In classic CPU client, "recognizing" a CPU meant that client can select best core for given CPU type automatically. Unfortunately, it's not possible in OpenCL where best core selection depend not only on device model but on numerous factors like tweaked GPU clocks (GPU/memory) and version of drivers (driver really is a full-scale C compiler for OpenCL program, different driver versions may generate different code).

When a CPU or GPU is "unrecognized", client automatically running small micro benchmark at startup to find fastest core. This is a only difference between "recognized" and "unrecognized" device. So, all these messages should be ignored. The "32-bit" line is a debugging message about features of the GPU, it should be ignored too.

We suggest to do a manual benchmark for your particular setup (e.g. running "dnetc -bench" few times) and select best core in configuration menu to avoid micro-benchmark and delay on startup.