Skip to content

Commit 4d1a497

Browse files
authored
Merge pull request #377 from Modalities/feat--support-NVIDIA-GH200-120GB
feat: supports now gh200 mfu calculations
2 parents 8270a7a + 38cf338 commit 4d1a497

File tree

1 file changed

+5
-4
lines changed

1 file changed

+5
-4
lines changed

src/modalities/utils/mfu.py

Lines changed: 5 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -14,10 +14,7 @@
1414
# https://www.nvidia.com/en-us/data-center/h100/
1515
#
1616
# NOTE: These values are valid for fp16 and bf16 only
17-
PEAK_PERFORMANCE = {
18-
"A100": 312e12,
19-
"H100": 989e12,
20-
}
17+
PEAK_PERFORMANCE = {"A100": 312e12, "H100": 989e12, "GH200": 989e12}
2118

2219

2320
class MFUCalculatorABC:
@@ -128,6 +125,10 @@ def _get_theoretical_gpu_peak_performance(wrapped_model: FSDPX, world_size: int)
128125
single_gpu_peak_performance = MFUCalculatorABC._get_theoretical_gpu_peak_performance_single(
129126
precision, "H100"
130127
)
128+
elif device_name.startswith("NVIDIA GH200"):
129+
single_gpu_peak_performance = MFUCalculatorABC._get_theoretical_gpu_peak_performance_single(
130+
precision, "GH200"
131+
)
131132
else:
132133
warnings.warn(f"Could not get theoretical GPU peak performance for unknown device = {device_name}.")
133134
return None

0 commit comments

Comments
 (0)