diff --git a/src/sagemaker/image_uri_config/instance_gpu_info.json b/src/sagemaker/image_uri_config/instance_gpu_info.json index 9fc005bc47..e64a9bcf88 100644 --- a/src/sagemaker/image_uri_config/instance_gpu_info.json +++ b/src/sagemaker/image_uri_config/instance_gpu_info.json @@ -23,7 +23,7 @@ "ml.g5.16xlarge": {"Count": 1, "TotalGpuMemoryInMiB": 24576}, "ml.g5.12xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, "ml.g5.24xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, - "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 196608} + "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 183104} }, "ap-east-1": { "ml.p5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 655360}, @@ -49,7 +49,7 @@ "ml.g5.16xlarge": {"Count": 1, "TotalGpuMemoryInMiB": 24576}, "ml.g5.12xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, "ml.g5.24xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, - "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 196608} + "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 183104} }, "ap-northeast-1": { "ml.p5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 655360}, @@ -75,7 +75,7 @@ "ml.g5.16xlarge": {"Count": 1, "TotalGpuMemoryInMiB": 24576}, "ml.g5.12xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, "ml.g5.24xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, - "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 196608} + "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 183104} }, "ap-northeast-2": { "ml.p5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 655360}, @@ -101,7 +101,7 @@ "ml.g5.16xlarge": {"Count": 1, "TotalGpuMemoryInMiB": 24576}, "ml.g5.12xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, "ml.g5.24xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, - "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 196608} + "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 183104} }, "ap-northeast-3": { "ml.p5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 655360}, @@ -127,7 +127,7 @@ "ml.g5.16xlarge": {"Count": 1, "TotalGpuMemoryInMiB": 24576}, "ml.g5.12xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, "ml.g5.24xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, - "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 196608} + "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 183104} }, "ap-south-1": { "ml.p5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 655360}, @@ -153,7 +153,7 @@ "ml.g5.16xlarge": {"Count": 1, "TotalGpuMemoryInMiB": 24576}, "ml.g5.12xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, "ml.g5.24xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, - "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 196608} + "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 183104} }, "ap-southeast-1": { "ml.p5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 655360}, @@ -179,7 +179,7 @@ "ml.g5.16xlarge": {"Count": 1, "TotalGpuMemoryInMiB": 24576}, "ml.g5.12xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, "ml.g5.24xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, - "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 196608} + "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 183104} }, "ap-southeast-2": { "ml.p5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 655360}, @@ -205,7 +205,7 @@ "ml.g5.16xlarge": {"Count": 1, "TotalGpuMemoryInMiB": 24576}, "ml.g5.12xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, "ml.g5.24xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, - "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 196608} + "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 183104} }, "ap-southeast-3": { "ml.p5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 655360}, @@ -231,7 +231,7 @@ "ml.g5.16xlarge": {"Count": 1, "TotalGpuMemoryInMiB": 24576}, "ml.g5.12xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, "ml.g5.24xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, - "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 196608} + "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 183104} }, "ca-central-1": { "ml.p5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 655360}, @@ -257,7 +257,7 @@ "ml.g5.16xlarge": {"Count": 1, "TotalGpuMemoryInMiB": 24576}, "ml.g5.12xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, "ml.g5.24xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, - "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 196608} + "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 183104} }, "cn-north-1": { "ml.p5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 655360}, @@ -283,7 +283,7 @@ "ml.g5.16xlarge": {"Count": 1, "TotalGpuMemoryInMiB": 24576}, "ml.g5.12xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, "ml.g5.24xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, - "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 196608} + "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 183104} }, "cn-northwest-1": { "ml.p5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 655360}, @@ -309,7 +309,7 @@ "ml.g5.16xlarge": {"Count": 1, "TotalGpuMemoryInMiB": 24576}, "ml.g5.12xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, "ml.g5.24xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, - "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 196608} + "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 183104} }, "eu-central-1": { "ml.p5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 655360}, @@ -335,7 +335,7 @@ "ml.g5.16xlarge": {"Count": 1, "TotalGpuMemoryInMiB": 24576}, "ml.g5.12xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, "ml.g5.24xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, - "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 196608} + "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 183104} }, "eu-central-2": { "ml.p5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 655360}, @@ -361,7 +361,7 @@ "ml.g5.16xlarge": {"Count": 1, "TotalGpuMemoryInMiB": 24576}, "ml.g5.12xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, "ml.g5.24xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, - "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 196608} + "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 183104} }, "eu-north-1": { "ml.p5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 655360}, @@ -387,7 +387,7 @@ "ml.g5.16xlarge": {"Count": 1, "TotalGpuMemoryInMiB": 24576}, "ml.g5.12xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, "ml.g5.24xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, - "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 196608} + "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 183104} }, "eu-south-1": { "ml.p5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 655360}, @@ -413,7 +413,7 @@ "ml.g5.16xlarge": {"Count": 1, "TotalGpuMemoryInMiB": 24576}, "ml.g5.12xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, "ml.g5.24xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, - "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 196608} + "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 183104} }, "eu-south-2": { "ml.p5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 655360}, @@ -439,7 +439,7 @@ "ml.g5.16xlarge": {"Count": 1, "TotalGpuMemoryInMiB": 24576}, "ml.g5.12xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, "ml.g5.24xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, - "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 196608} + "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 183104} }, "eu-west-1": { "ml.p5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 655360}, @@ -465,7 +465,7 @@ "ml.g5.16xlarge": {"Count": 1, "TotalGpuMemoryInMiB": 24576}, "ml.g5.12xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, "ml.g5.24xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, - "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 196608} + "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 183104} }, "eu-west-2": { "ml.p5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 655360}, @@ -491,7 +491,7 @@ "ml.g5.16xlarge": {"Count": 1, "TotalGpuMemoryInMiB": 24576}, "ml.g5.12xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, "ml.g5.24xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, - "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 196608} + "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 183104} }, "eu-west-3": { "ml.p5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 655360}, @@ -517,7 +517,7 @@ "ml.g5.16xlarge": {"Count": 1, "TotalGpuMemoryInMiB": 24576}, "ml.g5.12xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, "ml.g5.24xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, - "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 196608} + "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 183104} }, "il-central-1": { "ml.p5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 655360}, @@ -543,7 +543,7 @@ "ml.g5.16xlarge": {"Count": 1, "TotalGpuMemoryInMiB": 24576}, "ml.g5.12xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, "ml.g5.24xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, - "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 196608} + "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 183104} }, "me-central-1": { "ml.p5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 655360}, @@ -569,7 +569,7 @@ "ml.g5.16xlarge": {"Count": 1, "TotalGpuMemoryInMiB": 24576}, "ml.g5.12xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, "ml.g5.24xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, - "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 196608} + "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 183104} }, "me-south-1": { "ml.p5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 655360}, @@ -595,7 +595,7 @@ "ml.g5.16xlarge": {"Count": 1, "TotalGpuMemoryInMiB": 24576}, "ml.g5.12xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, "ml.g5.24xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, - "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 196608} + "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 183104} }, "sa-east-1": { "ml.p5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 655360}, @@ -621,7 +621,7 @@ "ml.g5.16xlarge": {"Count": 1, "TotalGpuMemoryInMiB": 24576}, "ml.g5.12xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, "ml.g5.24xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, - "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 196608} + "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 183104} }, "us-east-1": { "ml.p5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 655360}, @@ -647,7 +647,7 @@ "ml.g5.16xlarge": {"Count": 1, "TotalGpuMemoryInMiB": 24576}, "ml.g5.12xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, "ml.g5.24xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, - "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 196608} + "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 183104} }, "us-east-2": { "ml.p5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 655360}, @@ -673,7 +673,7 @@ "ml.g5.16xlarge": {"Count": 1, "TotalGpuMemoryInMiB": 24576}, "ml.g5.12xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, "ml.g5.24xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, - "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 196608} + "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 183104} }, "us-gov-east-1": { "ml.p5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 655360}, @@ -699,7 +699,7 @@ "ml.g5.16xlarge": {"Count": 1, "TotalGpuMemoryInMiB": 24576}, "ml.g5.12xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, "ml.g5.24xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, - "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 196608} + "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 183104} }, "us-gov-west-1": { "ml.p5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 655360}, @@ -725,7 +725,7 @@ "ml.g5.16xlarge": {"Count": 1, "TotalGpuMemoryInMiB": 24576}, "ml.g5.12xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, "ml.g5.24xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, - "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 196608} + "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 183104} }, "us-west-1": { "ml.p5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 655360}, @@ -751,7 +751,7 @@ "ml.g5.16xlarge": {"Count": 1, "TotalGpuMemoryInMiB": 24576}, "ml.g5.12xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, "ml.g5.24xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, - "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 196608} + "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 183104} }, "us-west-2": { "ml.p5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 655360}, @@ -777,6 +777,6 @@ "ml.g5.16xlarge": {"Count": 1, "TotalGpuMemoryInMiB": 24576}, "ml.g5.12xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, "ml.g5.24xlarge": {"Count": 4, "TotalGpuMemoryInMiB": 98304}, - "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 196608} + "ml.g5.48xlarge": {"Count": 8, "TotalGpuMemoryInMiB": 183104} } } \ No newline at end of file diff --git a/tests/integ/sagemaker/serve/utils/test_hardware_detector.py b/tests/integ/sagemaker/serve/utils/test_hardware_detector.py index 9102927c55..bab26a25d1 100644 --- a/tests/integ/sagemaker/serve/utils/test_hardware_detector.py +++ b/tests/integ/sagemaker/serve/utils/test_hardware_detector.py @@ -19,7 +19,7 @@ REGION = "us-west-2" VALID_INSTANCE_TYPE = "ml.g5.48xlarge" INVALID_INSTANCE_TYPE = "fl.c5.57xxlarge" -EXPECTED_INSTANCE_GPU_INFO = (8, 196608) +EXPECTED_INSTANCE_GPU_INFO = (8, 183104) def test_get_gpu_info_success(sagemaker_session): diff --git a/tests/unit/sagemaker/serve/utils/test_hardware_detector.py b/tests/unit/sagemaker/serve/utils/test_hardware_detector.py index d383f95809..58839bfc50 100644 --- a/tests/unit/sagemaker/serve/utils/test_hardware_detector.py +++ b/tests/unit/sagemaker/serve/utils/test_hardware_detector.py @@ -21,7 +21,7 @@ REGION = "us-west-2" VALID_INSTANCE_TYPE = "ml.g5.48xlarge" INVALID_INSTANCE_TYPE = "fl.c5.57xxlarge" -EXPECTED_INSTANCE_GPU_INFO = (8, 196608) +EXPECTED_INSTANCE_GPU_INFO = (8, 183104) MIB_CONVERSION_FACTOR = 0.00000095367431640625 MEMORY_BUFFER_MULTIPLIER = 1.2 # 20% buffer @@ -39,7 +39,7 @@ def test_get_gpu_info_success(sagemaker_session, boto_session): "MemoryInfo": {"SizeInMiB": 24576}, } ], - "TotalGpuMemoryInMiB": 196608, + "TotalGpuMemoryInMiB": 183104, }, } ]