|
@@ -89,6 +89,7 @@ CHIP_FLOPS = {
|
|
|
"NVIDIA GEFORCE RTX 2070": DeviceFlops(fp32=7.46*TFLOPS, fp16=14.93*TFLOPS, int8=29.86*TFLOPS),
|
|
|
"NVIDIA GEFORCE RTX 2070 SUPER": DeviceFlops(fp32=9.06*TFLOPS, fp16=18.12*TFLOPS, int8=36.24*TFLOPS),
|
|
|
"NVIDIA GEFORCE RTX 2080": DeviceFlops(fp32=10.07*TFLOPS, fp16=20.14*TFLOPS, int8=40.28*TFLOPS),
|
|
|
+ "NVIDIA GEFORCE RTX 2080 TI": DeviceFlops(fp32=13.45*TFLOPS, fp16=26.9*TFLOPS, int8=40.28*TFLOPS),
|
|
|
"NVIDIA GEFORCE RTX 2080 SUPER": DeviceFlops(fp32=11.15*TFLOPS, fp16=22.30*TFLOPS, int8=44.60*TFLOPS),
|
|
|
"NVIDIA TITAN RTX": DeviceFlops(fp32=16.31*TFLOPS, fp16=32.62*TFLOPS, int8=65.24*TFLOPS),
|
|
|
# QUADRO RTX Ampere series
|
|
@@ -178,7 +179,8 @@ def linux_device_capabilities() -> DeviceCapabilities:
|
|
|
|
|
|
pynvml.nvmlInit()
|
|
|
handle = pynvml.nvmlDeviceGetHandleByIndex(0)
|
|
|
- gpu_name = pynvml.nvmlDeviceGetName(handle).upper()
|
|
|
+ gpu_raw_name = pynvml.nvmlDeviceGetName(handle).upper()
|
|
|
+ gpu_name = gpu_raw_name.rsplit(" ", 1)[0] if gpu_raw_name.endswith("GB") else gpu_raw_name
|
|
|
gpu_memory_info = pynvml.nvmlDeviceGetMemoryInfo(handle)
|
|
|
|
|
|
if DEBUG >= 2: print(f"NVIDIA device {gpu_name=} {gpu_memory_info=}")
|