clore
/
hosting
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

V5.2.1 | add dict for core clocks for problematic gpus on "function not found" error

This commit is contained in:
clore 2024-05-28 00:56:26 +00:00
parent 5e733fd0d6
commit 73d19b5cd7
1 changed files with 39 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

39
lib/nvml.py
View File

@ -23,7 +23,7 @@ GPU_MEM_ALLOWED_OC_RANGES = { # Known to be problematic GPUs
"NVIDIA GeForce GTX 1070": [-2000, 2000], "NVIDIA GeForce GTX 1070": [-2000, 2000],
"NVIDIA GeForce GTX 1070 Ti": [-2000, 2000], "NVIDIA GeForce GTX 1070 Ti": [-2000, 2000],
"NVIDIA GeForce GTX 1080": [-2000, 2000], "NVIDIA GeForce GTX 1080": [-2000, 2000],
"NVIDIA GeForce GTX 1080 Ti":[-2000, 2000], "NVIDIA GeForce GTX 1080 Ti": [-2000, 2000],
"NVIDIA CMP 30HX": [-2000, 6000], "NVIDIA CMP 30HX": [-2000, 6000],
"NVIDIA CMP 40HX": [-2000, 6000], "NVIDIA CMP 40HX": [-2000, 6000],
"NVIDIA CMP 50HX": [-2000, 6000], "NVIDIA CMP 50HX": [-2000, 6000],
@ -39,6 +39,33 @@ GPU_MEM_ALLOWED_OC_RANGES = { # Known to be problematic GPUs
"NVIDIA GeForce RTX 2080 Ti": [-2000, 6000] "NVIDIA GeForce RTX 2080 Ti": [-2000, 6000]
} }
GPU_CORE_ALLOWED_OC_RANGES = { # Known to be problematic GPUs
"NVIDIA P102-100": [-200, 1200],
"NVIDIA P104-100": [-200, 1200],
"NVIDIA P106-090": [-200, 1200],
"NVIDIA P106-100": [-200, 1200],
"NVIDIA GeForce GTX 1050 Ti": [-200, 1200],
"NVIDIA GeForce GTX 1060 3GB": [-200, 1200],
"NVIDIA GeForce GTX 1060 6GB": [-200, 1200],
"NVIDIA GeForce GTX 1070": [-200, 1200],
"NVIDIA GeForce GTX 1070 Ti": [-200, 1200],
"NVIDIA GeForce GTX 1080": [-200, 1200],
"NVIDIA GeForce GTX 1080 Ti": [-200, 1200],
"NVIDIA CMP 30HX": [-1000, 1000],
"NVIDIA CMP 40HX": [-1000, 1000],
"NVIDIA CMP 50HX": [-1000, 1000],
"NVIDIA CMP 90HX": [-1000, 1000],
"NVIDIA GeForce GTX 1650": [-1000, 1000],
"NVIDIA GeForce GTX 1660 SUPER": [-1000, 1000],
"NVIDIA GeForce GTX 1660 Ti": [-1000, 1000],
"NVIDIA GeForce RTX 2060": [-1000, 1000],
"NVIDIA GeForce RTX 2060 SUPER": [-1000, 1000],
"NVIDIA GeForce RTX 2070": [-1000, 1000],
"NVIDIA GeForce RTX 2070 SUPER": [-1000, 1000],
"NVIDIA GeForce RTX 2080": [-1000, 1000],
"NVIDIA GeForce RTX 2080 Ti": [-1000, 1000]
}
is_hive = False is_hive = False
all_gpus_data_list=[] all_gpus_data_list=[]
get_data_fail=False get_data_fail=False
@ -134,6 +161,16 @@ def init(gpu_specs_file=None):
else: else:
mem_range=[0,0] mem_range=[0,0]
except Exception as e2: except Exception as e2:
if "function not found" in str(e2).lower():
if gpu_spec["name"] in GPU_MEM_ALLOWED_OC_RANGES:
mem_range = GPU_MEM_ALLOWED_OC_RANGES[gpu_spec["name"]]
else:
mem_range = [0,0]
if gpu_spec["name"] in GPU_CORE_ALLOWED_OC_RANGES:
core_range = GPU_CORE_ALLOWED_OC_RANGES[gpu_spec["name"]]
else:
core_range = [0,0]
else:
get_data_fail=True get_data_fail=True
if type(mem_range) == list and type(core_range) == list and len(mem_range)==2 and len(core_range)==2: if type(mem_range) == list and type(core_range) == list and len(mem_range)==2 and len(core_range)==2:
gpu_spec["mem"]=mem_range gpu_spec["mem"]=mem_range