Merge pull request #98 from QuanMPhm/97/h100_billing

Allow billing of H100 GPUs

Author: naved001

openshift_metrics/invoice.py

@@ -9,6 +9,7 @@
 GPU_A100 = "NVIDIA-A100-40GB"
 GPU_A100_SXM4 = "NVIDIA-A100-SXM4-40GB"
 GPU_V100 = "Tesla-V100-PCIE-32GB"
+GPU_H100 = "NVIDIA-H100-80GB-HBM3"
 GPU_UNKNOWN_TYPE = "GPU_UNKNOWN_TYPE"
 
 # GPU Resource - MIG Geometries
@@ -23,6 +24,7 @@
 SU_A100_GPU = "OpenShift GPUA100"
 SU_A100_SXM4_GPU = "OpenShift GPUA100SXM4"
 SU_V100_GPU = "OpenShift GPUV100"
+SU_H100_GPU = "OpenShift GPUH100"
 SU_UNKNOWN_GPU = "OpenShift Unknown GPU"
 SU_UNKNOWN_MIG_GPU = "OpenShift Unknown MIG GPU"
 SU_UNKNOWN = "Openshift Unknown"
@@ -65,6 +67,7 @@ def get_service_unit(self) -> ServiceUnit:
             GPU_A100: SU_A100_GPU,
             GPU_A100_SXM4: SU_A100_SXM4_GPU,
             GPU_V100: SU_V100_GPU,
+            GPU_H100: SU_H100_GPU,
         }
 
         A100_SXM4_MIG = {
@@ -79,6 +82,7 @@ def get_service_unit(self) -> ServiceUnit:
             SU_A100_GPU: {"gpu": 1, "cpu": 24, "ram": 74},
             SU_A100_SXM4_GPU: {"gpu": 1, "cpu": 31, "ram": 240},
             SU_V100_GPU: {"gpu": 1, "cpu": 48, "ram": 192},
+            SU_H100_GPU: {"gpu": 1, "cpu": 124, "ram": 360},
             SU_UNKNOWN_GPU: {"gpu": 1, "cpu": 8, "ram": 64},
             SU_UNKNOWN_MIG_GPU: {"gpu": 1, "cpu": 8, "ram": 64},
             SU_UNKNOWN: {"gpu": -1, "cpu": 1, "ram": 1},
@@ -179,6 +183,7 @@ class Rates:
     gpu_a100: Decimal
     gpu_a100sxm4: Decimal
     gpu_v100: Decimal
+    gpu_h100: Decimal
 
 
 @dataclass
@@ -201,6 +206,7 @@ class ProjectInvoce:
             SU_A100_GPU: 0,
             SU_A100_SXM4_GPU: 0,
             SU_V100_GPU: 0,
+            SU_H100_GPU: 0,
             SU_UNKNOWN_GPU: 0,
             SU_UNKNOWN_MIG_GPU: 0,
             SU_UNKNOWN: 0,
@@ -222,6 +228,8 @@ def get_rate(self, su_type) -> Decimal:
             return self.rates.gpu_a100sxm4
         if su_type == SU_V100_GPU:
             return self.rates.gpu_v100
+        if su_type == SU_H100_GPU:
+            return self.rates.gpu_h100
         return Decimal(0)
 
     def generate_invoice_rows(self, report_month) -> List[str]:

openshift_metrics/merge.py

@@ -69,6 +69,7 @@ def main():
     parser.add_argument("--rate-gpu-v100-su", type=Decimal)
     parser.add_argument("--rate-gpu-a100sxm4-su", type=Decimal)
     parser.add_argument("--rate-gpu-a100-su", type=Decimal)
+    parser.add_argument("--rate-gpu-h100-su", type=Decimal)
 
     args = parser.parse_args()
     files = args.files
@@ -115,13 +116,15 @@ def main():
             gpu_a100=Decimal(nerc_data.get_value_at("GPUA100 SU Rate", report_month)),
             gpu_a100sxm4=Decimal(nerc_data.get_value_at("GPUA100SXM4 SU Rate", report_month)),
             gpu_v100=Decimal(nerc_data.get_value_at("GPUV100 SU Rate", report_month)),
+            gpu_h100=Decimal(nerc_data.get_value_at("GPUH100 SU Rate", report_month)),
         )
     else:
         rates = invoice.Rates(
             cpu=Decimal(args.rate_cpu_su),
             gpu_a100=Decimal(args.rate_gpu_a100_su),
             gpu_a100sxm4=Decimal(args.rate_gpu_a100sxm4_su),
-            gpu_v100=Decimal(args.rate_gpu_v100_su)
+            gpu_v100=Decimal(args.rate_gpu_v100_su),
+            gpu_h100=Decimal(args.rate_gpu_h100_su),
         )
 
     if args.invoice_file:

openshift_metrics/tests/test_utils.py

@@ -23,7 +23,8 @@
     cpu = Decimal("0.013"),
     gpu_a100sxm4 = Decimal("2.078"),
     gpu_a100 = Decimal("1.803"),
-    gpu_v100 = Decimal("1.214")
+    gpu_v100 = Decimal("1.214"),
+    gpu_h100 = Decimal("6.04"),
     )
 
 class TestGetNamespaceAnnotations(TestCase):
@@ -567,6 +568,13 @@ def test_known_gpu_V100(self):
         self.assertEqual(su_count, 1)
         self.assertEqual(determining_resource, "GPU")
 
+    def test_known_gpu_H100(self):
+        pod = self.make_pod(124, 360, 1, invoice.GPU_H100, invoice.WHOLE_GPU)
+        su_type, su_count, determining_resource = pod.get_service_unit()
+        self.assertEqual(su_type, invoice.SU_H100_GPU)
+        self.assertEqual(su_count, 1)
+        self.assertEqual(determining_resource, "GPU")
+
     def test_known_gpu_high_cpu(self):
         pod = self.make_pod(50, 96, 1, invoice.GPU_A100, invoice.WHOLE_GPU)
         su_type, su_count, determining_resource = pod.get_service_unit()