In NVIDIA GPU profiling, what are sub-partitions, sectors and units?

Question

What are sub-partitions, sectors and DRAM units as related to the NVIDIA GPU performance counters for memory transactions?

I'm looking for information about how to interpret the following performance counters:

fb_subp0_read_sectors : Number of read requests sent to sub-partition 0 of all the DRAM units fb_subp1_read_sectors : Number of read requests sent to sub-partition 1 of all the DRAM units fb_subp0_write_sectors : Number of write requests sent to sub-partition 0 of all the DRAM units fb_subp1_write_sectors : Number of read requests sent to sub-partition 1 of all the DRAM units fb0_subp0_read_sectors : Number of read requests sent to sub-partition 0 of DRAM unit 0 fb0_subp1_read_sectors : Number of read requests sent to sub-partition 1 of DRAM unit 0 fb0_subp0_write_sectors : Number of write requests sent to sub-partition 0 of DRAM unit 0 fb0_subp1_write_sectors : Number of write requests sent to sub-partition 1 of DRAM unit 0 fb1_subp0_read_sectors : Number of read requests sent to sub-partition 0 of DRAM unit 1 fb1_subp1_read_sectors : Number of read requests sent to sub-partition 1 of DRAM unit 1 fb1_subp0_write_sectors : Number of write requests sent to sub-partition 0 of DRAM unit 1 fb1_subp1_write_sectors : Number of write requests sent to sub-partition 1 of DRAM unit 1

I checked the documentation for the NVIDIA Visual Profiler 5.0 but did not find any additional information there.

Answer 1

For most of the gpus you will see following four counters: fb_subp0_read_sectors : Number of read requests sent to sub-partition 0 of all the DRAM units fb_subp1_read_sectors : Number of read requests sent to sub-partition 1 of all the DRAM units fb_subp0_write_sectors : Number of write requests sent to sub-partition 0 of all the DRAM units fb_subp1_write_sectors : Number of read requests sent to sub-partition 1 of all the DRAM units

The DRAM is divided into multiple units(partitions) and each unit is divided into max 2 sub-partitions. When you profile fb_subp0_read_sectors or fb_subp0_write_sectors counter, you will get aggregate value for subparition 0 from all the units. Similarly when you profile fb_subp1_read_sectors or fb_subp1_write_sectors you will get aggregate value for subpartition 1 from all the DRAM units. Sector is the unit in which DRAM memory is accessed and it is 32 bytes.

Other counters are valid only for some gpus in fermi family(GeForce GT 440/430/420 and GeForce GT 520/530, GeForce GT 610). They were introduced due to some hardware limitation.