VINDICATORs Pour moi c’est bon, voici quelques infos pour aiguiller au pire mais je n’ai pas de cartes aussi récentes.
$ lspci -nnk | grep -iA3 "VGA"
01:00.0 VGA compatible controller [0300]: Advanced Micro Devices, Inc. [AMD/ATI] Curacao XT / Trinidad XT [Radeon R7 370 / R9 270X/370X] [1002:6810]
Subsystem: PC Partner Limited / Sapphire Technology Device [174b:e270]
Kernel driver in use: radeon
Kernel modules: radeon, amdgpu
$ rpm -qa rocm-opencl
rocm-opencl-5.5.0-1.fc38.x86_64
$ clpeak
Platform: Clover
Device: PITCAIRN (, LLVM 16.0.1, DRM 2.50, 6.3.3-200.fc38.x86_64)
Driver version : 23.0.3 (Linux x64)
Compute units : 20
Clock frequency : 1100 MHz
Global memory bandwidth (GBPS)
float : 140.63
float2 : 148.91
float4 : 151.36
float8 : 142.34
float16 : 68.74
Single-precision compute (GFLOPS)
float : 346.98
float2 : 346.92
float4 : 346.34
float8 : 345.06
float16 : 343.73
No half precision support! Skipped
Double-precision compute (GFLOPS)
double : 173.84
double2 : 173.66
double4 : 173.40
double8 : 172.85
double16 : 172.34
Integer compute (GIOPS)
int : 138.96
int2 : 138.87
int4 : 138.65
int8 : 138.22
int16 : 139.07
Integer compute Fast 24bit (GIOPS)
int : 686.73
int2 : 684.34
int4 : 679.20
int8 : 668.99
int16 : 649.53
Transfer bandwidth (GBPS)
enqueueWriteBuffer : 5.27
enqueueReadBuffer : 4.63
enqueueWriteBuffer non-blocking : 5.23
enqueueReadBuffer non-blocking : 4.69
enqueueMapBuffer(for read) : 2514.62
memcpy from mapped ptr : 4.96
enqueueUnmap(after write) : 2118.67
memcpy to mapped ptr : 5.44
Kernel launch latency : 97.40 us
Platform: rusticl
clCreateContextFromType (-1)