during my recent overclocking adventures with the titan x, ive found out that RT (cuda apps in general?) set the gpu to power state p2, as opposed to P0
P0 is used for games and features the highest clocks for core and memory.
P2 is a more conservative set of clocks, particularly on the memory.
i assume this is deliberate so critical software running on cuda gets guaranteed stability.
however, it might be worth (if you didnt already) seeing if its possible to persuade RT to use power state P0 instead?
i dont know if this is possible or if it is baked into the NV drivers.
when i use NVinspector to manually match the ram clock in p2 to the one in p0 i get a healthy speedup in RT, and no issues.
this also explains why overclocking the titan X has some strange issues with RT (no speedup etc) since the "gaming" overclocking applications (precision, afterburn etc) only affect power state P0.
P0 is used for games and features the highest clocks for core and memory.
P2 is a more conservative set of clocks, particularly on the memory.
i assume this is deliberate so critical software running on cuda gets guaranteed stability.
however, it might be worth (if you didnt already) seeing if its possible to persuade RT to use power state P0 instead?
i dont know if this is possible or if it is baked into the NV drivers.
when i use NVinspector to manually match the ram clock in p2 to the one in p0 i get a healthy speedup in RT, and no issues.
this also explains why overclocking the titan X has some strange issues with RT (no speedup etc) since the "gaming" overclocking applications (precision, afterburn etc) only affect power state P0.
Comment