Update 2x H100 superbench config (#223)

Author: yangpanMS

.pipelines/azure-pipelines-linux.yml

@@ -16,7 +16,7 @@ resources:
     options: --entrypoint=""
 
 variables:
-  VcVersion : 1.13.2
+  VcVersion : 1.13.5
   ROOT: $(Build.SourcesDirectory)
   CDP_DEFINITION_BUILD_COUNT: $[counter('', 0)] # needed for onebranch.pipeline.version task https://aka.ms/obpipelines/versioning
   ENABLE_PRS_DELAYSIGN: 1

.pipelines/azure-pipelines.yml

@@ -18,7 +18,7 @@ pool:
   vmImage: windows-latest
 
 variables:
-  VcVersion : 1.13.4
+  VcVersion : 1.13.5
   ROOT: $(Build.SourcesDirectory)
   CDP_DEFINITION_BUILD_COUNT: $[counter('', 0)] # needed for onebranch.pipeline.version task https://aka.ms/obpipelines/versioning
   ENABLE_PRS_DELAYSIGN: 1

src/VirtualClient/VirtualClient.Actions/SuperBenchmark/2xH100.yaml

@@ -5,7 +5,7 @@ superbench:
   - kernel-launch
   - gemm-flops
   - cublaslt-gemm
-  - cublas-function
+  # cublas-function
   - matmul
   - gpu-burn
   # microbenchmark - communication
@@ -67,7 +67,7 @@ superbench:
       timeout: *default_timeout
     gemm-flops:
       <<: *default_local_mode
-      timeout: 1800
+      timeout: 3600
     cublaslt-gemm:
       <<: *default_local_mode
       timeout: *default_timeout
@@ -83,7 +83,7 @@ superbench:
         - 12608,3072,1024
     cublas-function:
       <<: *default_local_mode
-      timeout: *default_timeout
+      timeout: 1200
     matmul:
       <<: *default_local_mode
       timeout: *default_timeout
@@ -167,7 +167,7 @@ superbench:
         seq_len: 224
     model-benchmarks:bert:
       <<: *default_pytorch_mode
-      timeout: 2400
+      timeout: 4800
       models:
       - bert-base
       - bert-large
@@ -227,4 +227,4 @@ superbench:
         <<: *model_ddp_param
         seq_len: 224
         duration: 1800
-        num_steps: -100
+        num_steps: -100