Support mlperf inference v6.0

script/app-mlperf-inference/meta.yaml

@@ -467,6 +467,35 @@ variations:
         docker:
           base_image: nvcr.io/nvidia/mlperf/mlperf-inference:mlpinf-v5.0-cuda12.8-pytorch25.01-ubuntu24.04-aarch64-Grace-release
 
+  nvidia-original,r6.0-dev_default:
+    env:
+      MLC_NVIDIA_MITTEN_FROM_SRC: 'yes'
+    docker:
+      os_version: "24.04"
+      user: 'ubuntu'
+      build_deps:
+        - tags: detect,os
+      image_name: mlperf-inference-nvidia-v5.0-common
+      build_env:
+        ENV: release
+    deps:
+    - names:
+      - numpy
+      tags: get,generic-python-lib,_package.numpy
+      version_max: "1.26.999"
+      version_max_usable: "1.26.4"
+    update_meta_if_env:
+      - enable_if_env:
+          MLC_HOST_PLATFORM_FLAVOR:
+            - x86_64
+        docker:
+          base_image: nvcr.io/nvidia/mlperf/mlperf-inference:mlpinf-v5.0-cuda12.8-pytorch25.01-ubuntu24.04-x86_64-release
+      - skip_if_env:
+          MLC_HOST_PLATFORM_FLAVOR:
+            - x86_64
+        docker:
+          base_image: nvcr.io/nvidia/mlperf/mlperf-inference:mlpinf-v5.0-cuda12.8-pytorch25.01-ubuntu24.04-aarch64-Grace-release
+
   nvidia-original,r5.1_default:
     env:
       MLC_NVIDIA_MITTEN_FROM_SRC: 'yes'
@@ -2377,6 +2406,13 @@ variations:
     group:
       reproducibility
     add_deps_recursive:
+      nvidia-inference-common-code:
+        tags: _mlcommons,_v6.0-dev
+      nvidia-inference-server:
+        version: r5.0
+        tags: _mlcommons
+      nvidia-harness:   
+        tags: _v5.0
       pycuda:
         version_min: "2024.1"
     default_env:

script/get-cuda-devices/print_cuda_devices.cu

@@ -54,7 +54,11 @@ int main(int argc, char *argv[])
      printf("CUDA runtime version: %d.%d\n", rtver/1000, (rtver%100)/10);
 
      printf("Global memory: %llu\n", (unsigned long long) features.totalGlobalMem);
-     printf("Max clock rate: %f MHz\n", features.clockRate * 0.001);
+
+     int clock;
+     cudaDeviceGetAttribute(&clock, cudaDevAttrClockRate, id);
+
+     printf("Max clock rate: %f MHz\n", 0.001 * clock);
 
      printf("Total amount of shared memory per block: %lu\n", features.sharedMemPerBlock);
      printf("Total number of registers available per block: %d\n", features.regsPerBlock);

script/get-mlperf-inference-nvidia-common-code/meta.yaml

@@ -57,6 +57,10 @@ variations:
     add_deps_recursive:
       mlperf-inference-results:
         tags: _code-only,_v5.1
+  v6.0-dev:
+    add_deps_recursive:
+      mlperf-inference-results:
+        tags: _code-only-for-v6.0,_v6.0-dev
 versions:
   r2.1:
     add_deps_recursive:

script/get-mlperf-inference-results/meta.yaml

@@ -47,6 +47,11 @@ variations:
       inference-results-repo:
         tags: _branch.mlc-code-only-for-v5.1
     group: repo-branch
+  code-only-for-v6.0:
+    adr:
+      inference-results-repo:
+        tags: _branch.mlc-code-only-for-v6.0
+    group: repo-branch
   ctuning:
     env:
       GITHUB_REPO_OWNER: ctuning
@@ -93,6 +98,12 @@ variations:
       MLC_GIT_URL: https://github.com/<<<GITHUB_REPO_OWNER>>>/inference_results_v5.1.git
       MLC_MLPERF_INFERENCE_RESULTS_VERSION_NAME: v5.1
       MLC_VERSION: "v5.1"
+  v6.0-dev:
+    group: version
+    env:
+      MLC_GIT_URL: https://github.com/<<<GITHUB_REPO_OWNER>>>/inference_results_v5.0.git
+      MLC_MLPERF_INFERENCE_RESULTS_VERSION_NAME: v6.0-dev
+      MLC_VERSION: "v5.0"
 versions:
   v2.1:
     env:

script/install-cuda-prebuilt/meta.yaml

@@ -97,3 +97,6 @@ versions:
   12.9.1:
     env:
       MLC_CUDA_LINUX_FILENAME: cuda_12.9.1_575.57.08_linux.run
+  13.1.1:
+    env:
+      MLC_CUDA_LINUX_FILENAME: cuda_13.1.1_590.48.01_linux.run

script/run-mlperf-inference-app/meta.yaml

@@ -103,6 +103,7 @@ input_mapping:
   skip_truncation: MLC_SKIP_TRUNCATE_ACCURACY
   submission_dir: MLC_MLPERF_INFERENCE_SUBMISSION_DIR
   submitter: MLC_MLPERF_SUBMITTER
+  submitter_id: MLC_MLPERF_SUBMITTER_ID
   sut_servers: MLC_NETWORK_LOADGEN_SUT_SERVERS
   sw_notes_extra: MLC_MLPERF_SUT_SW_NOTES_EXTRA
   system_type: MLC_MLPERF_SUBMISSION_SYSTEM_TYPE