matching the 8k1k local recipe layout
+# (sglang/<model>/<hw>-<precision>/<seq>/disagg/<variant>/...).
+
+name: b200-fp4-mtp-max-tpt-dep4-1p-dep8-1d
+model:
+  path: dsr1
+  container: dynamo-sglang
+  precision: fp4
+
+dynamo:
+  hash: "5b4bc1dd70965017a737c71b19db5a0aeaa88727"
+  install: true
+
+resources:
+  gpu_type: b200
+  prefill_nodes: 1
+  prefill_workers: 1
+  gpus_per_prefill: 4
+  decode_nodes: 1
+  decode_workers: 1
+  gpus_per_node: 8
+backend:
+  prefill_environment:
+    TORCH_DISTRIBUTED_DEFAULT_TIMEOUT: '1800'
+    PYTHONUNBUFFERED: '1'
+    DYN_SKIP_SGLANG_LOG_FORMATTING: '1'
+    SGLANG_ENABLE_JIT_DEEPGEMM: 'false'
+    SGLANG_DISAGGREGATION_HEARTBEAT_MAX_FAILURE: '100000'
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: '100000'
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: '100000'
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: 'True'
+    SGLANG_USE_MESSAGE_QUEUE_BROADCASTER: '0'
+    SGLANG_DISABLE_TP_MEMORY_INBALANCE_CHECK: '1'
+    MC_FORCE_MNNVL: '1'
+    NCCL_MNNVL_ENABLE: '1'
+    NCCL_CUMEM_ENABLE: '1'
+    SGLANG_PER_TOKEN_GROUP_QUANT_8BIT_V2: '1'
+    DYN_REQUEST_PLANE: nats
+    SGLANG_ENABLE_SPEC_V2: '1'
+  decode_environment:
+    TORCH_DISTRIBUTED_DEFAULT_TIMEOUT: '1800'
+    PYTHONUNBUFFERED: '1'
+    DYN_SKIP_SGLANG_LOG_FORMATTING: '1'
+    SGLANG_ENABLE_JIT_DEEPGEMM: 'false'
+    SGLANG_DISAGGREGATION_HEARTBEAT_MAX_FAILURE: '100000'
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: '100000'
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: '100000'
+    SGLANG_DECODE_BOOTSTRAP_TIMEOUT: '1000'
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: 'True'
+    SGLANG_USE_MESSAGE_QUEUE_BROADCASTER: '0'
+    SGLANG_DISABLE_TP_MEMORY_INBALANCE_CHECK: '1'
+    MC_FORCE_MNNVL: '1'
+    NCCL_MNNVL_ENABLE: '1'
+    NCCL_CUMEM_ENABLE: '1'
+    SGLANG_PER_TOKEN_GROUP_QUANT_8BIT_V2: '1'
+    DYN_REQUEST_PLANE: nats
+    SGLANG_MOE_NVFP4_DISPATCH: '1'
+    SGLANG_FLASHINFER_FP4_GEMM_BACKEND: cutlass
+    SGLANG_ENABLE_SPEC_V2: '1'
+  sglang_config:
+    prefill:
+      served-model-name: deepseek-ai/DeepSeek-R1
+      trust-remote-code: true
+      quantization: modelopt_fp4
+      disaggregation-mode: prefill
+      disaggregation-transfer-backend: nixl
+      mem-fraction-static: 0.85
+      max-prefill-tokens: 32768
+      chunked-prefill-size: 32768
+      context-length: 2200
+      max-running-requests: 1024
+      disable-cuda-graph: true
+      tensor-parallel-size: 4
+      data-parallel-size: 4
+      expert-parallel-size: 4
+      enable-dp-attention: true
+      enable-dp-lm-head: true
+      attention-backend: trtllm_mla
+      kv-cache-dtype: fp8_e4m3
+      moe-runner-backend: flashinfer_trtllm
+      moe-dense-tp-size: 1
+      stream-interval: 30
+      watchdog-timeout: 1000000
+      enable-flashinfer-allreduce-fusion: true
+      disable-radix-cache: true
+      fp4-gemm-backend: flashinfer_trtllm
+    decode:
+      served-model-name: deepseek-ai/DeepSeek-R1
+      trust-remote-code: true
+      quantization: modelopt_fp4
+      disaggregation-mode: decode
+      disaggregation-transfer-backend: nixl
+      mem-fraction-static: 0.75
+      max-prefill-tokens: 32768
+      chunked-prefill-size: 32768
+      context-length: 2200
+      max-running-requests: 1024
+      cuda-graph-max-bs: 1024
+      tensor-parallel-size: 8
+      data-parallel-size: 8
+      expert-parallel-size: 8
+      attention-backend: trtllm_mla
+      kv-cache-dtype: fp8_e4m3
+      moe-runner-backend: flashinfer_trtllm
+      stream-interval: 30
+      watchdog-timeout: 1000000
+      enable-flashinfer-allreduce-fusion: true
+      disable-radix-cache: true
+      fp4-gemm-backend: flashinfer_trtllm
+      enable-dp-attention: true
+      enable-dp-lm-head: true
+      moe-dense-tp-size: 1
+      speculative-algorithm: EAGLE
+      speculative-num-steps: 2
+      speculative-eagle-topk: 1
+      speculative-num-draft-tokens: 3
+health_check:
+  max_attempts: 360
+  interval_seconds: 10
+benchmark:
+  type: sa-bench
+  isl: 1024
+  osl: 1024
+  req_rate: inf
+  concurrencies: 512x1024
+  use_chat_template: true
diff --git a/benchmarks/multi_node/srt-slurm-recipes/sglang/dsr1/b200-fp4/1k1k/disagg/mtp/1k1k_mtp_maxtpt_1.yaml b/benchmarks/multi_node/srt-slurm-recipes/sglang/dsr1/b200-fp4/1k1k/disagg/mtp/1k1k_mtp_maxtpt_1.yaml
new file mode 100644
index 0000000000..36ceac6122
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/sglang/dsr1/b200-fp4/1k1k/disagg/mtp/1k1k_mtp_maxtpt_1.yaml
@@ -0,0 +1,128 @@
+# Derived from the srt-slurm b200-fp4 1k1k recipe (recipes/b200-fp4/1k1k.yaml
+# base + zip_override_mtp_maxtpt[1]): 1p2d max-throughput (dep4 prefill / dep8 decode, mem-fraction 0.85).
+# One flat YAML per concrete topology, matching the 8k1k local recipe layout
+# (sglang/<model>/<hw>-<precision>/<seq>/disagg/<variant>/...).
+
+name: b200-fp4-mtp-max-tpt-dep4-1p-dep8-2d
+model:
+  path: dsr1
+  container: dynamo-sglang
+  precision: fp4
+
+dynamo:
+  hash: "5b4bc1dd70965017a737c71b19db5a0aeaa88727"
+  install: true
+
+resources:
+  gpu_type: b200
+  prefill_nodes: 1
+  prefill_workers: 1
+  gpus_per_prefill: 4
+  decode_nodes: 2
+  decode_workers: 2
+  gpus_per_node: 8
+backend:
+  prefill_environment:
+    TORCH_DISTRIBUTED_DEFAULT_TIMEOUT: '1800'
+    PYTHONUNBUFFERED: '1'
+    DYN_SKIP_SGLANG_LOG_FORMATTING: '1'
+    SGLANG_ENABLE_JIT_DEEPGEMM: 'false'
+    SGLANG_DISAGGREGATION_HEARTBEAT_MAX_FAILURE: '100000'
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: '100000'
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: '100000'
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: 'True'
+    SGLANG_USE_MESSAGE_QUEUE_BROADCASTER: '0'
+    SGLANG_DISABLE_TP_MEMORY_INBALANCE_CHECK: '1'
+    MC_FORCE_MNNVL: '1'
+    NCCL_MNNVL_ENABLE: '1'
+    NCCL_CUMEM_ENABLE: '1'
+    SGLANG_PER_TOKEN_GROUP_QUANT_8BIT_V2: '1'
+    DYN_REQUEST_PLANE: nats
+    SGLANG_ENABLE_SPEC_V2: '1'
+  decode_environment:
+    TORCH_DISTRIBUTED_DEFAULT_TIMEOUT: '1800'
+    PYTHONUNBUFFERED: '1'
+    DYN_SKIP_SGLANG_LOG_FORMATTING: '1'
+    SGLANG_ENABLE_JIT_DEEPGEMM: 'false'
+    SGLANG_DISAGGREGATION_HEARTBEAT_MAX_FAILURE: '100000'
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: '100000'
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: '100000'
+    SGLANG_DECODE_BOOTSTRAP_TIMEOUT: '1000'
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: 'True'
+    SGLANG_USE_MESSAGE_QUEUE_BROADCASTER: '0'
+    SGLANG_DISABLE_TP_MEMORY_INBALANCE_CHECK: '1'
+    MC_FORCE_MNNVL: '1'
+    NCCL_MNNVL_ENABLE: '1'
+    NCCL_CUMEM_ENABLE: '1'
+    SGLANG_PER_TOKEN_GROUP_QUANT_8BIT_V2: '1'
+    DYN_REQUEST_PLANE: nats
+    SGLANG_MOE_NVFP4_DISPATCH: '1'
+    SGLANG_FLASHINFER_FP4_GEMM_BACKEND: cutlass
+    SGLANG_ENABLE_SPEC_V2: '1'
+  sglang_config:
+    prefill:
+      served-model-name: deepseek-ai/DeepSeek-R1
+      trust-remote-code: true
+      quantization: modelopt_fp4
+      disaggregation-mode: prefill
+      disaggregation-transfer-backend: nixl
+      mem-fraction-static: 0.85
+      max-prefill-tokens: 32768
+      chunked-prefill-size: 32768
+      context-length: 2200
+      max-running-requests: 512
+      disable-cuda-graph: true
+      tensor-parallel-size: 4
+      data-parallel-size: 4
+      expert-parallel-size: 4
+      enable-dp-attention: true
+      enable-dp-lm-head: true
+      attention-backend: trtllm_mla
+      kv-cache-dtype: fp8_e4m3
+      moe-runner-backend: flashinfer_trtllm
+      moe-dense-tp-size: 1
+      stream-interval: 30
+      watchdog-timeout: 1000000
+      enable-flashinfer-allreduce-fusion: true
+      disable-radix-cache: true
+      fp4-gemm-backend: flashinfer_trtllm
+    decode:
+      served-model-name: deepseek-ai/DeepSeek-R1
+      trust-remote-code: true
+      quantization: modelopt_fp4
+      disaggregation-mode: decode
+      disaggregation-transfer-backend: nixl
+      mem-fraction-static: 0.85
+      max-prefill-tokens: 32768
+      chunked-prefill-size: 32768
+      context-length: 2200
+      max-running-requests: 512
+      cuda-graph-max-bs: 512
+      tensor-parallel-size: 8
+      data-parallel-size: 8
+      expert-parallel-size: 8
+      attention-backend: trtllm_mla
+      kv-cache-dtype: fp8_e4m3
+      moe-runner-backend: flashinfer_trtllm
+      stream-interval: 30
+      watchdog-timeout: 1000000
+      enable-flashinfer-allreduce-fusion: true
+      disable-radix-cache: true
+      fp4-gemm-backend: flashinfer_trtllm
+      enable-dp-attention: true
+      enable-dp-lm-head: true
+      moe-dense-tp-size: 1
+      speculative-algorithm: EAGLE
+      speculative-num-steps: 2
+      speculative-eagle-topk: 1
+      speculative-num-draft-tokens: 3
+health_check:
+  max_attempts: 360
+  interval_seconds: 10
+benchmark:
+  type: sa-bench
+  isl: 1024
+  osl: 1024
+  req_rate: inf
+  concurrencies: '512'
+  use_chat_template: true
diff --git a/benchmarks/multi_node/srt-slurm-recipes/sglang/dsr1/b200-fp4/8k1k/disagg/mtp/8k1k_mtp2_throughput_1p1d.yaml b/benchmarks/multi_node/srt-slurm-recipes/sglang/dsr1/b200-fp4/8k1k/disagg/mtp/8k1k_mtp2_throughput_1p1d.yaml
new file mode 100644
index 0000000000..13a1a7d9d0
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/sglang/dsr1/b200-fp4/8k1k/disagg/mtp/8k1k_mtp2_throughput_1p1d.yaml
@@ -0,0 +1,143 @@
+name: "b200-fp4-mtp2-throughput-1p-dep4-1d-dep8"
+
+# Derived from the srt-slurm b200-fp4 8k1k recipe (recipes/b200-fp4/8k1k.yaml
+# base + override_mtp2_throughput_1p1d, DEP4 prefill / DEP8 decode, MTP2).
+# One flat YAML per concrete topology, matching the InferenceX glm5 disagg
+# layout (sglang/<model>/<hw>-<precision>/<seq>/disagg/<variant>/...).
+
+dynamo:
+  hash: "5b4bc1dd70965017a737c71b19db5a0aeaa88727"
+  install: true
+
+frontend:
+  type: dynamo
+  enable_multiple_frontends: true
+  num_additional_frontends: 3
+  nginx_container: nginx-sqsh
+
+model:
+  path: "dsr1"
+  container: "lmsysorg/sglang:v0.5.12.post1"
+  precision: "fp4"
+
+resources:
+  gpu_type: "b200"
+  prefill_nodes: 1
+  decode_nodes: 1
+  prefill_workers: 1
+  decode_workers: 1
+  gpus_per_prefill: 4
+  gpus_per_node: 8
+
+backend:
+  prefill_environment:
+    TORCH_DISTRIBUTED_DEFAULT_TIMEOUT: "1800"
+    PYTHONUNBUFFERED: "1"
+    DYN_SKIP_SGLANG_LOG_FORMATTING: "1"
+    SGLANG_USE_MESSAGE_QUEUE_BROADCASTER: "0"
+    SGLANG_DISABLE_TP_MEMORY_INBALANCE_CHECK: "1"
+    SGLANG_DISAGGREGATION_HEARTBEAT_MAX_FAILURE: "100000"
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: "100000"
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: "100000"
+    SGLANG_DECODE_BOOTSTRAP_TIMEOUT: "1000"
+    MC_FORCE_MNNVL: "1"
+    NCCL_MNNVL_ENABLE: "1"
+    NCCL_CUMEM_ENABLE: "1"
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: "True"
+    SGLANG_ENABLE_JIT_DEEPGEMM: "false"
+    SGLANG_ENABLE_SPEC_V2: "1"
+    UCX_TLS: "rc,cuda_ipc,cuda_copy,tcp,self"
+
+  decode_environment:
+    TORCH_DISTRIBUTED_DEFAULT_TIMEOUT: "1800"
+    PYTHONUNBUFFERED: "1"
+    DYN_SKIP_SGLANG_LOG_FORMATTING: "1"
+    SGLANG_USE_MESSAGE_QUEUE_BROADCASTER: "0"
+    SGLANG_DISABLE_TP_MEMORY_INBALANCE_CHECK: "1"
+    SGLANG_DISAGGREGATION_HEARTBEAT_MAX_FAILURE: "100000"
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: "100000"
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: "100000"
+    SGLANG_DECODE_BOOTSTRAP_TIMEOUT: "1000"
+    MC_FORCE_MNNVL: "1"
+    NCCL_MNNVL_ENABLE: "1"
+    NCCL_CUMEM_ENABLE: "1"
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: "True"
+    SGLANG_ENABLE_JIT_DEEPGEMM: "false"
+    SGLANG_ENABLE_SPEC_V2: "1"
+    UCX_TLS: "rc,cuda_ipc,cuda_copy,tcp,self"
+
+  sglang_config:
+    prefill:
+      disaggregation-mode: "prefill"
+      served-model-name: "deepseek-ai/DeepSeek-R1"
+      trust-remote-code: true
+      disable-radix-cache: true
+      kv-cache-dtype: "fp8_e4m3"
+      attention-backend: "trtllm_mla"
+      quantization: "modelopt_fp4"
+      moe-runner-backend: "flashinfer_trtllm"
+      stream-interval: 50
+      watchdog-timeout: 1000000
+      context-length: 9600
+      mem-fraction-static: 0.95
+      max-total-tokens: 32768
+      chunked-prefill-size: 24576
+      cuda-graph-max-bs: 256
+      max-running-requests: 2048
+      scheduler-recv-interval: 1
+      load-balance-method: "round_robin"
+      disaggregation-bootstrap-port: null
+      data-parallel-size: 4
+      tensor-parallel-size: 4
+      expert-parallel-size: 4
+      enable-dp-attention: true
+      enable-dp-lm-head: true
+      fp4-gemm-backend: "flashinfer_trtllm"
+      disaggregation-transfer-backend: nixl
+      speculative-algorithm: "EAGLE"
+      speculative-num-steps: 2
+      speculative-eagle-topk: 1
+      speculative-num-draft-tokens: 3
+
+    decode:
+      disaggregation-mode: "decode"
+      served-model-name: "deepseek-ai/DeepSeek-R1"
+      prefill-round-robin-balance: true
+      trust-remote-code: true
+      disable-radix-cache: true
+      kv-cache-dtype: "fp8_e4m3"
+      attention-backend: "trtllm_mla"
+      quantization: "modelopt_fp4"
+      moe-runner-backend: "flashinfer_trtllm"
+      disaggregation-bootstrap-port: 30001
+      stream-interval: 50
+      watchdog-timeout: 1000000
+      context-length: 9600
+      mem-fraction-static: 0.85
+      chunked-prefill-size: 8192
+      cuda-graph-max-bs: 1024
+      max-running-requests: 2048
+      scheduler-recv-interval: 1
+      data-parallel-size: 8
+      tensor-parallel-size: 8
+      expert-parallel-size: 8
+      enable-dp-attention: true
+      enable-dp-lm-head: true
+      fp4-gemm-backend: "flashinfer_trtllm"
+      disaggregation-transfer-backend: nixl
+      speculative-algorithm: "EAGLE"
+      speculative-num-steps: 2
+      speculative-eagle-topk: 1
+      speculative-num-draft-tokens: 3
+
+health_check:
+  max_attempts: 360
+  interval_seconds: 10
+
+benchmark:
+  type: "sa-bench"
+  isl: 8192
+  osl: 1024
+  req_rate: 300
+  concurrencies: "512"
+  use_chat_template: true
diff --git a/benchmarks/multi_node/srt-slurm-recipes/sglang/dsr1/b200-fp4/8k1k/disagg/mtp/8k1k_mtp2_throughput_4p1d.yaml b/benchmarks/multi_node/srt-slurm-recipes/sglang/dsr1/b200-fp4/8k1k/disagg/mtp/8k1k_mtp2_throughput_4p1d.yaml
new file mode 100644
index 0000000000..a3923b02db
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/sglang/dsr1/b200-fp4/8k1k/disagg/mtp/8k1k_mtp2_throughput_4p1d.yaml
@@ -0,0 +1,143 @@
+name: "b200-fp4-mtp2-throughput-4p-dep4-1d-dep8"
+
+# Derived from the srt-slurm b200-fp4 8k1k recipe (recipes/b200-fp4/8k1k.yaml
+# base + override_mtp2_throughput_4p1d, DEP4 prefill / DEP8 decode, MTP2).
+# One flat YAML per concrete topology, matching the InferenceX glm5 disagg
+# layout (sglang/<model>/<hw>-<precision>/<seq>/disagg/<variant>/...).
+
+dynamo:
+  hash: "5b4bc1dd70965017a737c71b19db5a0aeaa88727"
+  install: true
+
+frontend:
+  type: dynamo
+  enable_multiple_frontends: true
+  num_additional_frontends: 3
+  nginx_container: nginx-sqsh
+
+model:
+  path: "dsr1"
+  container: "lmsysorg/sglang:v0.5.12.post1"
+  precision: "fp4"
+
+resources:
+  gpu_type: "b200"
+  prefill_nodes: 2
+  decode_nodes: 1
+  prefill_workers: 4
+  decode_workers: 1
+  gpus_per_prefill: 4
+  gpus_per_node: 8
+
+backend:
+  prefill_environment:
+    TORCH_DISTRIBUTED_DEFAULT_TIMEOUT: "1800"
+    PYTHONUNBUFFERED: "1"
+    DYN_SKIP_SGLANG_LOG_FORMATTING: "1"
+    SGLANG_USE_MESSAGE_QUEUE_BROADCASTER: "0"
+    SGLANG_DISABLE_TP_MEMORY_INBALANCE_CHECK: "1"
+    SGLANG_DISAGGREGATION_HEARTBEAT_MAX_FAILURE: "100000"
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: "100000"
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: "100000"
+    SGLANG_DECODE_BOOTSTRAP_TIMEOUT: "1000"
+    MC_FORCE_MNNVL: "1"
+    NCCL_MNNVL_ENABLE: "1"
+    NCCL_CUMEM_ENABLE: "1"
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: "True"
+    SGLANG_ENABLE_JIT_DEEPGEMM: "false"
+    SGLANG_ENABLE_SPEC_V2: "1"
+    UCX_TLS: "rc,cuda_ipc,cuda_copy,tcp,self"
+
+  decode_environment:
+    TORCH_DISTRIBUTED_DEFAULT_TIMEOUT: "1800"
+    PYTHONUNBUFFERED: "1"
+    DYN_SKIP_SGLANG_LOG_FORMATTING: "1"
+    SGLANG_USE_MESSAGE_QUEUE_BROADCASTER: "0"
+    SGLANG_DISABLE_TP_MEMORY_INBALANCE_CHECK: "1"
+    SGLANG_DISAGGREGATION_HEARTBEAT_MAX_FAILURE: "100000"
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: "100000"
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: "100000"
+    SGLANG_DECODE_BOOTSTRAP_TIMEOUT: "1000"
+    MC_FORCE_MNNVL: "1"
+    NCCL_MNNVL_ENABLE: "1"
+    NCCL_CUMEM_ENABLE: "1"
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: "True"
+    SGLANG_ENABLE_JIT_DEEPGEMM: "false"
+    SGLANG_ENABLE_SPEC_V2: "1"
+    UCX_TLS: "rc,cuda_ipc,cuda_copy,tcp,self"
+
+  sglang_config:
+    prefill:
+      disaggregation-mode: "prefill"
+      served-model-name: "deepseek-ai/DeepSeek-R1"
+      trust-remote-code: true
+      disable-radix-cache: true
+      kv-cache-dtype: "fp8_e4m3"
+      attention-backend: "trtllm_mla"
+      quantization: "modelopt_fp4"
+      moe-runner-backend: "flashinfer_trtllm"
+      stream-interval: 50
+      watchdog-timeout: 1000000
+      context-length: 9600
+      mem-fraction-static: 0.95
+      max-total-tokens: 32768
+      chunked-prefill-size: 24576
+      cuda-graph-max-bs: 256
+      max-running-requests: 2048
+      scheduler-recv-interval: 1
+      load-balance-method: "round_robin"
+      disaggregation-bootstrap-port: null
+      data-parallel-size: 4
+      tensor-parallel-size: 4
+      expert-parallel-size: 4
+      enable-dp-attention: true
+      enable-dp-lm-head: true
+      fp4-gemm-backend: "flashinfer_trtllm"
+      disaggregation-transfer-backend: nixl
+      speculative-algorithm: "EAGLE"
+      speculative-num-steps: 2
+      speculative-eagle-topk: 1
+      speculative-num-draft-tokens: 3
+
+    decode:
+      disaggregation-mode: "decode"
+      served-model-name: "deepseek-ai/DeepSeek-R1"
+      prefill-round-robin-balance: true
+      trust-remote-code: true
+      disable-radix-cache: true
+      kv-cache-dtype: "fp8_e4m3"
+      attention-backend: "trtllm_mla"
+      quantization: "modelopt_fp4"
+      moe-runner-backend: "flashinfer_trtllm"
+      disaggregation-bootstrap-port: 30001
+      stream-interval: 50
+      watchdog-timeout: 1000000
+      context-length: 9600
+      mem-fraction-static: 0.85
+      chunked-prefill-size: 8192
+      cuda-graph-max-bs: 1024
+      max-running-requests: 2048
+      scheduler-recv-interval: 1
+      data-parallel-size: 8
+      tensor-parallel-size: 8
+      expert-parallel-size: 8
+      enable-dp-attention: true
+      enable-dp-lm-head: true
+      fp4-gemm-backend: "flashinfer_trtllm"
+      disaggregation-transfer-backend: nixl
+      speculative-algorithm: "EAGLE"
+      speculative-num-steps: 2
+      speculative-eagle-topk: 1
+      speculative-num-draft-tokens: 3
+
+health_check:
+  max_attempts: 360
+  interval_seconds: 10
+
+benchmark:
+  type: "sa-bench"
+  isl: 8192
+  osl: 1024
+  req_rate: 300
+  concurrencies: "512"
+  use_chat_template: true
diff --git a/benchmarks/multi_node/srt-slurm-recipes/sglang/dsr1/b200-fp4/8k1k/disagg/mtp/8k1k_mtp_4p1d_c2048.yaml b/benchmarks/multi_node/srt-slurm-recipes/sglang/dsr1/b200-fp4/8k1k/disagg/mtp/8k1k_mtp_4p1d_c2048.yaml
new file mode 100644
index 0000000000..fc2b0f3366
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/sglang/dsr1/b200-fp4/8k1k/disagg/mtp/8k1k_mtp_4p1d_c2048.yaml
@@ -0,0 +1,119 @@
+# DeepSeek-R1-0528 FP4 8k/1k B200 dynamo-sglang disagg — MTP, 4P:1D (c2048)
+# Topology: 4 prefill workers (TP4/EP4 dp-attn) + 1 decode worker (TP8/EP8 dp-attn) = 24 GPU.
+# Dedicated high-concurrency tune:
+#   SGLANG_HACK_SEQ_BOOTSTRAP_ROOM performance knob (not output-affecting; standard in GB200 recipes).
+#   max-running-requests raised to 2048 to avoid admission throttle at high concurrency.
+#   scheduler-recv-interval: 1 on decode.
+#   symm-mem + allreduce fusion retained.
+#   prefill stream-interval: 100; decode stream-interval: 34 (tuned for the c2048 batch cadence).
+name: b200-fp4-dsr1_8k1k_mtp_4p1d_c2048
+model:
+  path: "dsr1"
+  container: "lmsysorg/sglang:v0.5.12.post1"
+  precision: fp4
+resources:
+  gpu_type: b200
+  gpus_per_node: 8
+  prefill_nodes: 2
+  prefill_workers: 4
+  gpus_per_prefill: 4
+  decode_nodes: 1
+  decode_workers: 1
+frontend:
+  type: dynamo
+dynamo:
+  hash: "5b4bc1dd70965017a737c71b19db5a0aeaa88727"
+  install: true
+backend:
+  prefill_environment: &env
+    TORCH_DISTRIBUTED_DEFAULT_TIMEOUT: '1800'
+    PYTHONUNBUFFERED: '1'
+    DYN_SKIP_SGLANG_LOG_FORMATTING: '1'
+    DYN_REQUEST_PLANE: nats
+    SGLANG_ENABLE_JIT_DEEPGEMM: 'false'
+    SGLANG_DISAGGREGATION_HEARTBEAT_MAX_FAILURE: '100000'
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: '100000'
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: '100000'
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: 'True'
+    SGLANG_USE_MESSAGE_QUEUE_BROADCASTER: '0'
+    SGLANG_DISABLE_TP_MEMORY_INBALANCE_CHECK: '1'
+    SGLANG_PER_TOKEN_GROUP_QUANT_8BIT_V2: '1'
+    SGLANG_ENABLE_SPEC_V2: '1'
+    SGLANG_MOE_NVFP4_DISPATCH: '1'
+    SGLANG_FLASHINFER_FP4_GEMM_BACKEND: cutlass
+    MC_FORCE_MNNVL: '1'
+    NCCL_MNNVL_ENABLE: '1'
+    NCCL_CUMEM_ENABLE: '1'
+    SGLANG_NCCL_ALL_GATHER_IN_OVERLAP_SCHEDULER_SYNC_BATCH: '1'
+    SGLANG_HACK_SEQ_BOOTSTRAP_ROOM: '1'
+  decode_environment: *env
+  sglang_config:
+    prefill:
+      served-model-name: deepseek-ai/DeepSeek-R1
+      trust-remote-code: true
+      quantization: modelopt_fp4
+      disaggregation-mode: prefill
+      load-balance-method: round_robin
+      disaggregation-transfer-backend: nixl
+      mem-fraction-static: 0.85
+      max-prefill-tokens: 32768
+      chunked-prefill-size: 32768
+      context-length: 9600
+      max-running-requests: 2048
+      disable-cuda-graph: true
+      tensor-parallel-size: 4
+      data-parallel-size: 4
+      expert-parallel-size: 4
+      enable-dp-attention: true
+      enable-dp-lm-head: true
+      attention-backend: trtllm_mla
+      kv-cache-dtype: fp8_e4m3
+      moe-runner-backend: flashinfer_trtllm
+      moe-dense-tp-size: 1
+      enable-flashinfer-allreduce-fusion: true
+      enable-symm-mem: true
+      disable-radix-cache: true
+      stream-interval: 100
+      watchdog-timeout: 1000000
+    decode:
+      served-model-name: deepseek-ai/DeepSeek-R1
+      trust-remote-code: true
+      quantization: modelopt_fp4
+      disaggregation-mode: decode
+      prefill-round-robin-balance: true
+      disaggregation-transfer-backend: nixl
+      mem-fraction-static: 0.75
+      max-prefill-tokens: 32768
+      chunked-prefill-size: 32768
+      context-length: 9600
+      max-running-requests: 2048
+      cuda-graph-max-bs: 1024
+      tensor-parallel-size: 8
+      data-parallel-size: 8
+      expert-parallel-size: 8
+      enable-dp-attention: true
+      enable-dp-lm-head: true
+      attention-backend: trtllm_mla
+      kv-cache-dtype: fp8_e4m3
+      moe-runner-backend: flashinfer_trtllm
+      moe-dense-tp-size: 1
+      enable-flashinfer-allreduce-fusion: true
+      enable-symm-mem: true
+      disable-radix-cache: true
+      stream-interval: 34
+      scheduler-recv-interval: 1
+      watchdog-timeout: 1000000
+      speculative-algorithm: EAGLE
+      speculative-num-steps: 2
+      speculative-eagle-topk: 1
+      speculative-num-draft-tokens: 3
+health_check:
+  max_attempts: 360
+  interval_seconds: 10
+benchmark:
+  type: sa-bench
+  req_rate: inf
+  isl: 8192
+  osl: 1024
+  concurrencies: '2048'
+  use_chat_template: true
diff --git a/benchmarks/multi_node/srt-slurm-recipes/sglang/dsr1/b200-fp4/8k1k/disagg/mtp/8k1k_mtp_lowlat_0.yaml b/benchmarks/multi_node/srt-slurm-recipes/sglang/dsr1/b200-fp4/8k1k/disagg/mtp/8k1k_mtp_lowlat_0.yaml
index 160854ebc4..62bbddf84d 100644
--- a/benchmarks/multi_node/srt-slurm-recipes/sglang/dsr1/b200-fp4/8k1k/disagg/mtp/8k1k_mtp_lowlat_0.yaml
+++ b/benchmarks/multi_node/srt-slurm-recipes/sglang/dsr1/b200-fp4/8k1k/disagg/mtp/8k1k_mtp_lowlat_0.yaml
@@ -136,5 +136,5 @@ benchmark:
   isl: 8192
   osl: 1024
   req_rate: 300
-  concurrencies: "4x8x16x32x64"
+  concurrencies: "4x8x16x32"
   use_chat_template: true
diff --git a/benchmarks/multi_node/srt-slurm-recipes/sglang/dsr1/b200-fp4/8k1k/disagg/mtp/8k1k_mtp_lowlat_1.yaml b/benchmarks/multi_node/srt-slurm-recipes/sglang/dsr1/b200-fp4/8k1k/disagg/mtp/8k1k_mtp_lowlat_1.yaml
index fd1d4a4f53..521eddf931 100644
--- a/benchmarks/multi_node/srt-slurm-recipes/sglang/dsr1/b200-fp4/8k1k/disagg/mtp/8k1k_mtp_lowlat_1.yaml
+++ b/benchmarks/multi_node/srt-slurm-recipes/sglang/dsr1/b200-fp4/8k1k/disagg/mtp/8k1k_mtp_lowlat_1.yaml
@@ -136,5 +136,5 @@ benchmark:
   isl: 8192
   osl: 1024
   req_rate: 300
-  concurrencies: "32x64x128x256x512"
+  concurrencies: "32x64"
   use_chat_template: true
diff --git a/benchmarks/multi_node/srt-slurm-recipes/sglang/dsr1/b200-fp4/8k1k/disagg/mtp/8k1k_mtp_lowlat_2.yaml b/benchmarks/multi_node/srt-slurm-recipes/sglang/dsr1/b200-fp4/8k1k/disagg/mtp/8k1k_mtp_lowlat_2.yaml
index bcceaf872d..9dfc6331a8 100644
--- a/benchmarks/multi_node/srt-slurm-recipes/sglang/dsr1/b200-fp4/8k1k/disagg/mtp/8k1k_mtp_lowlat_2.yaml
+++ b/benchmarks/multi_node/srt-slurm-recipes/sglang/dsr1/b200-fp4/8k1k/disagg/mtp/8k1k_mtp_lowlat_2.yaml
@@ -136,5 +136,5 @@ benchmark:
   isl: 8192
   osl: 1024
   req_rate: 300
-  concurrencies: "32x64x128x256x512"
+  concurrencies: "32"
   use_chat_template: true
diff --git a/configs/nvidia-master.yaml b/configs/nvidia-master.yaml
index b6f68c6483..bc5dfb7939 100644
--- a/configs/nvidia-master.yaml
+++ b/configs/nvidia-master.yaml
@@ -7480,7 +7480,7 @@ dsr1-fp4-b200-dynamo-sglang-mtp:
           ep: 4
           dp-attn: true
           additional-settings:
-          - "CONFIG_FILE=recipes/b200-fp4/1k1k.yaml:zip_override_mtp_lowlat[0]"
+          - "CONFIG_FILE=recipes/sglang/dsr1/b200-fp4/1k1k/disagg/mtp/1k1k_mtp_lowlat_0.yaml"
         decode:
           num-worker: 5
           tp: 8
@@ -7494,7 +7494,7 @@ dsr1-fp4-b200-dynamo-sglang-mtp:
           ep: 4
           dp-attn: true
           additional-settings:
-          - "CONFIG_FILE=recipes/b200-fp4/1k1k.yaml:zip_override_mtp_lowlat[1]"
+          - "CONFIG_FILE=recipes/sglang/dsr1/b200-fp4/1k1k/disagg/mtp/1k1k_mtp_lowlat_1.yaml"
         decode:
           num-worker: 6
           tp: 8
@@ -7508,7 +7508,7 @@ dsr1-fp4-b200-dynamo-sglang-mtp:
           ep: 4
           dp-attn: true
           additional-settings:
-          - "CONFIG_FILE=recipes/b200-fp4/1k1k.yaml:zip_override_mtp_maxtpt[0]"
+          - "CONFIG_FILE=recipes/sglang/dsr1/b200-fp4/1k1k/disagg/mtp/1k1k_mtp_maxtpt_0.yaml"
         decode:
           num-worker: 1
           tp: 8
@@ -7522,7 +7522,7 @@ dsr1-fp4-b200-dynamo-sglang-mtp:
           ep: 4
           dp-attn: true
           additional-settings:
-          - "CONFIG_FILE=recipes/b200-fp4/1k1k.yaml:zip_override_mtp_maxtpt[1]"
+          - "CONFIG_FILE=recipes/sglang/dsr1/b200-fp4/1k1k/disagg/mtp/1k1k_mtp_maxtpt_1.yaml"
         decode:
           num-worker: 2
           tp: 8
@@ -7533,7 +7533,7 @@ dsr1-fp4-b200-dynamo-sglang-mtp:
       search-space:
       # 1p5d low-latency (decode-heavy).
       - spec-decoding: "mtp"
-        conc-list: [4, 8, 16, 32, 64]
+        conc-list: [4, 8, 16, 32]
         prefill:
           num-worker: 1
           tp: 4
@@ -7548,7 +7548,7 @@ dsr1-fp4-b200-dynamo-sglang-mtp:
           dp-attn: false
       # 1p3d low-latency.
       - spec-decoding: "mtp"
-        conc-list: [32, 64, 128, 256, 512]
+        conc-list: [32, 64]
         prefill:
           num-worker: 1
           tp: 4
@@ -7563,7 +7563,7 @@ dsr1-fp4-b200-dynamo-sglang-mtp:
           dp-attn: false
       # 1p1d low-latency.
       - spec-decoding: "mtp"
-        conc-list: [32, 64, 128, 256, 512]
+        conc-list: [32]
         prefill:
           num-worker: 1
           tp: 4
@@ -7577,6 +7577,21 @@ dsr1-fp4-b200-dynamo-sglang-mtp:
           ep: 1
           dp-attn: false
       # MTP2 high-throughput (DEP4 prefill / DEP8 decode), one Pareto point each.
+      # 1p1d throughput.
+      - spec-decoding: "mtp"
+        conc-list: [512]
+        prefill:
+          num-worker: 1
+          tp: 4
+          ep: 4
+          dp-attn: true
+          additional-settings:
+          - "CONFIG_FILE=recipes/sglang/dsr1/b200-fp4/8k1k/disagg/mtp/8k1k_mtp2_throughput_1p1d.yaml"
+        decode:
+          num-worker: 1
+          tp: 8
+          ep: 8
+          dp-attn: true
       # 2p1d throughput.
       - spec-decoding: "mtp"
         conc-list: [768]
@@ -7607,6 +7622,21 @@ dsr1-fp4-b200-dynamo-sglang-mtp:
           tp: 8
           ep: 8
           dp-attn: true
+      # 4p1d throughput.
+      - spec-decoding: "mtp"
+        conc-list: [512]
+        prefill:
+          num-worker: 4
+          tp: 4
+          ep: 4
+          dp-attn: true
+          additional-settings:
+          - "CONFIG_FILE=recipes/sglang/dsr1/b200-fp4/8k1k/disagg/mtp/8k1k_mtp2_throughput_4p1d.yaml"
+        decode:
+          num-worker: 1
+          tp: 8
+          ep: 8
+          dp-attn: true
       # 5p1d throughput.
       - spec-decoding: "mtp"
         conc-list: [2048]
@@ -7622,6 +7652,21 @@ dsr1-fp4-b200-dynamo-sglang-mtp:
           tp: 8
           ep: 8
           dp-attn: true
+      # 4p1d high-throughput (dedicated c2048 tune).
+      - spec-decoding: "mtp"
+        conc-list: [2048]
+        prefill:
+          num-worker: 4
+          tp: 4
+          ep: 4
+          dp-attn: true
+          additional-settings:
+          - "CONFIG_FILE=recipes/sglang/dsr1/b200-fp4/8k1k/disagg/mtp/8k1k_mtp_4p1d_c2048.yaml"
+        decode:
+          num-worker: 1
+          tp: 8
+          ep: 8
+          dp-attn: true
 
 kimik2.5-fp4-gb200-dynamo-trt:
   image: nvcr.io#nvidia/ai-dynamo/tensorrtllm-runtime:1.3.0-dev.1-cuda13
diff --git a/perf-changelog.yaml b/perf-changelog.yaml
index 336edc9a61..bb7440ca4d 100644
--- a/perf-changelog.yaml
+++ b/perf-changelog.yaml
@@ -4589,3 +4589,11 @@
     - "Image: lmsysorg/sglang:v0.5.12"
     - "14 topologies across 1k/1k and 8k/1k: prefill TP8 STP + decode wide-EP (DEP16/DEP32 high-throughput) and per-node TP8 low-latency, recipes under benchmarks/multi_node/srt-slurm-recipes/sglang/glm5/gb200-fp8/"
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1895
+
+- config-keys:
+    - dsr1-fp4-b200-dynamo-sglang-mtp
+  description:
+    - "Prune the 8k/1k MTP disagg low-latency points: 1p5d-tp8 to conc 4-32, 1p3d-tp8 to conc 32-64, and 1p1d-tp8 to conc 32."
+    - "Add 8k/1k MTP disagg high-throughput points: 1p1d and 4p1d (DEP4 prefill / DEP8 decode) at conc 512."
+    - "Add an 8k/1k MTP disagg 4p1d (DEP4 prefill / DEP8 decode) high-throughput point at conc 2048 with a dedicated engine tune (max-running-requests 2048, decode scheduler-recv-interval 1, prefill/decode stream-interval 100/34, SGLANG_HACK_SEQ_BOOTSTRAP_ROOM)."
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2113
diff --git a/runners/launch_b200-dgxc.sh b/runners/launch_b200-dgxc.sh
index e08441ebcc..8c7aa76bb2 100644
--- a/runners/launch_b200-dgxc.sh
+++ b/runners/launch_b200-dgxc.sh
@@ -120,7 +120,12 @@ if [[ "$IS_MULTINODE" == "true" ]]; then
     elif [[ $FRAMEWORK == "dynamo-sglang" && $MODEL_PREFIX == "dsr1" && $PRECISION == "fp4" ]]; then
         git clone https://github.com/NVIDIA/srt-slurm.git "$SRT_REPO_DIR"
         cd "$SRT_REPO_DIR" || exit 1
-        git checkout main
+        # Pin srt-slurm: newer commits stopped honoring the hash-pinned dynamo
+        # build and fall back to a dynamo release that is incompatible with this
+        # sglang image (worker fails at import). This is the last commit before
+        # that change. Do not float on main -- the srtctl + dynamo-install
+        # toolchain is unpinned there.
+        git checkout a98738de9b2233459b5456e9ed71af09ce893f92
         mkdir -p recipes/sglang/dsr1/b200-fp4
         cp -rT "$GITHUB_WORKSPACE/benchmarks/multi_node/srt-slurm-recipes/sglang/dsr1/b200-fp4" recipes/sglang/dsr1/b200-fp4
     else