Code
Platform: NVIDIA CUDA
Device: NVIDIA GeForce RTX 3090
Driver version : 527.56 (Win64)
Compute units : 82
Clock frequency : 1755 MHz
Global memory bandwidth (GBPS)
float : 797.56
float2 : 821.78
float4 : 816.65
float8 : 740.41
float16 : 812.96
Single-precision compute (GFLOPS)
float : 35310.54
float2 : 34988.51
float4 : 34958.87
float8 : 35000.56
float16 : 33970.65
No half precision support! Skipped
Double-precision compute (GFLOPS)
double : 626.31
double2 : 623.39
double4 : 625.22
double8 : 617.13
double16 : 616.63
Integer compute (GIOPS)
int : 18599.87
int2 : 18412.26
int4 : 18557.66
int8 : 18215.33
int16 : 18089.46
Integer compute Fast 24bit (GIOPS)
int : 17922.65
int2 : 17961.78
int4 : 18407.26
int8 : 17958.23
int16 : 18123.10
Transfer bandwidth (GBPS)
enqueueWriteBuffer : 11.98
enqueueReadBuffer : 12.22
enqueueWriteBuffer non-blocking : 11.96
enqueueReadBuffer non-blocking : 12.25
enqueueMapBuffer(for read) : 5.50
memcpy from mapped ptr : 16.07
enqueueUnmap(after write) : 12.96
memcpy to mapped ptr : 16.29
Kernel launch latency : 14.80 us
Alles anzeigen
NVIDIA mal wieder (wie immer) komplett unbrauchbar bei Gleitkomma.
























