diff --git a/examples/configs/recipes/llm/grpo-moonlight-16ba3b-4n8g-megatron-fp8-e2e.yaml b/examples/configs/recipes/llm/grpo-moonlight-16ba3b-4n8g-megatron-fp8-e2e.yaml
index 19c86c8a04..27108c55c7 100644
--- a/examples/configs/recipes/llm/grpo-moonlight-16ba3b-4n8g-megatron-fp8-e2e.yaml
+++ b/examples/configs/recipes/llm/grpo-moonlight-16ba3b-4n8g-megatron-fp8-e2e.yaml
@@ -16,8 +16,6 @@ policy:
   max_total_sequence_length: 8192
   dtensor_cfg:
     enabled: false
-  sequence_packing:
-    algorithm: modified_ffd
   make_sequence_length_divisible_by: ${policy.megatron_cfg.tensor_model_parallel_size}
   optimizer: null
   megatron_cfg:
@@ -45,7 +43,6 @@ policy:
       precision: fp8
       use_deep_gemm: true
       gpu_memory_utilization: 0.5
-      expert_parallel_size: 4
       quantization_ignored_layer_kws: [
         a_proj,
         b_proj
diff --git a/examples/configs/recipes/llm/performance/grpo-deepseek-v3-32n4g.yaml b/examples/configs/recipes/llm/performance/grpo-deepseek-v3-32n4g.yaml
new file mode 100644
index 0000000000..812fdfae9d
--- /dev/null
+++ b/examples/configs/recipes/llm/performance/grpo-deepseek-v3-32n4g.yaml
@@ -0,0 +1,21 @@
+defaults: ./grpo-deepseek-v3-32n8g.yaml
+checkpointing:
+  checkpoint_dir: results/grpo-deepseek-v3-32n4g
+policy:
+  sequence_packing:
+    enabled: false
+  megatron_cfg:
+    pipeline_model_parallel_size: 8
+    expert_model_parallel_size: 16
+    num_layers_in_first_pipeline_stage: 7
+    num_layers_in_last_pipeline_stage: 6
+  generation:
+    vllm_cfg:
+      tensor_parallel_size: 16
+logger:
+  log_dir: logs/grpo-deepseek-v3-32n4g
+  wandb:
+    name: grpo-deepseek-v3-32n4g
+cluster:
+  gpus_per_node: 4
+  num_nodes: 32
diff --git a/examples/configs/recipes/llm/performance/grpo-deepseek-v3-64n4g-async-1off.yaml b/examples/configs/recipes/llm/performance/grpo-deepseek-v3-64n4g-async-1off.yaml
new file mode 100644
index 0000000000..bf9a30a5d3
--- /dev/null
+++ b/examples/configs/recipes/llm/performance/grpo-deepseek-v3-64n4g-async-1off.yaml
@@ -0,0 +1,25 @@
+defaults: ./grpo-deepseek-v3-64n8g-async-1off.yaml
+checkpointing:
+  checkpoint_dir: results/grpo-deepseek-v3-64n4g-async-1off
+policy:
+  sequence_packing:
+    enabled: false
+  megatron_cfg:
+    pipeline_model_parallel_size: 8
+    expert_model_parallel_size: 16
+    num_layers_in_first_pipeline_stage: 7
+    num_layers_in_last_pipeline_stage: 6
+  generation:
+    colocated:
+      resources:
+        gpus_per_node: 4
+    vllm_cfg:
+      tensor_parallel_size: 16
+      gpu_memory_utilization: 0.8
+logger:
+  log_dir: logs/grpo-deepseek-v3-64n4g-async-32T32G-1off
+  wandb:
+    name: grpo-deepseek-v3-64n4g-async-32T32G-1off
+cluster:
+  gpus_per_node: 4
+  num_nodes: 64
diff --git a/examples/configs/recipes/llm/performance/grpo-deepseek-v3-64n8g-fp8-async-1off.yaml b/examples/configs/recipes/llm/performance/grpo-deepseek-v3-64n8g-fp8-async-1off.yaml
new file mode 100644
index 0000000000..7f6b5ae86b
--- /dev/null
+++ b/examples/configs/recipes/llm/performance/grpo-deepseek-v3-64n8g-fp8-async-1off.yaml
@@ -0,0 +1,28 @@
+defaults: ./grpo-deepseek-v3-64n8g-async-1off.yaml
+checkpointing:
+  checkpoint_dir: results/grpo-deepseek-v3-64n8g-fp8-async-1off
+policy:
+  megatron_cfg:
+    fp8_cfg:
+      enabled: true
+      fp8: "e4m3"
+      fp8_recipe: "blockwise"
+      fp8_param: false
+    moe_router_dtype: fp32
+    env_vars:
+      NVTE_FP8_BLOCK_SCALING_FP32_SCALES: "1"
+  generation:
+    vllm_cfg:
+      tensor_parallel_size: 16
+      precision: "fp8"
+      use_deep_gemm: true
+      quantization_ignored_layer_kws: [
+        a_proj,
+        b_proj
+      ]
+    vllm_kwargs:
+      max_num_seqs: 32
+logger:
+  log_dir: logs/grpo-deepseek-v3-64n8g-fp8-async-1off
+  wandb:
+    name: grpo-deepseek-v3-64n8g-fp8-async-1off
diff --git a/examples/configs/recipes/llm/performance/grpo-llama3.1-8b-instruct-2n4g-async-1off.yaml b/examples/configs/recipes/llm/performance/grpo-llama3.1-8b-instruct-2n4g-async-1off.yaml
new file mode 100644
index 0000000000..d906eda2b4
--- /dev/null
+++ b/examples/configs/recipes/llm/performance/grpo-llama3.1-8b-instruct-2n4g-async-1off.yaml
@@ -0,0 +1,32 @@
+defaults: ./grpo-llama3.1-8b-instruct-2n4g.yaml
+grpo:
+  async_grpo:
+    enabled: true
+    max_trajectory_age_steps: 1
+    in_flight_weight_updates: true
+loss_fn:
+  use_importance_sampling_correction: true
+checkpointing:
+  checkpoint_dir: results/grpo-llama3.1-8b-instruct-2n4g-async-1off
+policy:
+  megatron_cfg:
+    tensor_model_parallel_size: 1
+    pipeline_model_parallel_size: 1
+    sequence_parallel: false
+  generation:
+    colocated:
+      enabled: false
+      resources:
+        num_nodes: 1
+        gpus_per_node: 4
+    vllm_cfg:
+      async_engine: true
+      tensor_parallel_size: 1
+      gpu_memory_utilization: 0.8
+logger:
+  log_dir: logs/grpo-llama3.1-8b-instruct-2n4g-async-1off
+  wandb:
+    name: grpo-llama3.1-8b-instruct-2n4g-async-1off
+cluster:
+  gpus_per_node: 4
+  num_nodes: 2
diff --git a/examples/configs/recipes/llm/performance/grpo-llama3.1-8b-instruct-2n4g.yaml b/examples/configs/recipes/llm/performance/grpo-llama3.1-8b-instruct-2n4g.yaml
new file mode 100644
index 0000000000..a99f7c1498
--- /dev/null
+++ b/examples/configs/recipes/llm/performance/grpo-llama3.1-8b-instruct-2n4g.yaml
@@ -0,0 +1,58 @@
+defaults: ../../../grpo_math_1B.yaml
+grpo:
+  num_prompts_per_step: 64
+  num_generations_per_prompt: 32
+  max_num_steps: 500
+loss_fn:
+  use_importance_sampling_correction: true
+checkpointing:
+  checkpoint_dir: results/grpo-llama3.1-8b-instruct-2n4g
+policy:
+  model_name: meta-llama/Llama-3.1-8B-Instruct
+  tokenizer:
+    name: meta-llama/Llama-3.1-8B-Instruct
+  train_micro_batch_size: 1
+  logprob_batch_size: 2
+  max_total_sequence_length: 4096
+  make_sequence_length_divisible_by: 1
+  dtensor_cfg:
+    enabled: false
+  megatron_cfg:
+    enabled: true
+    empty_unused_memory_level: 1
+    converter_type: LlamaForCausalLM
+    tensor_model_parallel_size: 1
+    pipeline_model_parallel_size: 1
+    sequence_parallel: false
+    activation_checkpointing: true
+    defer_fp32_logits: true
+    optimizer:
+      lr: 5.0e-07
+      min_lr: 5.0e-08
+      weight_decay: 0.0
+      use_precision_aware_optimizer: true
+    scheduler:
+      lr_warmup_iters: 2
+      lr_warmup_init: 5.0e-08
+    fp8_cfg:
+      enabled: false
+  generation:
+    max_new_tokens: 4096
+    stop_token_ids:
+    - 128009
+    vllm_cfg:
+      max_model_len: 4096
+      tensor_parallel_size: 1
+data:
+  max_input_seq_length: 4096
+logger:
+  log_dir: logs/grpo-llama3.1-8b-instruct-2n4g
+  wandb_enabled: true
+  tensorboard_enabled: true
+  wandb:
+    project: nemo-rl
+    name: grpo-llama3.1-8b-instruct-2n4g
+cluster:
+  gpus_per_node: 4
+  num_nodes: 2
+
diff --git a/examples/configs/recipes/llm/performance/grpo-llama3.1-8b-instruct-2n8g-async-1off.yaml b/examples/configs/recipes/llm/performance/grpo-llama3.1-8b-instruct-2n8g-async-1off.yaml
index b6d7ed441d..c0263f68fb 100644
--- a/examples/configs/recipes/llm/performance/grpo-llama3.1-8b-instruct-2n8g-async-1off.yaml
+++ b/examples/configs/recipes/llm/performance/grpo-llama3.1-8b-instruct-2n8g-async-1off.yaml
@@ -9,6 +9,8 @@ loss_fn:
 checkpointing:
   checkpoint_dir: results/grpo-llama3.1-8b-instruct-2n8g-async-1off
 policy:
+  megatron_cfg:
+    pipeline_model_parallel_size: 1
   generation:
     colocated:
       enabled: false
diff --git a/examples/configs/recipes/llm/performance/grpo-llama3.1-8b-instruct-2n8g-fp8-async-1off.yaml b/examples/configs/recipes/llm/performance/grpo-llama3.1-8b-instruct-2n8g-fp8-async-1off.yaml
new file mode 100644
index 0000000000..b32786f7d7
--- /dev/null
+++ b/examples/configs/recipes/llm/performance/grpo-llama3.1-8b-instruct-2n8g-fp8-async-1off.yaml
@@ -0,0 +1,20 @@
+defaults: ./grpo-llama3.1-8b-instruct-2n8g-async-1off.yaml
+checkpointing:
+  checkpoint_dir: results/grpo-llama3.1-8b-instruct-2n8g-fp8-async-1off
+policy:
+  megatron_cfg:
+    fp8_cfg:
+      enabled: true
+      fp8: "e4m3"
+      fp8_recipe: "blockwise"
+      fp8_param: false
+    env_vars:
+      NVTE_FP8_BLOCK_SCALING_FP32_SCALES: "1"
+  generation:
+    vllm_cfg:
+      precision: "fp8"
+      use_deep_gemm: true
+logger:
+  log_dir: logs/grpo-llama3.1-8b-instruct-2n8g-fp8-async-1off
+  wandb:
+    name: grpo-llama3.1-8b-instruct-2n8g-fp8-async-1off
\ No newline at end of file
diff --git a/examples/configs/recipes/llm/performance/grpo-qwen3-235b-16n4g.yaml b/examples/configs/recipes/llm/performance/grpo-qwen3-235b-16n4g.yaml
new file mode 100644
index 0000000000..1640deda09
--- /dev/null
+++ b/examples/configs/recipes/llm/performance/grpo-qwen3-235b-16n4g.yaml
@@ -0,0 +1,18 @@
+defaults: ./grpo-qwen3-235b-16n8g.yaml
+checkpointing:
+  checkpoint_dir: results/grpo-qwen3-235b-16n4g
+policy:
+  megatron_cfg:
+    pipeline_model_parallel_size: 4
+    num_layers_in_first_pipeline_stage: 23
+    num_layers_in_last_pipeline_stage: 23
+  generation:
+    vllm_cfg:
+      tensor_parallel_size: 8
+logger:
+  log_dir: logs/grpo-qwen3-235b-16n4g
+  wandb:
+    name: grpo-qwen3-235b-16n4g
+cluster:
+  gpus_per_node: 4
+  num_nodes: 16
diff --git a/examples/configs/recipes/llm/performance/grpo-qwen3-235b-32n4g-async-1off.yaml b/examples/configs/recipes/llm/performance/grpo-qwen3-235b-32n4g-async-1off.yaml
new file mode 100644
index 0000000000..f55b383686
--- /dev/null
+++ b/examples/configs/recipes/llm/performance/grpo-qwen3-235b-32n4g-async-1off.yaml
@@ -0,0 +1,21 @@
+defaults: ./grpo-qwen3-235b-32n8g-async-1off.yaml
+checkpointing:
+  checkpoint_dir: results/grpo-qwen3-235b-32n4g-async-1off
+policy:
+  megatron_cfg:
+    pipeline_model_parallel_size: 4
+    num_layers_in_first_pipeline_stage: 23
+    num_layers_in_last_pipeline_stage: 23
+  generation:
+    colocated:
+      resources:
+        gpus_per_node: 4
+    vllm_cfg:
+      tensor_parallel_size: 8
+logger:
+  log_dir: logs/grpo-qwen3-235b-32n4g-async-1off
+  wandb:
+    name: grpo-qwen3-235b-32n4g-async-1off
+cluster:
+  gpus_per_node: 4
+  num_nodes: 32
diff --git a/examples/configs/recipes/llm/performance/grpo-qwen3-30ba3b-4n4g.yaml b/examples/configs/recipes/llm/performance/grpo-qwen3-30ba3b-4n4g.yaml
new file mode 100644
index 0000000000..21b9746f4b
--- /dev/null
+++ b/examples/configs/recipes/llm/performance/grpo-qwen3-30ba3b-4n4g.yaml
@@ -0,0 +1,45 @@
+defaults: ../../../grpo_math_1B.yaml
+grpo:
+  num_prompts_per_step: 64
+  num_generations_per_prompt: 32
+checkpointing:
+  enabled: false
+  checkpoint_dir: results/grpo-qwen3-30ba3b-4n4g
+policy:
+  model_name: Qwen/Qwen3-30B-A3B
+  train_micro_batch_size: 1
+  max_total_sequence_length: 4096
+  dtensor_cfg:
+    enabled: false
+  optimizer: null
+  scheduler: null
+  make_sequence_length_divisible_by: ${policy.megatron_cfg.tensor_model_parallel_size}
+  megatron_cfg:
+    enabled: true
+    empty_unused_memory_level: 1
+    tensor_model_parallel_size: 1
+    pipeline_model_parallel_size: 1
+    expert_model_parallel_size: 16
+    sequence_parallel: false
+    optimizer:
+      lr: 3.0e-07
+      min_lr: 3.0e-08
+    scheduler:
+      lr_warmup_iters: 50
+      lr_warmup_init: 3.0e-08
+    env_vars:
+      PYTORCH_CUDA_ALLOC_CONF: expandable_segments:False
+  generation:
+    vllm_cfg:
+      tensor_parallel_size: 1
+logger:
+  log_dir: logs/grpo-qwen3-30ba3b-4n4g
+  wandb_enabled: true
+  tensorboard_enabled: true
+  wandb:
+    project: nemo-rl
+    name: grpo-qwen3-30ba3b-4n4g
+cluster:
+  gpus_per_node: 4
+  num_nodes: 4
+
diff --git a/examples/configs/recipes/llm/performance/grpo-qwen3-30ba3b-8n4g-async-1off.yaml b/examples/configs/recipes/llm/performance/grpo-qwen3-30ba3b-8n4g-async-1off.yaml
new file mode 100644
index 0000000000..a9837c87f2
--- /dev/null
+++ b/examples/configs/recipes/llm/performance/grpo-qwen3-30ba3b-8n4g-async-1off.yaml
@@ -0,0 +1,33 @@
+defaults: ./grpo-qwen3-30ba3b-4n4g.yaml
+grpo:
+  async_grpo:
+    enabled: true
+    max_trajectory_age_steps: 1
+    in_flight_weight_updates: true
+loss_fn:
+  use_importance_sampling_correction: true
+checkpointing:
+  checkpoint_dir: results/grpo-qwen3-30ba3b-8n4g-async-1off
+policy:
+  megatron_cfg:
+    tensor_model_parallel_size: 1
+    pipeline_model_parallel_size: 1
+    expert_model_parallel_size: 16
+    sequence_parallel: false
+  generation:
+    colocated:
+      enabled: false
+      resources:
+        num_nodes: 4
+        gpus_per_node: 4
+    vllm_cfg:
+      async_engine: true
+      tensor_parallel_size: 1
+      gpu_memory_utilization: 0.8
+logger:
+  log_dir: logs/grpo-qwen3-30ba3b-8n4g-async-1off
+  wandb:
+    name: grpo-qwen3-30ba3b-8n4g-async-1off
+cluster:
+  gpus_per_node: 4
+  num_nodes: 8
diff --git a/examples/configs/recipes/llm/performance/grpo-qwen3-32b-4n4g.yaml b/examples/configs/recipes/llm/performance/grpo-qwen3-32b-4n4g.yaml
new file mode 100644
index 0000000000..2e441cdb5f
--- /dev/null
+++ b/examples/configs/recipes/llm/performance/grpo-qwen3-32b-4n4g.yaml
@@ -0,0 +1,42 @@
+defaults: ../../../grpo_math_1B.yaml
+grpo:
+  num_prompts_per_step: 64
+  num_generations_per_prompt: 32
+checkpointing:
+  enabled: false
+  checkpoint_dir: results/grpo-qwen3-32b-4n4g
+policy:
+  model_name: Qwen/Qwen3-32B
+  train_micro_batch_size: 1
+  max_total_sequence_length: 4096
+  dtensor_cfg:
+    enabled: false
+  optimizer: null
+  scheduler: null
+  make_sequence_length_divisible_by: ${policy.megatron_cfg.tensor_model_parallel_size}
+  megatron_cfg:
+    enabled: true
+    empty_unused_memory_level: 1
+    tensor_model_parallel_size: 2
+    pipeline_model_parallel_size: 1
+    sequence_parallel: true
+    optimizer:
+      lr: 3.0e-07
+      min_lr: 3.0e-08
+    scheduler:
+      lr_warmup_iters: 2
+      lr_warmup_init: 3.0e-08
+  generation:
+    vllm_cfg:
+      tensor_parallel_size: 1
+logger:
+  log_dir: logs/grpo-qwen3-32b-4n4g
+  wandb_enabled: true
+  tensorboard_enabled: true
+  wandb:
+    project: nemo-rl
+    name: grpo-qwen3-32b-4n4g
+cluster:
+  gpus_per_node: 4
+  num_nodes: 4
+
diff --git a/examples/configs/recipes/llm/performance/grpo-qwen3-32b-8n4g-async-1off.yaml b/examples/configs/recipes/llm/performance/grpo-qwen3-32b-8n4g-async-1off.yaml
new file mode 100644
index 0000000000..4f8a0a03bb
--- /dev/null
+++ b/examples/configs/recipes/llm/performance/grpo-qwen3-32b-8n4g-async-1off.yaml
@@ -0,0 +1,33 @@
+defaults: ./grpo-qwen3-32b-4n4g.yaml
+grpo:
+  async_grpo:
+    enabled: true
+    max_trajectory_age_steps: 1
+    in_flight_weight_updates: true
+loss_fn:
+  use_importance_sampling_correction: true
+checkpointing:
+  checkpoint_dir: results/grpo-qwen3-32b-8n4g-async-1off
+policy:
+  megatron_cfg:
+    tensor_model_parallel_size: 2
+    pipeline_model_parallel_size: 1
+    sequence_parallel: true
+  generation:
+    colocated:
+      enabled: false
+      resources:
+        num_nodes: 4
+        gpus_per_node: 4
+    vllm_cfg:
+      async_engine: true
+      tensor_parallel_size: 1
+      gpu_memory_utilization: 0.8
+logger:
+  log_dir: logs/grpo-qwen3-32b-8n4g-async-1off
+  wandb:
+    name: grpo-qwen3-32b-8n4g-async-1off
+cluster:
+  gpus_per_node: 4
+  num_nodes: 8
+
diff --git a/tests/test_suites/llm/performance/grpo-deepseek-v3-32n4g.sh b/tests/test_suites/llm/performance/grpo-deepseek-v3-32n4g.sh
new file mode 100755
index 0000000000..738b38dd5b
--- /dev/null
+++ b/tests/test_suites/llm/performance/grpo-deepseek-v3-32n4g.sh
@@ -0,0 +1,45 @@
+#!/bin/bash
+SCRIPT_DIR=$( cd -- "$( dirname -- "${BASH_SOURCE[0]}" )" &> /dev/null && pwd)
+source $SCRIPT_DIR/common.env
+# disable NVLS to avoid OOM issue
+export NCCL_NVLS_ENABLE=0
+# allow user to pass an existing HF checkpoint path based on instruction in https://github.com/NVIDIA-NeMo/RL/blob/main/docs/guides/deepseek.md
+export MODEL_NAME=${NRL_DEEPSEEK_V3_HF_CKPT:-"unsloth/DeepSeek-V3-0324-BF16"}
+
+# ===== BEGIN CONFIG =====
+NUM_NODES=32
+STEPS_PER_RUN=10
+MAX_STEPS=10
+NUM_RUNS=$(( (MAX_STEPS + STEPS_PER_RUN - 1) / STEPS_PER_RUN ))  # Round up
+NUM_MINUTES=240
+# ===== END CONFIG =====
+
+exit_if_max_steps_reached
+
+# Run the experiment
+cd $PROJECT_ROOT
+uv run examples/run_grpo_math.py \
+    --config $CONFIG_PATH \
+    grpo.max_num_steps=$MAX_STEPS \
+    policy.model_name=$MODEL_NAME \
+    policy.tokenizer.name=$MODEL_NAME \
+    logger.log_dir=$LOG_DIR \
+    logger.wandb_enabled=True \
+    logger.wandb.project=nemo-rl \
+    logger.wandb.name=$EXP_NAME \
+    logger.monitor_gpus=True \
+    logger.tensorboard_enabled=True \
+    checkpointing.enabled=True \
+    checkpointing.checkpoint_dir=$CKPT_DIR \
+    $@ \
+    2>&1 | tee $RUN_LOG
+
+# Convert tensorboard logs to json
+uv run tests/json_dump_tb_logs.py $LOG_DIR --output_path $JSON_METRICS
+
+# Only run metrics if the target step is reached
+if [[ $(jq 'to_entries | .[] | select(.key == "train/loss") | .value | keys | map(tonumber) | max' $JSON_METRICS) -ge $MAX_STEPS ]]; then
+    uv run tests/check_metrics.py $JSON_METRICS \
+        'mean(data["train/token_mult_prob_error"]) < 1.1' \
+        'data["train/token_mult_prob_error"]["10"] < 1.1'
+fi
diff --git a/tests/test_suites/llm/performance/grpo-deepseek-v3-64n4g-async-1off.sh b/tests/test_suites/llm/performance/grpo-deepseek-v3-64n4g-async-1off.sh
new file mode 100755
index 0000000000..14138486e1
--- /dev/null
+++ b/tests/test_suites/llm/performance/grpo-deepseek-v3-64n4g-async-1off.sh
@@ -0,0 +1,45 @@
+#!/bin/bash
+SCRIPT_DIR=$( cd -- "$( dirname -- "${BASH_SOURCE[0]}" )" &> /dev/null && pwd)
+source $SCRIPT_DIR/common.env
+# disable NVLS to avoid OOM issue
+export NCCL_NVLS_ENABLE=0
+# allow user to pass an existing HF checkpoint path based on instruction in https://github.com/NVIDIA-NeMo/RL/blob/main/docs/guides/deepseek.md
+export MODEL_NAME=${NRL_DEEPSEEK_V3_HF_CKPT:-"unsloth/DeepSeek-V3-0324-BF16"}
+
+# ===== BEGIN CONFIG =====
+NUM_NODES=64
+STEPS_PER_RUN=10
+MAX_STEPS=10
+NUM_RUNS=$(( (MAX_STEPS + STEPS_PER_RUN - 1) / STEPS_PER_RUN ))  # Round up
+NUM_MINUTES=240
+# ===== END CONFIG =====
+
+exit_if_max_steps_reached
+
+# Run the experiment
+cd $PROJECT_ROOT
+uv run examples/run_grpo_math.py \
+    --config $CONFIG_PATH \
+    grpo.max_num_steps=$MAX_STEPS \
+    policy.model_name=$MODEL_NAME \
+    policy.tokenizer.name=$MODEL_NAME \
+    logger.log_dir=$LOG_DIR \
+    logger.wandb_enabled=True \
+    logger.wandb.project=nemo-rl \
+    logger.wandb.name=$EXP_NAME \
+    logger.monitor_gpus=True \
+    logger.tensorboard_enabled=True \
+    checkpointing.enabled=True \
+    checkpointing.checkpoint_dir=$CKPT_DIR \
+    $@ \
+    2>&1 | tee $RUN_LOG
+
+# Convert tensorboard logs to json
+uv run tests/json_dump_tb_logs.py $LOG_DIR --output_path $JSON_METRICS
+
+# Only run metrics if the target step is reached
+if [[ $(jq 'to_entries | .[] | select(.key == "train/loss") | .value | keys | map(tonumber) | max' $JSON_METRICS) -ge $MAX_STEPS ]]; then
+    uv run tests/check_metrics.py $JSON_METRICS \
+        'mean(data["train/token_mult_prob_error"]) < 1.1' \
+        'data["train/token_mult_prob_error"]["10"] < 1.1'
+fi
diff --git a/tests/test_suites/llm/performance/grpo-deepseek-v3-64n8g-fp8-async-1off.sh b/tests/test_suites/llm/performance/grpo-deepseek-v3-64n8g-fp8-async-1off.sh
new file mode 100755
index 0000000000..14138486e1
--- /dev/null
+++ b/tests/test_suites/llm/performance/grpo-deepseek-v3-64n8g-fp8-async-1off.sh
@@ -0,0 +1,45 @@
+#!/bin/bash
+SCRIPT_DIR=$( cd -- "$( dirname -- "${BASH_SOURCE[0]}" )" &> /dev/null && pwd)
+source $SCRIPT_DIR/common.env
+# disable NVLS to avoid OOM issue
+export NCCL_NVLS_ENABLE=0
+# allow user to pass an existing HF checkpoint path based on instruction in https://github.com/NVIDIA-NeMo/RL/blob/main/docs/guides/deepseek.md
+export MODEL_NAME=${NRL_DEEPSEEK_V3_HF_CKPT:-"unsloth/DeepSeek-V3-0324-BF16"}
+
+# ===== BEGIN CONFIG =====
+NUM_NODES=64
+STEPS_PER_RUN=10
+MAX_STEPS=10
+NUM_RUNS=$(( (MAX_STEPS + STEPS_PER_RUN - 1) / STEPS_PER_RUN ))  # Round up
+NUM_MINUTES=240
+# ===== END CONFIG =====
+
+exit_if_max_steps_reached
+
+# Run the experiment
+cd $PROJECT_ROOT
+uv run examples/run_grpo_math.py \
+    --config $CONFIG_PATH \
+    grpo.max_num_steps=$MAX_STEPS \
+    policy.model_name=$MODEL_NAME \
+    policy.tokenizer.name=$MODEL_NAME \
+    logger.log_dir=$LOG_DIR \
+    logger.wandb_enabled=True \
+    logger.wandb.project=nemo-rl \
+    logger.wandb.name=$EXP_NAME \
+    logger.monitor_gpus=True \
+    logger.tensorboard_enabled=True \
+    checkpointing.enabled=True \
+    checkpointing.checkpoint_dir=$CKPT_DIR \
+    $@ \
+    2>&1 | tee $RUN_LOG
+
+# Convert tensorboard logs to json
+uv run tests/json_dump_tb_logs.py $LOG_DIR --output_path $JSON_METRICS
+
+# Only run metrics if the target step is reached
+if [[ $(jq 'to_entries | .[] | select(.key == "train/loss") | .value | keys | map(tonumber) | max' $JSON_METRICS) -ge $MAX_STEPS ]]; then
+    uv run tests/check_metrics.py $JSON_METRICS \
+        'mean(data["train/token_mult_prob_error"]) < 1.1' \
+        'data["train/token_mult_prob_error"]["10"] < 1.1'
+fi
diff --git a/tests/test_suites/llm/performance/grpo-llama3.1-8b-instruct-2n4g-async-1off.sh b/tests/test_suites/llm/performance/grpo-llama3.1-8b-instruct-2n4g-async-1off.sh
new file mode 100755
index 0000000000..e7636f3e93
--- /dev/null
+++ b/tests/test_suites/llm/performance/grpo-llama3.1-8b-instruct-2n4g-async-1off.sh
@@ -0,0 +1,39 @@
+#!/bin/bash
+SCRIPT_DIR=$( cd -- "$( dirname -- "${BASH_SOURCE[0]}" )" &> /dev/null && pwd)
+source $SCRIPT_DIR/common.env
+
+# ===== BEGIN CONFIG =====
+NUM_NODES=2
+STEPS_PER_RUN=10
+MAX_STEPS=10
+NUM_RUNS=$(( (MAX_STEPS + STEPS_PER_RUN - 1) / STEPS_PER_RUN ))  # Round up
+NUM_MINUTES=100
+# ===== END CONFIG =====
+
+exit_if_max_steps_reached
+
+# Run the experiment
+cd $PROJECT_ROOT
+uv run examples/run_grpo_math.py \
+    --config $CONFIG_PATH \
+    grpo.max_num_steps=$MAX_STEPS \
+    logger.log_dir=$LOG_DIR \
+    logger.wandb_enabled=True \
+    logger.wandb.project=nemo-rl \
+    logger.wandb.name=$EXP_NAME \
+    logger.monitor_gpus=True \
+    logger.tensorboard_enabled=True \
+    checkpointing.enabled=True \
+    checkpointing.checkpoint_dir=$CKPT_DIR \
+    $@ \
+    2>&1 | tee $RUN_LOG
+
+# Convert tensorboard logs to json
+uv run tests/json_dump_tb_logs.py $LOG_DIR --output_path $JSON_METRICS
+
+# Only run metrics if the target step is reached
+if [[ $(jq 'to_entries | .[] | select(.key == "train/loss") | .value | keys | map(tonumber) | max' $JSON_METRICS) -ge $MAX_STEPS ]]; then
+    uv run tests/check_metrics.py $JSON_METRICS \
+        'mean(data["train/token_mult_prob_error"]) < 1.1' \
+        'data["train/token_mult_prob_error"]["10"] < 1.1'
+fi
diff --git a/tests/test_suites/llm/performance/grpo-llama3.1-8b-instruct-2n4g.sh b/tests/test_suites/llm/performance/grpo-llama3.1-8b-instruct-2n4g.sh
new file mode 100755
index 0000000000..e7636f3e93
--- /dev/null
+++ b/tests/test_suites/llm/performance/grpo-llama3.1-8b-instruct-2n4g.sh
@@ -0,0 +1,39 @@
+#!/bin/bash
+SCRIPT_DIR=$( cd -- "$( dirname -- "${BASH_SOURCE[0]}" )" &> /dev/null && pwd)
+source $SCRIPT_DIR/common.env
+
+# ===== BEGIN CONFIG =====
+NUM_NODES=2
+STEPS_PER_RUN=10
+MAX_STEPS=10
+NUM_RUNS=$(( (MAX_STEPS + STEPS_PER_RUN - 1) / STEPS_PER_RUN ))  # Round up
+NUM_MINUTES=100
+# ===== END CONFIG =====
+
+exit_if_max_steps_reached
+
+# Run the experiment
+cd $PROJECT_ROOT
+uv run examples/run_grpo_math.py \
+    --config $CONFIG_PATH \
+    grpo.max_num_steps=$MAX_STEPS \
+    logger.log_dir=$LOG_DIR \
+    logger.wandb_enabled=True \
+    logger.wandb.project=nemo-rl \
+    logger.wandb.name=$EXP_NAME \
+    logger.monitor_gpus=True \
+    logger.tensorboard_enabled=True \
+    checkpointing.enabled=True \
+    checkpointing.checkpoint_dir=$CKPT_DIR \
+    $@ \
+    2>&1 | tee $RUN_LOG
+
+# Convert tensorboard logs to json
+uv run tests/json_dump_tb_logs.py $LOG_DIR --output_path $JSON_METRICS
+
+# Only run metrics if the target step is reached
+if [[ $(jq 'to_entries | .[] | select(.key == "train/loss") | .value | keys | map(tonumber) | max' $JSON_METRICS) -ge $MAX_STEPS ]]; then
+    uv run tests/check_metrics.py $JSON_METRICS \
+        'mean(data["train/token_mult_prob_error"]) < 1.1' \
+        'data["train/token_mult_prob_error"]["10"] < 1.1'
+fi
diff --git a/tests/test_suites/llm/performance/grpo-llama3.1-8b-instruct-2n8g-fp8-async-1off.sh b/tests/test_suites/llm/performance/grpo-llama3.1-8b-instruct-2n8g-fp8-async-1off.sh
new file mode 100755
index 0000000000..e7636f3e93
--- /dev/null
+++ b/tests/test_suites/llm/performance/grpo-llama3.1-8b-instruct-2n8g-fp8-async-1off.sh
@@ -0,0 +1,39 @@
+#!/bin/bash
+SCRIPT_DIR=$( cd -- "$( dirname -- "${BASH_SOURCE[0]}" )" &> /dev/null && pwd)
+source $SCRIPT_DIR/common.env
+
+# ===== BEGIN CONFIG =====
+NUM_NODES=2
+STEPS_PER_RUN=10
+MAX_STEPS=10
+NUM_RUNS=$(( (MAX_STEPS + STEPS_PER_RUN - 1) / STEPS_PER_RUN ))  # Round up
+NUM_MINUTES=100
+# ===== END CONFIG =====
+
+exit_if_max_steps_reached
+
+# Run the experiment
+cd $PROJECT_ROOT
+uv run examples/run_grpo_math.py \
+    --config $CONFIG_PATH \
+    grpo.max_num_steps=$MAX_STEPS \
+    logger.log_dir=$LOG_DIR \
+    logger.wandb_enabled=True \
+    logger.wandb.project=nemo-rl \
+    logger.wandb.name=$EXP_NAME \
+    logger.monitor_gpus=True \
+    logger.tensorboard_enabled=True \
+    checkpointing.enabled=True \
+    checkpointing.checkpoint_dir=$CKPT_DIR \
+    $@ \
+    2>&1 | tee $RUN_LOG
+
+# Convert tensorboard logs to json
+uv run tests/json_dump_tb_logs.py $LOG_DIR --output_path $JSON_METRICS
+
+# Only run metrics if the target step is reached
+if [[ $(jq 'to_entries | .[] | select(.key == "train/loss") | .value | keys | map(tonumber) | max' $JSON_METRICS) -ge $MAX_STEPS ]]; then
+    uv run tests/check_metrics.py $JSON_METRICS \
+        'mean(data["train/token_mult_prob_error"]) < 1.1' \
+        'data["train/token_mult_prob_error"]["10"] < 1.1'
+fi
diff --git a/tests/test_suites/llm/performance/grpo-qwen3-235b-16n4g.sh b/tests/test_suites/llm/performance/grpo-qwen3-235b-16n4g.sh
new file mode 100755
index 0000000000..0f9bf9289f
--- /dev/null
+++ b/tests/test_suites/llm/performance/grpo-qwen3-235b-16n4g.sh
@@ -0,0 +1,40 @@
+#!/bin/bash
+SCRIPT_DIR=$( cd -- "$( dirname -- "${BASH_SOURCE[0]}" )" &> /dev/null && pwd)
+source $SCRIPT_DIR/common.env
+# disable NVLS to avoid OOM issue
+export NCCL_NVLS_ENABLE=0
+
+# ===== BEGIN CONFIG =====
+NUM_NODES=16
+STEPS_PER_RUN=10
+MAX_STEPS=10
+NUM_RUNS=$(( (MAX_STEPS + STEPS_PER_RUN - 1) / STEPS_PER_RUN ))  # Round up
+NUM_MINUTES=100
+# ===== END CONFIG =====
+
+exit_if_max_steps_reached
+
+# Run the experiment
+cd $PROJECT_ROOT
+uv run examples/run_grpo_math.py \
+    --config $CONFIG_PATH \
+    grpo.max_num_steps=$MAX_STEPS \
+    logger.log_dir=$LOG_DIR \
+    logger.wandb_enabled=True \
+    logger.wandb.project=nemo-rl \
+    logger.wandb.name=$EXP_NAME \
+    logger.monitor_gpus=True \
+    checkpointing.enabled=True \
+    checkpointing.checkpoint_dir=$CKPT_DIR \
+    $@ \
+    2>&1 | tee $RUN_LOG
+
+# Convert tensorboard logs to json
+uv run tests/json_dump_tb_logs.py $LOG_DIR --output_path $JSON_METRICS
+
+# Only run metrics if the target step is reached
+if [[ $(jq 'to_entries | .[] | select(.key == "train/loss") | .value | keys | map(tonumber) | max' $JSON_METRICS) -ge $MAX_STEPS ]]; then
+    uv run tests/check_metrics.py $JSON_METRICS \
+        'mean(data["train/token_mult_prob_error"]) < 1.1' \
+        'data["train/token_mult_prob_error"]["10"] < 1.1'
+fi
diff --git a/tests/test_suites/llm/performance/grpo-qwen3-235b-32n4g-async-1off.sh b/tests/test_suites/llm/performance/grpo-qwen3-235b-32n4g-async-1off.sh
new file mode 100755
index 0000000000..f7dac553af
--- /dev/null
+++ b/tests/test_suites/llm/performance/grpo-qwen3-235b-32n4g-async-1off.sh
@@ -0,0 +1,40 @@
+#!/bin/bash
+SCRIPT_DIR=$( cd -- "$( dirname -- "${BASH_SOURCE[0]}" )" &> /dev/null && pwd)
+source $SCRIPT_DIR/common.env
+# disable NVLS to avoid OOM issue
+export NCCL_NVLS_ENABLE=0
+
+# ===== BEGIN CONFIG =====
+NUM_NODES=32
+STEPS_PER_RUN=10
+MAX_STEPS=10
+NUM_RUNS=$(( (MAX_STEPS + STEPS_PER_RUN - 1) / STEPS_PER_RUN ))  # Round up
+NUM_MINUTES=100
+# ===== END CONFIG =====
+
+exit_if_max_steps_reached
+
+# Run the experiment
+cd $PROJECT_ROOT
+uv run examples/run_grpo_math.py \
+    --config $CONFIG_PATH \
+    grpo.max_num_steps=$MAX_STEPS \
+    logger.log_dir=$LOG_DIR \
+    logger.wandb_enabled=True \
+    logger.wandb.project=nemo-rl \
+    logger.wandb.name=$EXP_NAME \
+    logger.monitor_gpus=True \
+    checkpointing.enabled=True \
+    checkpointing.checkpoint_dir=$CKPT_DIR \
+    $@ \
+    2>&1 | tee $RUN_LOG
+
+# Convert tensorboard logs to json
+uv run tests/json_dump_tb_logs.py $LOG_DIR --output_path $JSON_METRICS
+
+# Only run metrics if the target step is reached
+if [[ $(jq 'to_entries | .[] | select(.key == "train/loss") | .value | keys | map(tonumber) | max' $JSON_METRICS) -ge $MAX_STEPS ]]; then
+    uv run tests/check_metrics.py $JSON_METRICS \
+        'mean(data["train/token_mult_prob_error"]) < 1.1' \
+        'data["train/token_mult_prob_error"]["10"] < 1.1'
+fi
diff --git a/tests/test_suites/llm/performance/grpo-qwen3-30ba3b-4n4g.sh b/tests/test_suites/llm/performance/grpo-qwen3-30ba3b-4n4g.sh
new file mode 100755
index 0000000000..2a56609ffd
--- /dev/null
+++ b/tests/test_suites/llm/performance/grpo-qwen3-30ba3b-4n4g.sh
@@ -0,0 +1,40 @@
+#!/bin/bash
+SCRIPT_DIR=$( cd -- "$( dirname -- "${BASH_SOURCE[0]}" )" &> /dev/null && pwd)
+source $SCRIPT_DIR/common.env
+
+# ===== BEGIN CONFIG =====
+NUM_NODES=4
+STEPS_PER_RUN=10
+MAX_STEPS=10
+NUM_RUNS=$(( (MAX_STEPS + STEPS_PER_RUN - 1) / STEPS_PER_RUN ))  # Round up
+NUM_MINUTES=100
+# ===== END CONFIG =====
+
+exit_if_max_steps_reached
+
+# Run the experiment
+cd $PROJECT_ROOT
+uv run examples/run_grpo_math.py \
+    --config $CONFIG_PATH \
+    grpo.max_num_steps=$MAX_STEPS \
+    logger.log_dir=$LOG_DIR \
+    logger.wandb_enabled=True \
+    logger.wandb.project=nemo-rl \
+    logger.wandb.name=$EXP_NAME \
+    logger.monitor_gpus=True \
+    logger.tensorboard_enabled=True \
+    checkpointing.enabled=True \
+    checkpointing.checkpoint_dir=$CKPT_DIR \
+    $@ \
+    2>&1 | tee $RUN_LOG
+
+# Convert tensorboard logs to json
+uv run tests/json_dump_tb_logs.py $LOG_DIR --output_path $JSON_METRICS
+
+# Only run metrics if the target step is reached
+if [[ $(jq 'to_entries | .[] | select(.key == "train/loss") | .value | keys | map(tonumber) | max' $JSON_METRICS) -ge $MAX_STEPS ]]; then
+    uv run tests/check_metrics.py $JSON_METRICS \
+        'mean(data["train/token_mult_prob_error"]) < 1.1' \
+        'data["train/token_mult_prob_error"]["10"] < 1.1'
+fi
+
diff --git a/tests/test_suites/llm/performance/grpo-qwen3-30ba3b-8n4g-async-1off.sh b/tests/test_suites/llm/performance/grpo-qwen3-30ba3b-8n4g-async-1off.sh
new file mode 100755
index 0000000000..8350d128e8
--- /dev/null
+++ b/tests/test_suites/llm/performance/grpo-qwen3-30ba3b-8n4g-async-1off.sh
@@ -0,0 +1,40 @@
+#!/bin/bash
+SCRIPT_DIR=$( cd -- "$( dirname -- "${BASH_SOURCE[0]}" )" &> /dev/null && pwd)
+source $SCRIPT_DIR/common.env
+
+# ===== BEGIN CONFIG =====
+NUM_NODES=8
+STEPS_PER_RUN=10
+MAX_STEPS=10
+NUM_RUNS=$(( (MAX_STEPS + STEPS_PER_RUN - 1) / STEPS_PER_RUN ))  # Round up
+NUM_MINUTES=100
+# ===== END CONFIG =====
+
+exit_if_max_steps_reached
+
+# Run the experiment
+cd $PROJECT_ROOT
+uv run examples/run_grpo_math.py \
+    --config $CONFIG_PATH \
+    grpo.max_num_steps=$MAX_STEPS \
+    logger.log_dir=$LOG_DIR \
+    logger.wandb_enabled=True \
+    logger.wandb.project=nemo-rl \
+    logger.wandb.name=$EXP_NAME \
+    logger.monitor_gpus=True \
+    logger.tensorboard_enabled=True \
+    checkpointing.enabled=True \
+    checkpointing.checkpoint_dir=$CKPT_DIR \
+    $@ \
+    2>&1 | tee $RUN_LOG
+
+# Convert tensorboard logs to json
+uv run tests/json_dump_tb_logs.py $LOG_DIR --output_path $JSON_METRICS
+
+# Only run metrics if the target step is reached
+if [[ $(jq 'to_entries | .[] | select(.key == "train/loss") | .value | keys | map(tonumber) | max' $JSON_METRICS) -ge $MAX_STEPS ]]; then
+    uv run tests/check_metrics.py $JSON_METRICS \
+        'mean(data["train/token_mult_prob_error"]) < 1.1' \
+        'data["train/token_mult_prob_error"]["10"] < 1.1'
+fi
+
diff --git a/tests/test_suites/llm/performance/grpo-qwen3-32b-4n4g.sh b/tests/test_suites/llm/performance/grpo-qwen3-32b-4n4g.sh
new file mode 100755
index 0000000000..2a56609ffd
--- /dev/null
+++ b/tests/test_suites/llm/performance/grpo-qwen3-32b-4n4g.sh
@@ -0,0 +1,40 @@
+#!/bin/bash
+SCRIPT_DIR=$( cd -- "$( dirname -- "${BASH_SOURCE[0]}" )" &> /dev/null && pwd)
+source $SCRIPT_DIR/common.env
+
+# ===== BEGIN CONFIG =====
+NUM_NODES=4
+STEPS_PER_RUN=10
+MAX_STEPS=10
+NUM_RUNS=$(( (MAX_STEPS + STEPS_PER_RUN - 1) / STEPS_PER_RUN ))  # Round up
+NUM_MINUTES=100
+# ===== END CONFIG =====
+
+exit_if_max_steps_reached
+
+# Run the experiment
+cd $PROJECT_ROOT
+uv run examples/run_grpo_math.py \
+    --config $CONFIG_PATH \
+    grpo.max_num_steps=$MAX_STEPS \
+    logger.log_dir=$LOG_DIR \
+    logger.wandb_enabled=True \
+    logger.wandb.project=nemo-rl \
+    logger.wandb.name=$EXP_NAME \
+    logger.monitor_gpus=True \
+    logger.tensorboard_enabled=True \
+    checkpointing.enabled=True \
+    checkpointing.checkpoint_dir=$CKPT_DIR \
+    $@ \
+    2>&1 | tee $RUN_LOG
+
+# Convert tensorboard logs to json
+uv run tests/json_dump_tb_logs.py $LOG_DIR --output_path $JSON_METRICS
+
+# Only run metrics if the target step is reached
+if [[ $(jq 'to_entries | .[] | select(.key == "train/loss") | .value | keys | map(tonumber) | max' $JSON_METRICS) -ge $MAX_STEPS ]]; then
+    uv run tests/check_metrics.py $JSON_METRICS \
+        'mean(data["train/token_mult_prob_error"]) < 1.1' \
+        'data["train/token_mult_prob_error"]["10"] < 1.1'
+fi
+
diff --git a/tests/test_suites/llm/performance/grpo-qwen3-32b-8n4g-async-1off.sh b/tests/test_suites/llm/performance/grpo-qwen3-32b-8n4g-async-1off.sh
new file mode 100755
index 0000000000..35d58c98f7
--- /dev/null
+++ b/tests/test_suites/llm/performance/grpo-qwen3-32b-8n4g-async-1off.sh
@@ -0,0 +1,39 @@
+#!/bin/bash
+SCRIPT_DIR=$( cd -- "$( dirname -- "${BASH_SOURCE[0]}" )" &> /dev/null && pwd)
+source $SCRIPT_DIR/common.env
+
+# ===== BEGIN CONFIG =====
+NUM_NODES=8
+STEPS_PER_RUN=10
+MAX_STEPS=10
+NUM_RUNS=$(( (MAX_STEPS + STEPS_PER_RUN - 1) / STEPS_PER_RUN ))  # Round up
+NUM_MINUTES=100
+# ===== END CONFIG =====
+
+exit_if_max_steps_reached
+
+# Run the experiment
+cd $PROJECT_ROOT
+uv run examples/run_grpo_math.py \
+    --config $CONFIG_PATH \
+    grpo.max_num_steps=$MAX_STEPS \
+    logger.log_dir=$LOG_DIR \
+    logger.wandb_enabled=True \
+    logger.wandb.project=nemo-rl \
+    logger.wandb.name=$EXP_NAME \
+    logger.monitor_gpus=True \
+    logger.tensorboard_enabled=True \
+    checkpointing.enabled=True \
+    checkpointing.checkpoint_dir=$CKPT_DIR \
+    $@ \
+    2>&1 | tee $RUN_LOG
+
+# Convert tensorboard logs to json
+uv run tests/json_dump_tb_logs.py $LOG_DIR --output_path $JSON_METRICS
+
+# Only run metrics if the target step is reached
+if [[ $(jq 'to_entries | .[] | select(.key == "train/loss") | .value | keys | map(tonumber) | max' $JSON_METRICS) -ge $MAX_STEPS ]]; then
+    uv run tests/check_metrics.py $JSON_METRICS \
+        'mean(data["train/token_mult_prob_error"]) < 1.1' \
+        'data["train/token_mult_prob_error"]["10"] < 1.1'
+fi
diff --git a/tests/test_suites/performance_gb200.txt b/tests/test_suites/performance_gb200.txt
new file mode 100644
index 0000000000..d958386001
--- /dev/null
+++ b/tests/test_suites/performance_gb200.txt
@@ -0,0 +1,19 @@
+########
+# GRPO #
+########
+
+# GB200 BF16
+
+## SYNC
+tests/test_suites/llm/performance/grpo-llama3.1-8b-instruct-2n4g.sh
+tests/test_suites/llm/performance/grpo-qwen3-32b-4n4g.sh
+tests/test_suites/llm/performance/grpo-qwen3-30ba3b-4n4g.sh
+tests/test_suites/llm/performance/grpo-deepseek-v3-32n4g.sh
+tests/test_suites/llm/performance/grpo-qwen3-235b-16n4g.sh
+
+## ASYNC 1-off
+tests/test_suites/llm/performance/grpo-llama3.1-8b-instruct-2n4g-async-1off.sh
+tests/test_suites/llm/performance/grpo-qwen3-32b-8n4g-async-1off.sh
+tests/test_suites/llm/performance/grpo-qwen3-30ba3b-8n4g-async-1off.sh
+tests/test_suites/llm/performance/grpo-deepseek-v3-64n4g-async-1off.sh
+tests/test_suites/llm/performance/grpo-qwen3-235b-32n4g-async-1off.sh
diff --git a/tests/test_suites/performance.txt b/tests/test_suites/performance_h100.txt
similarity index 75%
rename from tests/test_suites/performance.txt
rename to tests/test_suites/performance_h100.txt
index bf714b0e74..9e3eb208ce 100644
--- a/tests/test_suites/performance.txt
+++ b/tests/test_suites/performance_h100.txt
@@ -2,16 +2,28 @@
 # GRPO #
 ########
 
+# H100 BF16
+
+## SYNC
 tests/test_suites/llm/performance/grpo-llama3.1-8b-instruct-2n8g.sh
 tests/test_suites/llm/performance/grpo-qwen3-30ba3b-4n8g.sh
 tests/test_suites/llm/performance/grpo-deepseek-v3-32n8g.sh
 tests/test_suites/llm/performance/grpo-qwen3-32b-4n8g.sh
 tests/test_suites/llm/performance/grpo-qwen3-235b-16n8g.sh
 
+## ASYNC 1-off
 tests/test_suites/llm/performance/grpo-deepseek-v3-64n8g-async-1off.sh
 tests/test_suites/llm/performance/grpo-llama3.1-8b-instruct-2n8g-async-1off.sh
 tests/test_suites/llm/performance/grpo-qwen3-32b-8n8g-async-1off.sh
 tests/test_suites/llm/performance/grpo-qwen3-235b-32n8g-async-1off.sh
 tests/test_suites/llm/performance/grpo-qwen3-30ba3b-4n8g-async-1off.sh
 
-tests/test_suites/llm/performance/grpo-qwen3-30ba3b-24n8g-async-8off.sh
\ No newline at end of file
+## ASYNC many-off
+tests/test_suites/llm/performance/grpo-qwen3-30ba3b-24n8g-async-8off.sh
+
+# H100 FP8
+
+## ASYNC 1-off
+tests/test_suites/llm/performance/grpo-deepseek-v3-64n8g-fp8-async-1off.sh
+tests/test_suites/llm/performance/grpo-llama3.1-8b-instruct-2n8g-fp8-async-1off.sh
+
diff --git a/tests/unit/test_recipes_and_test_suites.py b/tests/unit/test_recipes_and_test_suites.py
index 435742ec4a..ade6d49d87 100644
--- a/tests/unit/test_recipes_and_test_suites.py
+++ b/tests/unit/test_recipes_and_test_suites.py
@@ -28,7 +28,10 @@
 
 nightly_test_suite_path = os.path.join(test_suites_dir, "nightly.txt")
 release_test_suite_path = os.path.join(test_suites_dir, "release.txt")
-performance_test_suite_path = os.path.join(test_suites_dir, "performance.txt")
+h100_performance_test_suite_path = os.path.join(test_suites_dir, "performance_h100.txt")
+gb200_performance_test_suite_path = os.path.join(
+    test_suites_dir, "performance_gb200.txt"
+)
 
 # Relative to project root
 ALGO_MAPPING_TO_BASE_YAML = {
@@ -72,7 +75,12 @@ def release_test_suite():
 @pytest.fixture
 def performance_test_suite():
     performance_suite = []
-    with open(performance_test_suite_path, "r") as f:
+    with open(h100_performance_test_suite_path, "r") as f:
+        for line in f:
+            line = line.strip()
+            if line and not line.startswith("#"):
+                performance_suite.append(line)
+    with open(gb200_performance_test_suite_path, "r") as f:
         for line in f:
             line = line.strip()
             if line and not line.startswith("#"):
@@ -104,12 +112,14 @@ def all_recipe_yaml_rel_paths():
     [
         nightly_test_suite_path,
         release_test_suite_path,
-        performance_test_suite_path,
+        h100_performance_test_suite_path,
+        gb200_performance_test_suite_path,
     ],
     ids=[
         "nightly_test_suite",
         "release_test_suite",
-        "performance_test_suite",
+        "h100_performance_test_suite",
+        "gb200_performance_test_suite",
     ],
 )
 def test_test_suites_exist(test_suite_path):