For current Nvidia GPUs, the L1 cache system is not coherent across different SMs. Only the L2 cache is coherent across the entire chip.
kayvonf
@200. You are correct that the L1's are not coherent, but there is only a single shared L2 on modern NVIDIA GPUs. Why does this eliminate the need for cache coherence at this level of the memory hierarchy?
200
@kayvonf. Because all the processors access contents from the same source, so there is no data replication.
For current Nvidia GPUs, the L1 cache system is not coherent across different SMs. Only the L2 cache is coherent across the entire chip.
@200. You are correct that the L1's are not coherent, but there is only a single shared L2 on modern NVIDIA GPUs. Why does this eliminate the need for cache coherence at this level of the memory hierarchy?
@kayvonf. Because all the processors access contents from the same source, so there is no data replication.