diff --git a/benchmarks/multi_node/srt-slurm-recipes/sglang/glm5.1/gb300-fp8/1k1k/disagg/mtp/1k1k_mtp_hightpt_3.yaml b/benchmarks/multi_node/srt-slurm-recipes/sglang/glm5.1/gb300-fp8/1k1k/disagg/mtp/1k1k_mtp_hightpt_3.yaml
new file mode 100644
index 0000000000..ba9907840f
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/sglang/glm5.1/gb300-fp8/1k1k/disagg/mtp/1k1k_mtp_hightpt_3.yaml
@@ -0,0 +1,152 @@
+name: gb300-fp8-glm5-mtp_1k1k_hightpt_3
+
+model:
+  path: glm-5.1-fp8
+  container: "lmsysorg/sglang:v0.5.11-cu130"
+  precision: fp8
+
+resources:
+  gpu_type: gb300
+  gpus_per_node: 4
+  prefill_nodes: 6
+  prefill_workers: 6
+  decode_nodes: 12
+  decode_workers: 1
+frontend:
+  type: dynamo
+  enable_multiple_frontends: true
+  num_additional_frontends: 9
+dynamo:
+  version: 1.2.1
+
+backend:
+  prefill_environment:
+    TORCH_DISTRIBUTED_DEFAULT_TIMEOUT: '1800'
+    PYTHONUNBUFFERED: '1'
+    DYN_SKIP_SGLANG_LOG_FORMATTING: '1'
+    SGLANG_DISAGGREGATION_HEARTBEAT_MAX_FAILURE: '100000'
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: '100000'
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: '100000'
+    MC_TE_METRIC: 'true'
+    MC_FORCE_MNNVL: '1'
+    NCCL_MNNVL_ENABLE: '1'
+    NCCL_CUMEM_ENABLE: '1'
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: 'True'
+    SGLANG_USE_MESSAGE_QUEUE_BROADCASTER: '0'
+    SGLANG_DISABLE_TP_MEMORY_INBALANCE_CHECK: '1'
+    DYN_REQUEST_PLANE: nats
+
+  decode_environment:
+    TORCH_DISTRIBUTED_DEFAULT_TIMEOUT: '1800'
+    PYTHONUNBUFFERED: '1'
+    DYN_SKIP_SGLANG_LOG_FORMATTING: '1'
+    SGLANG_DISAGGREGATION_HEARTBEAT_MAX_FAILURE: '100000'
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: '100000'
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: '100000'
+    MC_TE_METRIC: 'true'
+    MC_FORCE_MNNVL: '1'
+    NCCL_MNNVL_ENABLE: '1'
+    NCCL_CUMEM_ENABLE: '1'
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: 'True'
+    SGLANG_USE_MESSAGE_QUEUE_BROADCASTER: '0'
+    SGLANG_DISABLE_TP_MEMORY_INBALANCE_CHECK: '1'
+    DYN_REQUEST_PLANE: nats
+      # DeepEP per-rank dispatch buffer; must be >= ceil(cuda_graph_max_bs / dp_size).
+      # Default 128 overflows with large DP + batch (e.g. 4096/24 ~= 171 > 128). Limit 1024.
+    SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK: '1024'
+
+  sglang_config:
+    prefill:
+        # Model configuration
+      served-model-name: GLM-5.1-FP8
+      trust-remote-code: true
+      quantization: fp8
+      kv-cache-dtype: fp8_e4m3
+
+        # Disaggregation mode
+      disaggregation-mode: prefill
+      disaggregation-transfer-backend: nixl
+
+        # Size limits
+      max-running-requests: 256
+      cuda-graph-max-bs: 256
+      mem-fraction-static: 0.7
+      context-length: 9600
+      chunked-prefill-size: 32768
+      max-prefill-tokens: 8192
+
+        # Parallelism
+      tensor-parallel-size: 4
+      data-parallel-size: 4
+      expert-parallel-size: 1
+      enable-dp-attention: true
+      enable-dp-lm-head: true
+      load-balance-method: total_tokens
+
+        # Backend
+      nsa-decode-backend: trtllm
+      nsa-prefill-backend: trtllm
+      moe-runner-backend: flashinfer_trtllm
+
+        # Other flags
+      enable-flashinfer-allreduce-fusion: true
+      disable-radix-cache: true
+      weight-loader-prefetch-checkpoints: true
+      model-loader-extra-config: '{"enable_multithread_load": true}'
+
+    decode:
+        # Model configuration
+      served-model-name: GLM-5.1-FP8
+      trust-remote-code: true
+
+      quantization: fp8
+      kv-cache-dtype: fp8_e4m3
+
+        # Disaggregation mode
+      disaggregation-mode: decode
+      disaggregation-transfer-backend: nixl
+
+        # Memory and token limits
+      mem-fraction-static: 0.8
+      context-length: 9600
+
+        # Backend
+      nsa-decode-backend: trtllm
+      nsa-prefill-backend: trtllm
+        # moe-runner-backend: "cutedsl"
+
+        # Detokenizer
+      skip-tokenizer-init: true
+      stream-interval: 30
+
+        # Other flags
+      disable-radix-cache: true
+      weight-loader-prefetch-checkpoints: true
+      model-loader-extra-config: '{"enable_multithread_load": true}'
+      tensor-parallel-size: 48
+      expert-parallel-size: 48
+      data-parallel-size: 48
+      enable-dp-lm-head: true
+      enable-dp-attention: true
+      moe-dense-tp-size: 1
+      ep-num-redundant-experts: 32
+      ep-dispatch-algorithm: static
+      moe-a2a-backend: deepep
+      deepep-mode: low_latency
+      deepep-config: /configs/deepep_config.json
+      max-running-requests: 6144
+      cuda-graph-max-bs: 128
+      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: 1024
+  osl: 1024
+  concurrencies: '6500'
diff --git a/benchmarks/multi_node/srt-slurm-recipes/sglang/glm5.1/gb300-fp8/1k1k/disagg/mtp/1k1k_mtp_hightpt_4.yaml b/benchmarks/multi_node/srt-slurm-recipes/sglang/glm5.1/gb300-fp8/1k1k/disagg/mtp/1k1k_mtp_hightpt_4.yaml
new file mode 100644
index 0000000000..ee2b0a2026
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/sglang/glm5.1/gb300-fp8/1k1k/disagg/mtp/1k1k_mtp_hightpt_4.yaml
@@ -0,0 +1,152 @@
+name: gb300-fp8-glm5-mtp_1k1k_hightpt_4
+
+model:
+  path: glm-5.1-fp8
+  container: "lmsysorg/sglang:v0.5.11-cu130"
+  precision: fp8
+
+resources:
+  gpu_type: gb300
+  gpus_per_node: 4
+  prefill_nodes: 4
+  prefill_workers: 4
+  decode_nodes: 14
+  decode_workers: 1
+frontend:
+  type: dynamo
+  enable_multiple_frontends: true
+  num_additional_frontends: 9
+dynamo:
+  version: 1.2.1
+
+backend:
+  prefill_environment:
+    TORCH_DISTRIBUTED_DEFAULT_TIMEOUT: '1800'
+    PYTHONUNBUFFERED: '1'
+    DYN_SKIP_SGLANG_LOG_FORMATTING: '1'
+    SGLANG_DISAGGREGATION_HEARTBEAT_MAX_FAILURE: '100000'
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: '100000'
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: '100000'
+    MC_TE_METRIC: 'true'
+    MC_FORCE_MNNVL: '1'
+    NCCL_MNNVL_ENABLE: '1'
+    NCCL_CUMEM_ENABLE: '1'
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: 'True'
+    SGLANG_USE_MESSAGE_QUEUE_BROADCASTER: '0'
+    SGLANG_DISABLE_TP_MEMORY_INBALANCE_CHECK: '1'
+    DYN_REQUEST_PLANE: nats
+
+  decode_environment:
+    TORCH_DISTRIBUTED_DEFAULT_TIMEOUT: '1800'
+    PYTHONUNBUFFERED: '1'
+    DYN_SKIP_SGLANG_LOG_FORMATTING: '1'
+    SGLANG_DISAGGREGATION_HEARTBEAT_MAX_FAILURE: '100000'
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: '100000'
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: '100000'
+    MC_TE_METRIC: 'true'
+    MC_FORCE_MNNVL: '1'
+    NCCL_MNNVL_ENABLE: '1'
+    NCCL_CUMEM_ENABLE: '1'
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: 'True'
+    SGLANG_USE_MESSAGE_QUEUE_BROADCASTER: '0'
+    SGLANG_DISABLE_TP_MEMORY_INBALANCE_CHECK: '1'
+    DYN_REQUEST_PLANE: nats
+      # DeepEP per-rank dispatch buffer; must be >= ceil(cuda_graph_max_bs / dp_size).
+      # Default 128 overflows with large DP + batch (e.g. 4096/24 ~= 171 > 128). Limit 1024.
+    SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK: '1024'
+
+  sglang_config:
+    prefill:
+        # Model configuration
+      served-model-name: GLM-5.1-FP8
+      trust-remote-code: true
+      quantization: fp8
+      kv-cache-dtype: fp8_e4m3
+
+        # Disaggregation mode
+      disaggregation-mode: prefill
+      disaggregation-transfer-backend: nixl
+
+        # Size limits
+      max-running-requests: 256
+      cuda-graph-max-bs: 256
+      mem-fraction-static: 0.7
+      context-length: 9600
+      chunked-prefill-size: 32768
+      max-prefill-tokens: 8192
+
+        # Parallelism
+      tensor-parallel-size: 4
+      data-parallel-size: 4
+      expert-parallel-size: 1
+      enable-dp-attention: true
+      enable-dp-lm-head: true
+      load-balance-method: total_tokens
+
+        # Backend
+      nsa-decode-backend: trtllm
+      nsa-prefill-backend: trtllm
+      moe-runner-backend: flashinfer_trtllm
+
+        # Other flags
+      enable-flashinfer-allreduce-fusion: true
+      disable-radix-cache: true
+      weight-loader-prefetch-checkpoints: true
+      model-loader-extra-config: '{"enable_multithread_load": true}'
+
+    decode:
+        # Model configuration
+      served-model-name: GLM-5.1-FP8
+      trust-remote-code: true
+
+      quantization: fp8
+      kv-cache-dtype: fp8_e4m3
+
+        # Disaggregation mode
+      disaggregation-mode: decode
+      disaggregation-transfer-backend: nixl
+
+        # Memory and token limits
+      mem-fraction-static: 0.8
+      context-length: 9600
+
+        # Backend
+      nsa-decode-backend: trtllm
+      nsa-prefill-backend: trtllm
+        # moe-runner-backend: "cutedsl"
+
+        # Detokenizer
+      skip-tokenizer-init: true
+      stream-interval: 30
+
+        # Other flags
+      disable-radix-cache: true
+      weight-loader-prefetch-checkpoints: true
+      model-loader-extra-config: '{"enable_multithread_load": true}'
+      tensor-parallel-size: 56
+      expert-parallel-size: 56
+      data-parallel-size: 56
+      enable-dp-lm-head: true
+      enable-dp-attention: true
+      moe-dense-tp-size: 1
+      ep-num-redundant-experts: 24
+      ep-dispatch-algorithm: static
+      moe-a2a-backend: deepep
+      deepep-mode: low_latency
+      deepep-config: /configs/deepep_config.json
+      max-running-requests: 5600
+      cuda-graph-max-bs: 100
+      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: 1024
+  osl: 1024
+  concurrencies: '5700'
diff --git a/benchmarks/multi_node/srt-slurm-recipes/sglang/glm5.1/gb300-fp8/1k1k/disagg/mtp/1k1k_mtp_lowlat_0.yaml b/benchmarks/multi_node/srt-slurm-recipes/sglang/glm5.1/gb300-fp8/1k1k/disagg/mtp/1k1k_mtp_lowlat_0.yaml
new file mode 100644
index 0000000000..cd01ba6b73
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/sglang/glm5.1/gb300-fp8/1k1k/disagg/mtp/1k1k_mtp_lowlat_0.yaml
@@ -0,0 +1,144 @@
+name: gb300-fp8-glm5-mtp_1k1k_lowlat_0
+
+model:
+  path: glm-5.1-fp8
+  container: "lmsysorg/sglang:v0.5.11-cu130"
+  precision: fp8
+
+resources:
+  gpu_type: gb300
+  gpus_per_node: 4
+  prefill_nodes: 1
+  prefill_workers: 1
+  decode_nodes: 17
+  decode_workers: 17
+frontend:
+  type: dynamo
+dynamo:
+  version: 1.2.1
+
+backend:
+  prefill_environment:
+    TORCH_DISTRIBUTED_DEFAULT_TIMEOUT: '1800'
+    PYTHONUNBUFFERED: '1'
+    DYN_SKIP_SGLANG_LOG_FORMATTING: '1'
+    SGLANG_DISAGGREGATION_HEARTBEAT_MAX_FAILURE: '100000'
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: '100000'
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: '100000'
+    MC_TE_METRIC: 'true'
+    MC_FORCE_MNNVL: '1'
+    NCCL_MNNVL_ENABLE: '1'
+    NCCL_CUMEM_ENABLE: '1'
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: 'True'
+    SGLANG_USE_MESSAGE_QUEUE_BROADCASTER: '0'
+    SGLANG_DISABLE_TP_MEMORY_INBALANCE_CHECK: '1'
+    DYN_REQUEST_PLANE: nats
+
+  decode_environment:
+    TORCH_DISTRIBUTED_DEFAULT_TIMEOUT: '1800'
+    PYTHONUNBUFFERED: '1'
+    DYN_SKIP_SGLANG_LOG_FORMATTING: '1'
+    SGLANG_DISAGGREGATION_HEARTBEAT_MAX_FAILURE: '100000'
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: '100000'
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: '100000'
+    MC_TE_METRIC: 'true'
+    MC_FORCE_MNNVL: '1'
+    NCCL_MNNVL_ENABLE: '1'
+    NCCL_CUMEM_ENABLE: '1'
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: 'True'
+    SGLANG_USE_MESSAGE_QUEUE_BROADCASTER: '0'
+    SGLANG_DISABLE_TP_MEMORY_INBALANCE_CHECK: '1'
+    DYN_REQUEST_PLANE: nats
+      # DeepEP per-rank dispatch buffer; must be >= ceil(cuda_graph_max_bs / dp_size).
+      # Default 128 overflows with large DP + batch (e.g. 4096/24 ~= 171 > 128). Limit 1024.
+    SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK: '1024'
+
+  sglang_config:
+    prefill:
+        # Model configuration
+      served-model-name: GLM-5.1-FP8
+      trust-remote-code: true
+      quantization: fp8
+      kv-cache-dtype: fp8_e4m3
+
+        # Disaggregation mode
+      disaggregation-mode: prefill
+      disaggregation-transfer-backend: nixl
+
+        # Size limits
+      max-running-requests: 256
+      cuda-graph-max-bs: 256
+      mem-fraction-static: 0.7
+      context-length: 9600
+      chunked-prefill-size: 32768
+      max-prefill-tokens: 8192
+
+        # Parallelism
+      tensor-parallel-size: 4
+      data-parallel-size: 4
+      expert-parallel-size: 1
+      enable-dp-attention: true
+      enable-dp-lm-head: true
+      load-balance-method: total_tokens
+
+        # Backend
+      nsa-decode-backend: trtllm
+      nsa-prefill-backend: trtllm
+      moe-runner-backend: flashinfer_trtllm
+
+        # Other flags
+      enable-flashinfer-allreduce-fusion: true
+      disable-radix-cache: true
+      weight-loader-prefetch-checkpoints: true
+      model-loader-extra-config: '{"enable_multithread_load": true}'
+
+    decode:
+        # Model configuration
+      served-model-name: GLM-5.1-FP8
+      trust-remote-code: true
+
+      quantization: fp8
+      kv-cache-dtype: fp8_e4m3
+
+        # Disaggregation mode
+      disaggregation-mode: decode
+      disaggregation-transfer-backend: nixl
+
+        # Memory and token limits
+      mem-fraction-static: 0.8
+      context-length: 9600
+
+        # Backend
+      nsa-decode-backend: trtllm
+      nsa-prefill-backend: trtllm
+        # moe-runner-backend: "cutedsl"
+
+        # Detokenizer
+      skip-tokenizer-init: true
+      stream-interval: 30
+
+        # Other flags
+      disable-radix-cache: true
+      weight-loader-prefetch-checkpoints: true
+      model-loader-extra-config: '{"enable_multithread_load": true}'
+      tensor-parallel-size: 4
+      expert-parallel-size: 1
+      data-parallel-size: 1
+      enable-flashinfer-allreduce-fusion: true
+      moe-runner-backend: flashinfer_trtllm
+      max-running-requests: 32
+      cuda-graph-max-bs: 32
+      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: 1024
+  osl: 1024
+  concurrencies: 512x256x128x64
diff --git a/benchmarks/multi_node/srt-slurm-recipes/sglang/glm5.1/gb300-fp8/1k1k/disagg/mtp/1k1k_mtp_lowlat_1.yaml b/benchmarks/multi_node/srt-slurm-recipes/sglang/glm5.1/gb300-fp8/1k1k/disagg/mtp/1k1k_mtp_lowlat_1.yaml
new file mode 100644
index 0000000000..1ec2d053a3
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/sglang/glm5.1/gb300-fp8/1k1k/disagg/mtp/1k1k_mtp_lowlat_1.yaml
@@ -0,0 +1,144 @@
+name: gb300-fp8-glm5-mtp_1k1k_lowlat_1
+
+model:
+  path: glm-5.1-fp8
+  container: "lmsysorg/sglang:v0.5.11-cu130"
+  precision: fp8
+
+resources:
+  gpu_type: gb300
+  gpus_per_node: 4
+  prefill_nodes: 1
+  prefill_workers: 1
+  decode_nodes: 17
+  decode_workers: 17
+frontend:
+  type: dynamo
+dynamo:
+  version: 1.2.1
+
+backend:
+  prefill_environment:
+    TORCH_DISTRIBUTED_DEFAULT_TIMEOUT: '1800'
+    PYTHONUNBUFFERED: '1'
+    DYN_SKIP_SGLANG_LOG_FORMATTING: '1'
+    SGLANG_DISAGGREGATION_HEARTBEAT_MAX_FAILURE: '100000'
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: '100000'
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: '100000'
+    MC_TE_METRIC: 'true'
+    MC_FORCE_MNNVL: '1'
+    NCCL_MNNVL_ENABLE: '1'
+    NCCL_CUMEM_ENABLE: '1'
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: 'True'
+    SGLANG_USE_MESSAGE_QUEUE_BROADCASTER: '0'
+    SGLANG_DISABLE_TP_MEMORY_INBALANCE_CHECK: '1'
+    DYN_REQUEST_PLANE: nats
+
+  decode_environment:
+    TORCH_DISTRIBUTED_DEFAULT_TIMEOUT: '1800'
+    PYTHONUNBUFFERED: '1'
+    DYN_SKIP_SGLANG_LOG_FORMATTING: '1'
+    SGLANG_DISAGGREGATION_HEARTBEAT_MAX_FAILURE: '100000'
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: '100000'
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: '100000'
+    MC_TE_METRIC: 'true'
+    MC_FORCE_MNNVL: '1'
+    NCCL_MNNVL_ENABLE: '1'
+    NCCL_CUMEM_ENABLE: '1'
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: 'True'
+    SGLANG_USE_MESSAGE_QUEUE_BROADCASTER: '0'
+    SGLANG_DISABLE_TP_MEMORY_INBALANCE_CHECK: '1'
+    DYN_REQUEST_PLANE: nats
+      # DeepEP per-rank dispatch buffer; must be >= ceil(cuda_graph_max_bs / dp_size).
+      # Default 128 overflows with large DP + batch (e.g. 4096/24 ~= 171 > 128). Limit 1024.
+    SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK: '1024'
+
+  sglang_config:
+    prefill:
+        # Model configuration
+      served-model-name: GLM-5.1-FP8
+      trust-remote-code: true
+      quantization: fp8
+      kv-cache-dtype: fp8_e4m3
+
+        # Disaggregation mode
+      disaggregation-mode: prefill
+      disaggregation-transfer-backend: nixl
+
+        # Size limits
+      max-running-requests: 256
+      cuda-graph-max-bs: 256
+      mem-fraction-static: 0.7
+      context-length: 9600
+      chunked-prefill-size: 32768
+      max-prefill-tokens: 8192
+
+        # Parallelism
+      tensor-parallel-size: 4
+      data-parallel-size: 4
+      expert-parallel-size: 1
+      enable-dp-attention: true
+      enable-dp-lm-head: true
+      load-balance-method: total_tokens
+
+        # Backend
+      nsa-decode-backend: trtllm
+      nsa-prefill-backend: trtllm
+      moe-runner-backend: flashinfer_trtllm
+
+        # Other flags
+      enable-flashinfer-allreduce-fusion: true
+      disable-radix-cache: true
+      weight-loader-prefetch-checkpoints: true
+      model-loader-extra-config: '{"enable_multithread_load": true}'
+
+    decode:
+        # Model configuration
+      served-model-name: GLM-5.1-FP8
+      trust-remote-code: true
+
+      quantization: fp8
+      kv-cache-dtype: fp8_e4m3
+
+        # Disaggregation mode
+      disaggregation-mode: decode
+      disaggregation-transfer-backend: nixl
+
+        # Memory and token limits
+      mem-fraction-static: 0.8
+      context-length: 9600
+
+        # Backend
+      nsa-decode-backend: trtllm
+      nsa-prefill-backend: trtllm
+        # moe-runner-backend: "cutedsl"
+
+        # Detokenizer
+      skip-tokenizer-init: true
+      stream-interval: 30
+
+        # Other flags
+      disable-radix-cache: true
+      weight-loader-prefetch-checkpoints: true
+      model-loader-extra-config: '{"enable_multithread_load": true}'
+      tensor-parallel-size: 4
+      expert-parallel-size: 1
+      data-parallel-size: 1
+      enable-flashinfer-allreduce-fusion: true
+      moe-runner-backend: flashinfer_trtllm
+      max-running-requests: 1
+      cuda-graph-max-bs: 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
+  req_rate: inf
+  isl: 1024
+  osl: 1024
+  concurrencies: '32'
diff --git a/benchmarks/multi_node/srt-slurm-recipes/sglang/glm5.1/gb300-fp8/8k1k/disagg/mtp/8k1k_mtp_hightpt_0.yaml b/benchmarks/multi_node/srt-slurm-recipes/sglang/glm5.1/gb300-fp8/8k1k/disagg/mtp/8k1k_mtp_hightpt_0.yaml
new file mode 100644
index 0000000000..6fbd1df10b
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/sglang/glm5.1/gb300-fp8/8k1k/disagg/mtp/8k1k_mtp_hightpt_0.yaml
@@ -0,0 +1,152 @@
+name: gb300-fp8-glm5-mtp_8k1k_hightpt_0
+
+model:
+  path: glm-5.1-fp8
+  container: "lmsysorg/sglang:v0.5.11-cu130"
+  precision: fp8
+
+resources:
+  gpu_type: gb300
+  gpus_per_node: 4
+  prefill_nodes: 14
+  prefill_workers: 14
+  decode_nodes: 4
+  decode_workers: 1
+frontend:
+  type: dynamo
+  enable_multiple_frontends: true
+  num_additional_frontends: 9
+dynamo:
+  version: 1.2.1
+
+backend:
+  prefill_environment:
+    TORCH_DISTRIBUTED_DEFAULT_TIMEOUT: '1800'
+    PYTHONUNBUFFERED: '1'
+    DYN_SKIP_SGLANG_LOG_FORMATTING: '1'
+    SGLANG_DISAGGREGATION_HEARTBEAT_MAX_FAILURE: '100000'
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: '100000'
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: '100000'
+    MC_TE_METRIC: 'true'
+    MC_FORCE_MNNVL: '1'
+    NCCL_MNNVL_ENABLE: '1'
+    NCCL_CUMEM_ENABLE: '1'
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: 'True'
+    SGLANG_USE_MESSAGE_QUEUE_BROADCASTER: '0'
+    SGLANG_DISABLE_TP_MEMORY_INBALANCE_CHECK: '1'
+    DYN_REQUEST_PLANE: nats
+
+  decode_environment:
+    TORCH_DISTRIBUTED_DEFAULT_TIMEOUT: '1800'
+    PYTHONUNBUFFERED: '1'
+    DYN_SKIP_SGLANG_LOG_FORMATTING: '1'
+    SGLANG_DISAGGREGATION_HEARTBEAT_MAX_FAILURE: '100000'
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: '100000'
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: '100000'
+    MC_TE_METRIC: 'true'
+    MC_FORCE_MNNVL: '1'
+    NCCL_MNNVL_ENABLE: '1'
+    NCCL_CUMEM_ENABLE: '1'
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: 'True'
+    SGLANG_USE_MESSAGE_QUEUE_BROADCASTER: '0'
+    SGLANG_DISABLE_TP_MEMORY_INBALANCE_CHECK: '1'
+    DYN_REQUEST_PLANE: nats
+      # DeepEP per-rank dispatch buffer; must be >= ceil(cuda_graph_max_bs / dp_size).
+      # Default 128 overflows with large DP + batch (e.g. 4096/24 ~= 171 > 128). Limit 1024.
+    SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK: '1024'
+
+  sglang_config:
+    prefill:
+        # Model configuration
+      served-model-name: GLM-5.1-FP8
+      trust-remote-code: true
+      quantization: fp8
+      kv-cache-dtype: fp8_e4m3
+
+        # Disaggregation mode
+      disaggregation-mode: prefill
+      disaggregation-transfer-backend: nixl
+
+        # Size limits
+      max-running-requests: 256
+      cuda-graph-max-bs: 256
+      mem-fraction-static: 0.7
+      context-length: 9600
+      chunked-prefill-size: 32768
+      max-prefill-tokens: 8192
+
+        # Parallelism
+      tensor-parallel-size: 4
+      data-parallel-size: 4
+      expert-parallel-size: 1
+      enable-dp-attention: true
+      enable-dp-lm-head: true
+      load-balance-method: total_tokens
+
+        # Backend
+      nsa-decode-backend: trtllm
+      nsa-prefill-backend: trtllm
+      moe-runner-backend: flashinfer_trtllm
+
+        # Other flags
+      enable-flashinfer-allreduce-fusion: true
+      disable-radix-cache: true
+      weight-loader-prefetch-checkpoints: true
+      model-loader-extra-config: '{"enable_multithread_load": true}'
+
+    decode:
+        # Model configuration
+      served-model-name: GLM-5.1-FP8
+      trust-remote-code: true
+
+      quantization: fp8
+      kv-cache-dtype: fp8_e4m3
+
+        # Disaggregation mode
+      disaggregation-mode: decode
+      disaggregation-transfer-backend: nixl
+
+        # Memory and token limits
+      mem-fraction-static: 0.8
+      context-length: 9600
+
+        # Backend
+      nsa-decode-backend: trtllm
+      nsa-prefill-backend: trtllm
+        # moe-runner-backend: "cutedsl"
+
+        # Detokenizer
+      skip-tokenizer-init: true
+      stream-interval: 30
+
+        # Other flags
+      disable-radix-cache: true
+      weight-loader-prefetch-checkpoints: true
+      model-loader-extra-config: '{"enable_multithread_load": true}'
+      tensor-parallel-size: 16
+      expert-parallel-size: 16
+      data-parallel-size: 16
+      enable-dp-lm-head: true
+      enable-dp-attention: true
+      moe-dense-tp-size: 1
+      ep-num-redundant-experts: 32
+      ep-dispatch-algorithm: static
+      moe-a2a-backend: deepep
+      deepep-mode: low_latency
+      deepep-config: /configs/deepep_config.json
+      max-running-requests: 2800
+      cuda-graph-max-bs: 175
+      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: '2800'
diff --git a/benchmarks/multi_node/srt-slurm-recipes/sglang/glm5.1/gb300-fp8/8k1k/disagg/mtp/8k1k_mtp_hightpt_1.yaml b/benchmarks/multi_node/srt-slurm-recipes/sglang/glm5.1/gb300-fp8/8k1k/disagg/mtp/8k1k_mtp_hightpt_1.yaml
new file mode 100644
index 0000000000..3c1a641a82
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/sglang/glm5.1/gb300-fp8/8k1k/disagg/mtp/8k1k_mtp_hightpt_1.yaml
@@ -0,0 +1,152 @@
+name: gb300-fp8-glm5-mtp_8k1k_hightpt_1
+
+model:
+  path: glm-5.1-fp8
+  container: "lmsysorg/sglang:v0.5.11-cu130"
+  precision: fp8
+
+resources:
+  gpu_type: gb300
+  gpus_per_node: 4
+  prefill_nodes: 12
+  prefill_workers: 12
+  decode_nodes: 6
+  decode_workers: 1
+frontend:
+  type: dynamo
+  enable_multiple_frontends: true
+  num_additional_frontends: 9
+dynamo:
+  version: 1.2.1
+
+backend:
+  prefill_environment:
+    TORCH_DISTRIBUTED_DEFAULT_TIMEOUT: '1800'
+    PYTHONUNBUFFERED: '1'
+    DYN_SKIP_SGLANG_LOG_FORMATTING: '1'
+    SGLANG_DISAGGREGATION_HEARTBEAT_MAX_FAILURE: '100000'
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: '100000'
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: '100000'
+    MC_TE_METRIC: 'true'
+    MC_FORCE_MNNVL: '1'
+    NCCL_MNNVL_ENABLE: '1'
+    NCCL_CUMEM_ENABLE: '1'
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: 'True'
+    SGLANG_USE_MESSAGE_QUEUE_BROADCASTER: '0'
+    SGLANG_DISABLE_TP_MEMORY_INBALANCE_CHECK: '1'
+    DYN_REQUEST_PLANE: nats
+
+  decode_environment:
+    TORCH_DISTRIBUTED_DEFAULT_TIMEOUT: '1800'
+    PYTHONUNBUFFERED: '1'
+    DYN_SKIP_SGLANG_LOG_FORMATTING: '1'
+    SGLANG_DISAGGREGATION_HEARTBEAT_MAX_FAILURE: '100000'
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: '100000'
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: '100000'
+    MC_TE_METRIC: 'true'
+    MC_FORCE_MNNVL: '1'
+    NCCL_MNNVL_ENABLE: '1'
+    NCCL_CUMEM_ENABLE: '1'
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: 'True'
+    SGLANG_USE_MESSAGE_QUEUE_BROADCASTER: '0'
+    SGLANG_DISABLE_TP_MEMORY_INBALANCE_CHECK: '1'
+    DYN_REQUEST_PLANE: nats
+      # DeepEP per-rank dispatch buffer; must be >= ceil(cuda_graph_max_bs / dp_size).
+      # Default 128 overflows with large DP + batch (e.g. 4096/24 ~= 171 > 128). Limit 1024.
+    SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK: '1024'
+
+  sglang_config:
+    prefill:
+        # Model configuration
+      served-model-name: GLM-5.1-FP8
+      trust-remote-code: true
+      quantization: fp8
+      kv-cache-dtype: fp8_e4m3
+
+        # Disaggregation mode
+      disaggregation-mode: prefill
+      disaggregation-transfer-backend: nixl
+
+        # Size limits
+      max-running-requests: 256
+      cuda-graph-max-bs: 256
+      mem-fraction-static: 0.7
+      context-length: 9600
+      chunked-prefill-size: 32768
+      max-prefill-tokens: 8192
+
+        # Parallelism
+      tensor-parallel-size: 4
+      data-parallel-size: 4
+      expert-parallel-size: 1
+      enable-dp-attention: true
+      enable-dp-lm-head: true
+      load-balance-method: total_tokens
+
+        # Backend
+      nsa-decode-backend: trtllm
+      nsa-prefill-backend: trtllm
+      moe-runner-backend: flashinfer_trtllm
+
+        # Other flags
+      enable-flashinfer-allreduce-fusion: true
+      disable-radix-cache: true
+      weight-loader-prefetch-checkpoints: true
+      model-loader-extra-config: '{"enable_multithread_load": true}'
+
+    decode:
+        # Model configuration
+      served-model-name: GLM-5.1-FP8
+      trust-remote-code: true
+
+      quantization: fp8
+      kv-cache-dtype: fp8_e4m3
+
+        # Disaggregation mode
+      disaggregation-mode: decode
+      disaggregation-transfer-backend: nixl
+
+        # Memory and token limits
+      mem-fraction-static: 0.8
+      context-length: 9600
+
+        # Backend
+      nsa-decode-backend: trtllm
+      nsa-prefill-backend: trtllm
+        # moe-runner-backend: "cutedsl"
+
+        # Detokenizer
+      skip-tokenizer-init: true
+      stream-interval: 30
+
+        # Other flags
+      disable-radix-cache: true
+      weight-loader-prefetch-checkpoints: true
+      model-loader-extra-config: '{"enable_multithread_load": true}'
+      tensor-parallel-size: 24
+      expert-parallel-size: 24
+      data-parallel-size: 24
+      enable-dp-lm-head: true
+      enable-dp-attention: true
+      moe-dense-tp-size: 1
+      ep-num-redundant-experts: 32
+      ep-dispatch-algorithm: static
+      moe-a2a-backend: deepep
+      deepep-mode: low_latency
+      deepep-config: /configs/deepep_config.json
+      max-running-requests: 1680
+      cuda-graph-max-bs: 70
+      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: '1700'
diff --git a/benchmarks/multi_node/srt-slurm-recipes/sglang/glm5.1/gb300-fp8/8k1k/disagg/mtp/8k1k_mtp_hightpt_2.yaml b/benchmarks/multi_node/srt-slurm-recipes/sglang/glm5.1/gb300-fp8/8k1k/disagg/mtp/8k1k_mtp_hightpt_2.yaml
new file mode 100644
index 0000000000..1ee087dfa2
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/sglang/glm5.1/gb300-fp8/8k1k/disagg/mtp/8k1k_mtp_hightpt_2.yaml
@@ -0,0 +1,152 @@
+name: gb300-fp8-glm5-mtp_8k1k_hightpt_2
+
+model:
+  path: glm-5.1-fp8
+  container: "lmsysorg/sglang:v0.5.11-cu130"
+  precision: fp8
+
+resources:
+  gpu_type: gb300
+  gpus_per_node: 4
+  prefill_nodes: 10
+  prefill_workers: 10
+  decode_nodes: 8
+  decode_workers: 1
+frontend:
+  type: dynamo
+  enable_multiple_frontends: true
+  num_additional_frontends: 9
+dynamo:
+  version: 1.2.1
+
+backend:
+  prefill_environment:
+    TORCH_DISTRIBUTED_DEFAULT_TIMEOUT: '1800'
+    PYTHONUNBUFFERED: '1'
+    DYN_SKIP_SGLANG_LOG_FORMATTING: '1'
+    SGLANG_DISAGGREGATION_HEARTBEAT_MAX_FAILURE: '100000'
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: '100000'
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: '100000'
+    MC_TE_METRIC: 'true'
+    MC_FORCE_MNNVL: '1'
+    NCCL_MNNVL_ENABLE: '1'
+    NCCL_CUMEM_ENABLE: '1'
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: 'True'
+    SGLANG_USE_MESSAGE_QUEUE_BROADCASTER: '0'
+    SGLANG_DISABLE_TP_MEMORY_INBALANCE_CHECK: '1'
+    DYN_REQUEST_PLANE: nats
+
+  decode_environment:
+    TORCH_DISTRIBUTED_DEFAULT_TIMEOUT: '1800'
+    PYTHONUNBUFFERED: '1'
+    DYN_SKIP_SGLANG_LOG_FORMATTING: '1'
+    SGLANG_DISAGGREGATION_HEARTBEAT_MAX_FAILURE: '100000'
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: '100000'
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: '100000'
+    MC_TE_METRIC: 'true'
+    MC_FORCE_MNNVL: '1'
+    NCCL_MNNVL_ENABLE: '1'
+    NCCL_CUMEM_ENABLE: '1'
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: 'True'
+    SGLANG_USE_MESSAGE_QUEUE_BROADCASTER: '0'
+    SGLANG_DISABLE_TP_MEMORY_INBALANCE_CHECK: '1'
+    DYN_REQUEST_PLANE: nats
+      # DeepEP per-rank dispatch buffer; must be >= ceil(cuda_graph_max_bs / dp_size).
+      # Default 128 overflows with large DP + batch (e.g. 4096/24 ~= 171 > 128). Limit 1024.
+    SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK: '1024'
+
+  sglang_config:
+    prefill:
+        # Model configuration
+      served-model-name: GLM-5.1-FP8
+      trust-remote-code: true
+      quantization: fp8
+      kv-cache-dtype: fp8_e4m3
+
+        # Disaggregation mode
+      disaggregation-mode: prefill
+      disaggregation-transfer-backend: nixl
+
+        # Size limits
+      max-running-requests: 256
+      cuda-graph-max-bs: 256
+      mem-fraction-static: 0.7
+      context-length: 9600
+      chunked-prefill-size: 32768
+      max-prefill-tokens: 8192
+
+        # Parallelism
+      tensor-parallel-size: 4
+      data-parallel-size: 4
+      expert-parallel-size: 1
+      enable-dp-attention: true
+      enable-dp-lm-head: true
+      load-balance-method: total_tokens
+
+        # Backend
+      nsa-decode-backend: trtllm
+      nsa-prefill-backend: trtllm
+      moe-runner-backend: flashinfer_trtllm
+
+        # Other flags
+      enable-flashinfer-allreduce-fusion: true
+      disable-radix-cache: true
+      weight-loader-prefetch-checkpoints: true
+      model-loader-extra-config: '{"enable_multithread_load": true}'
+
+    decode:
+        # Model configuration
+      served-model-name: GLM-5.1-FP8
+      trust-remote-code: true
+
+      quantization: fp8
+      kv-cache-dtype: fp8_e4m3
+
+        # Disaggregation mode
+      disaggregation-mode: decode
+      disaggregation-transfer-backend: nixl
+
+        # Memory and token limits
+      mem-fraction-static: 0.8
+      context-length: 9600
+
+        # Backend
+      nsa-decode-backend: trtllm
+      nsa-prefill-backend: trtllm
+        # moe-runner-backend: "cutedsl"
+
+        # Detokenizer
+      skip-tokenizer-init: true
+      stream-interval: 30
+
+        # Other flags
+      disable-radix-cache: true
+      weight-loader-prefetch-checkpoints: true
+      model-loader-extra-config: '{"enable_multithread_load": true}'
+      tensor-parallel-size: 32
+      expert-parallel-size: 32
+      data-parallel-size: 32
+      enable-dp-lm-head: true
+      enable-dp-attention: true
+      moe-dense-tp-size: 1
+      ep-num-redundant-experts: 32
+      ep-dispatch-algorithm: static
+      moe-a2a-backend: deepep
+      deepep-mode: low_latency
+      deepep-config: /configs/deepep_config.json
+      max-running-requests: 1280
+      cuda-graph-max-bs: 40
+      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: '1300'
diff --git a/benchmarks/multi_node/srt-slurm-recipes/sglang/glm5.1/gb300-fp8/8k1k/disagg/mtp/8k1k_mtp_hightpt_3.yaml b/benchmarks/multi_node/srt-slurm-recipes/sglang/glm5.1/gb300-fp8/8k1k/disagg/mtp/8k1k_mtp_hightpt_3.yaml
new file mode 100644
index 0000000000..a03cdef22d
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/sglang/glm5.1/gb300-fp8/8k1k/disagg/mtp/8k1k_mtp_hightpt_3.yaml
@@ -0,0 +1,152 @@
+name: gb300-fp8-glm5-mtp_8k1k_hightpt_3
+
+model:
+  path: glm-5.1-fp8
+  container: "lmsysorg/sglang:v0.5.11-cu130"
+  precision: fp8
+
+resources:
+  gpu_type: gb300
+  gpus_per_node: 4
+  prefill_nodes: 8
+  prefill_workers: 8
+  decode_nodes: 10
+  decode_workers: 1
+frontend:
+  type: dynamo
+  enable_multiple_frontends: true
+  num_additional_frontends: 9
+dynamo:
+  version: 1.2.1
+
+backend:
+  prefill_environment:
+    TORCH_DISTRIBUTED_DEFAULT_TIMEOUT: '1800'
+    PYTHONUNBUFFERED: '1'
+    DYN_SKIP_SGLANG_LOG_FORMATTING: '1'
+    SGLANG_DISAGGREGATION_HEARTBEAT_MAX_FAILURE: '100000'
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: '100000'
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: '100000'
+    MC_TE_METRIC: 'true'
+    MC_FORCE_MNNVL: '1'
+    NCCL_MNNVL_ENABLE: '1'
+    NCCL_CUMEM_ENABLE: '1'
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: 'True'
+    SGLANG_USE_MESSAGE_QUEUE_BROADCASTER: '0'
+    SGLANG_DISABLE_TP_MEMORY_INBALANCE_CHECK: '1'
+    DYN_REQUEST_PLANE: nats
+
+  decode_environment:
+    TORCH_DISTRIBUTED_DEFAULT_TIMEOUT: '1800'
+    PYTHONUNBUFFERED: '1'
+    DYN_SKIP_SGLANG_LOG_FORMATTING: '1'
+    SGLANG_DISAGGREGATION_HEARTBEAT_MAX_FAILURE: '100000'
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: '100000'
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: '100000'
+    MC_TE_METRIC: 'true'
+    MC_FORCE_MNNVL: '1'
+    NCCL_MNNVL_ENABLE: '1'
+    NCCL_CUMEM_ENABLE: '1'
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: 'True'
+    SGLANG_USE_MESSAGE_QUEUE_BROADCASTER: '0'
+    SGLANG_DISABLE_TP_MEMORY_INBALANCE_CHECK: '1'
+    DYN_REQUEST_PLANE: nats
+      # DeepEP per-rank dispatch buffer; must be >= ceil(cuda_graph_max_bs / dp_size).
+      # Default 128 overflows with large DP + batch (e.g. 4096/24 ~= 171 > 128). Limit 1024.
+    SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK: '1024'
+
+  sglang_config:
+    prefill:
+        # Model configuration
+      served-model-name: GLM-5.1-FP8
+      trust-remote-code: true
+      quantization: fp8
+      kv-cache-dtype: fp8_e4m3
+
+        # Disaggregation mode
+      disaggregation-mode: prefill
+      disaggregation-transfer-backend: nixl
+
+        # Size limits
+      max-running-requests: 256
+      cuda-graph-max-bs: 256
+      mem-fraction-static: 0.7
+      context-length: 9600
+      chunked-prefill-size: 32768
+      max-prefill-tokens: 8192
+
+        # Parallelism
+      tensor-parallel-size: 4
+      data-parallel-size: 4
+      expert-parallel-size: 1
+      enable-dp-attention: true
+      enable-dp-lm-head: true
+      load-balance-method: total_tokens
+
+        # Backend
+      nsa-decode-backend: trtllm
+      nsa-prefill-backend: trtllm
+      moe-runner-backend: flashinfer_trtllm
+
+        # Other flags
+      enable-flashinfer-allreduce-fusion: true
+      disable-radix-cache: true
+      weight-loader-prefetch-checkpoints: true
+      model-loader-extra-config: '{"enable_multithread_load": true}'
+
+    decode:
+        # Model configuration
+      served-model-name: GLM-5.1-FP8
+      trust-remote-code: true
+
+      quantization: fp8
+      kv-cache-dtype: fp8_e4m3
+
+        # Disaggregation mode
+      disaggregation-mode: decode
+      disaggregation-transfer-backend: nixl
+
+        # Memory and token limits
+      mem-fraction-static: 0.8
+      context-length: 9600
+
+        # Backend
+      nsa-decode-backend: trtllm
+      nsa-prefill-backend: trtllm
+        # moe-runner-backend: "cutedsl"
+
+        # Detokenizer
+      skip-tokenizer-init: true
+      stream-interval: 30
+
+        # Other flags
+      disable-radix-cache: true
+      weight-loader-prefetch-checkpoints: true
+      model-loader-extra-config: '{"enable_multithread_load": true}'
+      tensor-parallel-size: 40
+      expert-parallel-size: 40
+      data-parallel-size: 40
+      enable-dp-lm-head: true
+      enable-dp-attention: true
+      moe-dense-tp-size: 1
+      ep-num-redundant-experts: 24
+      ep-dispatch-algorithm: static
+      moe-a2a-backend: deepep
+      deepep-mode: low_latency
+      deepep-config: /configs/deepep_config.json
+      max-running-requests: 880
+      cuda-graph-max-bs: 22
+      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: '900'
diff --git a/benchmarks/multi_node/srt-slurm-recipes/sglang/glm5.1/gb300-fp8/8k1k/disagg/mtp/8k1k_mtp_lowlat_0.yaml b/benchmarks/multi_node/srt-slurm-recipes/sglang/glm5.1/gb300-fp8/8k1k/disagg/mtp/8k1k_mtp_lowlat_0.yaml
new file mode 100644
index 0000000000..baf3aa3609
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/sglang/glm5.1/gb300-fp8/8k1k/disagg/mtp/8k1k_mtp_lowlat_0.yaml
@@ -0,0 +1,144 @@
+name: gb300-fp8-glm5-mtp_8k1k_lowlat_0
+
+model:
+  path: glm-5.1-fp8
+  container: "lmsysorg/sglang:v0.5.11-cu130"
+  precision: fp8
+
+resources:
+  gpu_type: gb300
+  gpus_per_node: 4
+  prefill_nodes: 1
+  prefill_workers: 1
+  decode_nodes: 9
+  decode_workers: 9
+frontend:
+  type: dynamo
+dynamo:
+  version: 1.2.1
+
+backend:
+  prefill_environment:
+    TORCH_DISTRIBUTED_DEFAULT_TIMEOUT: '1800'
+    PYTHONUNBUFFERED: '1'
+    DYN_SKIP_SGLANG_LOG_FORMATTING: '1'
+    SGLANG_DISAGGREGATION_HEARTBEAT_MAX_FAILURE: '100000'
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: '100000'
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: '100000'
+    MC_TE_METRIC: 'true'
+    MC_FORCE_MNNVL: '1'
+    NCCL_MNNVL_ENABLE: '1'
+    NCCL_CUMEM_ENABLE: '1'
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: 'True'
+    SGLANG_USE_MESSAGE_QUEUE_BROADCASTER: '0'
+    SGLANG_DISABLE_TP_MEMORY_INBALANCE_CHECK: '1'
+    DYN_REQUEST_PLANE: nats
+
+  decode_environment:
+    TORCH_DISTRIBUTED_DEFAULT_TIMEOUT: '1800'
+    PYTHONUNBUFFERED: '1'
+    DYN_SKIP_SGLANG_LOG_FORMATTING: '1'
+    SGLANG_DISAGGREGATION_HEARTBEAT_MAX_FAILURE: '100000'
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: '100000'
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: '100000'
+    MC_TE_METRIC: 'true'
+    MC_FORCE_MNNVL: '1'
+    NCCL_MNNVL_ENABLE: '1'
+    NCCL_CUMEM_ENABLE: '1'
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: 'True'
+    SGLANG_USE_MESSAGE_QUEUE_BROADCASTER: '0'
+    SGLANG_DISABLE_TP_MEMORY_INBALANCE_CHECK: '1'
+    DYN_REQUEST_PLANE: nats
+      # DeepEP per-rank dispatch buffer; must be >= ceil(cuda_graph_max_bs / dp_size).
+      # Default 128 overflows with large DP + batch (e.g. 4096/24 ~= 171 > 128). Limit 1024.
+    SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK: '1024'
+
+  sglang_config:
+    prefill:
+        # Model configuration
+      served-model-name: GLM-5.1-FP8
+      trust-remote-code: true
+      quantization: fp8
+      kv-cache-dtype: fp8_e4m3
+
+        # Disaggregation mode
+      disaggregation-mode: prefill
+      disaggregation-transfer-backend: nixl
+
+        # Size limits
+      max-running-requests: 256
+      cuda-graph-max-bs: 256
+      mem-fraction-static: 0.7
+      context-length: 9600
+      chunked-prefill-size: 32768
+      max-prefill-tokens: 8192
+
+        # Parallelism
+      tensor-parallel-size: 4
+      data-parallel-size: 4
+      expert-parallel-size: 1
+      enable-dp-attention: true
+      enable-dp-lm-head: true
+      load-balance-method: total_tokens
+
+        # Backend
+      nsa-decode-backend: trtllm
+      nsa-prefill-backend: trtllm
+      moe-runner-backend: flashinfer_trtllm
+
+        # Other flags
+      enable-flashinfer-allreduce-fusion: true
+      disable-radix-cache: true
+      weight-loader-prefetch-checkpoints: true
+      model-loader-extra-config: '{"enable_multithread_load": true}'
+
+    decode:
+        # Model configuration
+      served-model-name: GLM-5.1-FP8
+      trust-remote-code: true
+
+      quantization: fp8
+      kv-cache-dtype: fp8_e4m3
+
+        # Disaggregation mode
+      disaggregation-mode: decode
+      disaggregation-transfer-backend: nixl
+
+        # Memory and token limits
+      mem-fraction-static: 0.8
+      context-length: 9600
+
+        # Backend
+      nsa-decode-backend: trtllm
+      nsa-prefill-backend: trtllm
+        # moe-runner-backend: "cutedsl"
+
+        # Detokenizer
+      skip-tokenizer-init: true
+      stream-interval: 30
+
+        # Other flags
+      disable-radix-cache: true
+      weight-loader-prefetch-checkpoints: true
+      model-loader-extra-config: '{"enable_multithread_load": true}'
+      tensor-parallel-size: 4
+      expert-parallel-size: 1
+      data-parallel-size: 1
+      enable-flashinfer-allreduce-fusion: true
+      moe-runner-backend: flashinfer_trtllm
+      max-running-requests: 15
+      cuda-graph-max-bs: 15
+      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: '150'
diff --git a/benchmarks/multi_node/srt-slurm-recipes/sglang/glm5.1/gb300-fp8/8k1k/disagg/mtp/8k1k_mtp_lowlat_1.yaml b/benchmarks/multi_node/srt-slurm-recipes/sglang/glm5.1/gb300-fp8/8k1k/disagg/mtp/8k1k_mtp_lowlat_1.yaml
new file mode 100644
index 0000000000..c386534d01
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/sglang/glm5.1/gb300-fp8/8k1k/disagg/mtp/8k1k_mtp_lowlat_1.yaml
@@ -0,0 +1,144 @@
+name: gb300-fp8-glm5-mtp_8k1k_lowlat_1
+
+model:
+  path: glm-5.1-fp8
+  container: "lmsysorg/sglang:v0.5.11-cu130"
+  precision: fp8
+
+resources:
+  gpu_type: gb300
+  gpus_per_node: 4
+  prefill_nodes: 1
+  prefill_workers: 1
+  decode_nodes: 17
+  decode_workers: 17
+frontend:
+  type: dynamo
+dynamo:
+  version: 1.2.1
+
+backend:
+  prefill_environment:
+    TORCH_DISTRIBUTED_DEFAULT_TIMEOUT: '1800'
+    PYTHONUNBUFFERED: '1'
+    DYN_SKIP_SGLANG_LOG_FORMATTING: '1'
+    SGLANG_DISAGGREGATION_HEARTBEAT_MAX_FAILURE: '100000'
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: '100000'
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: '100000'
+    MC_TE_METRIC: 'true'
+    MC_FORCE_MNNVL: '1'
+    NCCL_MNNVL_ENABLE: '1'
+    NCCL_CUMEM_ENABLE: '1'
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: 'True'
+    SGLANG_USE_MESSAGE_QUEUE_BROADCASTER: '0'
+    SGLANG_DISABLE_TP_MEMORY_INBALANCE_CHECK: '1'
+    DYN_REQUEST_PLANE: nats
+
+  decode_environment:
+    TORCH_DISTRIBUTED_DEFAULT_TIMEOUT: '1800'
+    PYTHONUNBUFFERED: '1'
+    DYN_SKIP_SGLANG_LOG_FORMATTING: '1'
+    SGLANG_DISAGGREGATION_HEARTBEAT_MAX_FAILURE: '100000'
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: '100000'
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: '100000'
+    MC_TE_METRIC: 'true'
+    MC_FORCE_MNNVL: '1'
+    NCCL_MNNVL_ENABLE: '1'
+    NCCL_CUMEM_ENABLE: '1'
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: 'True'
+    SGLANG_USE_MESSAGE_QUEUE_BROADCASTER: '0'
+    SGLANG_DISABLE_TP_MEMORY_INBALANCE_CHECK: '1'
+    DYN_REQUEST_PLANE: nats
+      # DeepEP per-rank dispatch buffer; must be >= ceil(cuda_graph_max_bs / dp_size).
+      # Default 128 overflows with large DP + batch (e.g. 4096/24 ~= 171 > 128). Limit 1024.
+    SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK: '1024'
+
+  sglang_config:
+    prefill:
+        # Model configuration
+      served-model-name: GLM-5.1-FP8
+      trust-remote-code: true
+      quantization: fp8
+      kv-cache-dtype: fp8_e4m3
+
+        # Disaggregation mode
+      disaggregation-mode: prefill
+      disaggregation-transfer-backend: nixl
+
+        # Size limits
+      max-running-requests: 256
+      cuda-graph-max-bs: 256
+      mem-fraction-static: 0.7
+      context-length: 9600
+      chunked-prefill-size: 32768
+      max-prefill-tokens: 8192
+
+        # Parallelism
+      tensor-parallel-size: 4
+      data-parallel-size: 4
+      expert-parallel-size: 1
+      enable-dp-attention: true
+      enable-dp-lm-head: true
+      load-balance-method: total_tokens
+
+        # Backend
+      nsa-decode-backend: trtllm
+      nsa-prefill-backend: trtllm
+      moe-runner-backend: flashinfer_trtllm
+
+        # Other flags
+      enable-flashinfer-allreduce-fusion: true
+      disable-radix-cache: true
+      weight-loader-prefetch-checkpoints: true
+      model-loader-extra-config: '{"enable_multithread_load": true}'
+
+    decode:
+        # Model configuration
+      served-model-name: GLM-5.1-FP8
+      trust-remote-code: true
+
+      quantization: fp8
+      kv-cache-dtype: fp8_e4m3
+
+        # Disaggregation mode
+      disaggregation-mode: decode
+      disaggregation-transfer-backend: nixl
+
+        # Memory and token limits
+      mem-fraction-static: 0.8
+      context-length: 9600
+
+        # Backend
+      nsa-decode-backend: trtllm
+      nsa-prefill-backend: trtllm
+        # moe-runner-backend: "cutedsl"
+
+        # Detokenizer
+      skip-tokenizer-init: true
+      stream-interval: 30
+
+        # Other flags
+      disable-radix-cache: true
+      weight-loader-prefetch-checkpoints: true
+      model-loader-extra-config: '{"enable_multithread_load": true}'
+      tensor-parallel-size: 4
+      expert-parallel-size: 1
+      data-parallel-size: 1
+      enable-flashinfer-allreduce-fusion: true
+      moe-runner-backend: flashinfer_trtllm
+      max-running-requests: 8
+      cuda-graph-max-bs: 8
+      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: 128x64x32
diff --git a/benchmarks/multi_node/srt-slurm-recipes/sglang/glm5.1/gb300-fp8/8k1k/disagg/mtp/8k1k_mtp_lowlat_2.yaml b/benchmarks/multi_node/srt-slurm-recipes/sglang/glm5.1/gb300-fp8/8k1k/disagg/mtp/8k1k_mtp_lowlat_2.yaml
new file mode 100644
index 0000000000..561ac280cc
--- /dev/null
+++ b/benchmarks/multi_node/srt-slurm-recipes/sglang/glm5.1/gb300-fp8/8k1k/disagg/mtp/8k1k_mtp_lowlat_2.yaml
@@ -0,0 +1,144 @@
+name: gb300-fp8-glm5-mtp_8k1k_lowlat_2
+
+model:
+  path: glm-5.1-fp8
+  container: "lmsysorg/sglang:v0.5.11-cu130"
+  precision: fp8
+
+resources:
+  gpu_type: gb300
+  gpus_per_node: 4
+  prefill_nodes: 1
+  prefill_workers: 1
+  decode_nodes: 17
+  decode_workers: 17
+frontend:
+  type: dynamo
+dynamo:
+  version: 1.2.1
+
+backend:
+  prefill_environment:
+    TORCH_DISTRIBUTED_DEFAULT_TIMEOUT: '1800'
+    PYTHONUNBUFFERED: '1'
+    DYN_SKIP_SGLANG_LOG_FORMATTING: '1'
+    SGLANG_DISAGGREGATION_HEARTBEAT_MAX_FAILURE: '100000'
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: '100000'
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: '100000'
+    MC_TE_METRIC: 'true'
+    MC_FORCE_MNNVL: '1'
+    NCCL_MNNVL_ENABLE: '1'
+    NCCL_CUMEM_ENABLE: '1'
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: 'True'
+    SGLANG_USE_MESSAGE_QUEUE_BROADCASTER: '0'
+    SGLANG_DISABLE_TP_MEMORY_INBALANCE_CHECK: '1'
+    DYN_REQUEST_PLANE: nats
+
+  decode_environment:
+    TORCH_DISTRIBUTED_DEFAULT_TIMEOUT: '1800'
+    PYTHONUNBUFFERED: '1'
+    DYN_SKIP_SGLANG_LOG_FORMATTING: '1'
+    SGLANG_DISAGGREGATION_HEARTBEAT_MAX_FAILURE: '100000'
+    SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: '100000'
+    SGLANG_DISAGGREGATION_WAITING_TIMEOUT: '100000'
+    MC_TE_METRIC: 'true'
+    MC_FORCE_MNNVL: '1'
+    NCCL_MNNVL_ENABLE: '1'
+    NCCL_CUMEM_ENABLE: '1'
+    SGLANG_MOONCAKE_CUSTOM_MEM_POOL: 'True'
+    SGLANG_USE_MESSAGE_QUEUE_BROADCASTER: '0'
+    SGLANG_DISABLE_TP_MEMORY_INBALANCE_CHECK: '1'
+    DYN_REQUEST_PLANE: nats
+      # DeepEP per-rank dispatch buffer; must be >= ceil(cuda_graph_max_bs / dp_size).
+      # Default 128 overflows with large DP + batch (e.g. 4096/24 ~= 171 > 128). Limit 1024.
+    SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK: '1024'
+
+  sglang_config:
+    prefill:
+        # Model configuration
+      served-model-name: GLM-5.1-FP8
+      trust-remote-code: true
+      quantization: fp8
+      kv-cache-dtype: fp8_e4m3
+
+        # Disaggregation mode
+      disaggregation-mode: prefill
+      disaggregation-transfer-backend: nixl
+
+        # Size limits
+      max-running-requests: 256
+      cuda-graph-max-bs: 256
+      mem-fraction-static: 0.7
+      context-length: 9600
+      chunked-prefill-size: 32768
+      max-prefill-tokens: 8192
+
+        # Parallelism
+      tensor-parallel-size: 4
+      data-parallel-size: 4
+      expert-parallel-size: 1
+      enable-dp-attention: true
+      enable-dp-lm-head: true
+      load-balance-method: total_tokens
+
+        # Backend
+      nsa-decode-backend: trtllm
+      nsa-prefill-backend: trtllm
+      moe-runner-backend: flashinfer_trtllm
+
+        # Other flags
+      enable-flashinfer-allreduce-fusion: true
+      disable-radix-cache: true
+      weight-loader-prefetch-checkpoints: true
+      model-loader-extra-config: '{"enable_multithread_load": true}'
+
+    decode:
+        # Model configuration
+      served-model-name: GLM-5.1-FP8
+      trust-remote-code: true
+
+      quantization: fp8
+      kv-cache-dtype: fp8_e4m3
+
+        # Disaggregation mode
+      disaggregation-mode: decode
+      disaggregation-transfer-backend: nixl
+
+        # Memory and token limits
+      mem-fraction-static: 0.8
+      context-length: 9600
+
+        # Backend
+      nsa-decode-backend: trtllm
+      nsa-prefill-backend: trtllm
+        # moe-runner-backend: "cutedsl"
+
+        # Detokenizer
+      skip-tokenizer-init: true
+      stream-interval: 30
+
+        # Other flags
+      disable-radix-cache: true
+      weight-loader-prefetch-checkpoints: true
+      model-loader-extra-config: '{"enable_multithread_load": true}'
+      tensor-parallel-size: 4
+      expert-parallel-size: 1
+      data-parallel-size: 1
+      enable-flashinfer-allreduce-fusion: true
+      moe-runner-backend: flashinfer_trtllm
+      max-running-requests: 1
+      cuda-graph-max-bs: 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
+  req_rate: inf
+  isl: 8192
+  osl: 1024
+  concurrencies: '24'
diff --git a/configs/nvidia-master.yaml b/configs/nvidia-master.yaml
index 79d785bba7..6be8e006f9 100644
--- a/configs/nvidia-master.yaml
+++ b/configs/nvidia-master.yaml
@@ -12292,6 +12292,188 @@ glm5-fp8-gb300-dynamo-sglang:
           ep: 1
           dp-attn: false
 
+glm5-fp8-gb300-dynamo-sglang-mtp:
+  image: lmsysorg/sglang:v0.5.11-cu130
+  model: zai-org/GLM-5.1-FP8
+  model-prefix: glm5.1
+  runner: gb300-nv
+  precision: fp8
+  framework: dynamo-sglang
+  multinode: true
+  disagg: true
+  scenarios:
+    fixed-seq-len:
+    # ---------- 8k1k high-throughput (wide-EP decode, EAGLE MTP) ----------
+    - isl: 8192
+      osl: 1024
+      search-space:
+      - spec-decoding: "mtp"
+        conc-list: [2800]
+        prefill:
+          num-worker: 14
+          tp: 4
+          ep: 1
+          dp-attn: true
+          additional-settings:
+          - "CONFIG_FILE=recipes/sglang/glm5.1/gb300-fp8/8k1k/disagg/mtp/8k1k_mtp_hightpt_0.yaml"
+        decode:
+          num-worker: 1
+          tp: 16
+          ep: 16
+          dp-attn: true
+      - spec-decoding: "mtp"
+        conc-list: [1700]
+        prefill:
+          num-worker: 12
+          tp: 4
+          ep: 1
+          dp-attn: true
+          additional-settings:
+          - "CONFIG_FILE=recipes/sglang/glm5.1/gb300-fp8/8k1k/disagg/mtp/8k1k_mtp_hightpt_1.yaml"
+        decode:
+          num-worker: 1
+          tp: 24
+          ep: 24
+          dp-attn: true
+      - spec-decoding: "mtp"
+        conc-list: [1300]
+        prefill:
+          num-worker: 10
+          tp: 4
+          ep: 1
+          dp-attn: true
+          additional-settings:
+          - "CONFIG_FILE=recipes/sglang/glm5.1/gb300-fp8/8k1k/disagg/mtp/8k1k_mtp_hightpt_2.yaml"
+        decode:
+          num-worker: 1
+          tp: 32
+          ep: 32
+          dp-attn: true
+      - spec-decoding: "mtp"
+        conc-list: [900]
+        prefill:
+          num-worker: 8
+          tp: 4
+          ep: 1
+          dp-attn: true
+          additional-settings:
+          - "CONFIG_FILE=recipes/sglang/glm5.1/gb300-fp8/8k1k/disagg/mtp/8k1k_mtp_hightpt_3.yaml"
+        decode:
+          num-worker: 1
+          tp: 40
+          ep: 40
+          dp-attn: true
+    # ---------- 8k1k low-latency (per-node TP=4 decode workers, EAGLE MTP) ----------
+    - isl: 8192
+      osl: 1024
+      search-space:
+      - spec-decoding: "mtp"
+        conc-list: [150]
+        prefill:
+          num-worker: 1
+          tp: 4
+          ep: 1
+          dp-attn: true
+          additional-settings:
+          - "CONFIG_FILE=recipes/sglang/glm5.1/gb300-fp8/8k1k/disagg/mtp/8k1k_mtp_lowlat_0.yaml"
+        decode:
+          num-worker: 9
+          tp: 4
+          ep: 1
+          dp-attn: false
+      - spec-decoding: "mtp"
+        conc-list: [128, 64, 32]
+        prefill:
+          num-worker: 1
+          tp: 4
+          ep: 1
+          dp-attn: true
+          additional-settings:
+          - "CONFIG_FILE=recipes/sglang/glm5.1/gb300-fp8/8k1k/disagg/mtp/8k1k_mtp_lowlat_1.yaml"
+        decode:
+          num-worker: 17
+          tp: 4
+          ep: 1
+          dp-attn: false
+      - spec-decoding: "mtp"
+        conc-list: [24]
+        prefill:
+          num-worker: 1
+          tp: 4
+          ep: 1
+          dp-attn: true
+          additional-settings:
+          - "CONFIG_FILE=recipes/sglang/glm5.1/gb300-fp8/8k1k/disagg/mtp/8k1k_mtp_lowlat_2.yaml"
+        decode:
+          num-worker: 17
+          tp: 4
+          ep: 1
+          dp-attn: false
+    # ---------- 1k1k high-throughput (wide-EP decode, EAGLE MTP) ----------
+    - isl: 1024
+      osl: 1024
+      search-space:
+      - spec-decoding: "mtp"
+        conc-list: [6500]
+        prefill:
+          num-worker: 6
+          tp: 4
+          ep: 1
+          dp-attn: true
+          additional-settings:
+          - "CONFIG_FILE=recipes/sglang/glm5.1/gb300-fp8/1k1k/disagg/mtp/1k1k_mtp_hightpt_3.yaml"
+        decode:
+          num-worker: 1
+          tp: 48
+          ep: 48
+          dp-attn: true
+      - spec-decoding: "mtp"
+        conc-list: [5700]
+        prefill:
+          num-worker: 4
+          tp: 4
+          ep: 1
+          dp-attn: true
+          additional-settings:
+          - "CONFIG_FILE=recipes/sglang/glm5.1/gb300-fp8/1k1k/disagg/mtp/1k1k_mtp_hightpt_4.yaml"
+        decode:
+          num-worker: 1
+          tp: 56
+          ep: 56
+          dp-attn: true
+    # ---------- 1k1k low-latency (per-node TP=4 decode workers, EAGLE MTP) ----------
+    - isl: 1024
+      osl: 1024
+      search-space:
+      - spec-decoding: "mtp"
+        conc-list: [512, 256, 128, 64]
+        prefill:
+          num-worker: 1
+          tp: 4
+          ep: 1
+          dp-attn: true
+          additional-settings:
+          - "CONFIG_FILE=recipes/sglang/glm5.1/gb300-fp8/1k1k/disagg/mtp/1k1k_mtp_lowlat_0.yaml"
+        decode:
+          num-worker: 17
+          tp: 4
+          ep: 1
+          dp-attn: false
+      - spec-decoding: "mtp"
+        conc-list: [32]
+        prefill:
+          num-worker: 1
+          tp: 4
+          ep: 1
+          dp-attn: true
+          additional-settings:
+          - "CONFIG_FILE=recipes/sglang/glm5.1/gb300-fp8/1k1k/disagg/mtp/1k1k_mtp_lowlat_1.yaml"
+        decode:
+          num-worker: 17
+          tp: 4
+          ep: 1
+          dp-attn: false
+
 # ============================================================================
 # Net-new agentic recipes from chore/agentx-v0.3 (no overlap with main entries).
 # Recipes that ALREADY existed on main were intentionally left at main's version
diff --git a/perf-changelog.yaml b/perf-changelog.yaml
index 05d5b021be..3425fd7c7c 100644
--- a/perf-changelog.yaml
+++ b/perf-changelog.yaml
@@ -4661,3 +4661,11 @@
     - "High-tpt 2P+1D DEP8 (conc 4096)"
     - "Max-tpt 3P+1D DEP8 (conc 4096); same engine config as high-tpt, one more prefill instance"
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2050
+
+- config-keys:
+    - glm5-fp8-gb300-dynamo-sglang-mtp
+  description:
+    - "Add GLM-5.1-FP8 GB300 multinode dynamo-sglang benchmark with EAGLE MTP speculative decoding"
+    - "Image: lmsysorg/sglang:v0.5.11-cu130"
+    - "11 topologies across 1k/1k and 8k/1k (prefill TP4 + decode wide-EP: 8k/1k DEP16/24/32/40, 1k/1k DEP48/56; per-node TP4 low-latency); MTP flags: speculative-algorithm EAGLE, num-steps 2, eagle-topk 1, num-draft-tokens 3"
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1907
diff --git a/runners/launch_gb300-nv.sh b/runners/launch_gb300-nv.sh
index 3180d9d215..5328ded02d 100644
--- a/runners/launch_gb300-nv.sh
+++ b/runners/launch_gb300-nv.sh
@@ -169,6 +169,14 @@ elif [[ $FRAMEWORK == "dynamo-sglang" && $MODEL_PREFIX == "glm5" ]]; then
         mkdir -p recipes/sglang/glm5/gb300-fp4
         cp -rT "$GITHUB_WORKSPACE/benchmarks/multi_node/srt-slurm-recipes/sglang/glm5/gb300-fp4" recipes/sglang/glm5/gb300-fp4
     fi
+elif [[ $FRAMEWORK == "dynamo-sglang" && $MODEL_PREFIX == "glm5.1" && $PRECISION == "fp8" ]]; then
+    # GLM-5.1 FP8 (gb300) recipes are version-controlled in-repo; overlay them
+    # onto the pinned submission branch.
+    git clone https://github.com/NVIDIA/srt-slurm.git "$SRT_REPO_DIR"
+    cd "$SRT_REPO_DIR"
+    git checkout sa-submission-q2-2026
+    mkdir -p recipes/sglang/glm5.1
+    cp -rT "$GITHUB_WORKSPACE/benchmarks/multi_node/srt-slurm-recipes/sglang/glm5.1" recipes/sglang/glm5.1
 elif [[ $FRAMEWORK == "dynamo-sglang" && $MODEL_PREFIX == "glm5.1" ]]; then
     # GLM-5.1 MTP recipe (recipes/gb300-fp4/glm5-mtp.yaml) lives on
     # NVIDIA/srt-slurm:main — check it out; no in-repo overlay needed.