P-State issue
Driver Version
550.127.05
CUDA Version
12.4
Power Limit
450 W
P-State
P2
Driver Version
565.57.01
CUDA Version
12.7
Power Limit
450 W
P-State
P0
We have two serverless machines. The first one is very slow, while the second one performs at the expected speed for a standard 4090. I'm wondering—what does P-state mean? Does it affect the GPU's performance?
10 Replies
Unknown User•2mo ago
Message Not Public
Sign In & Join Server To View
what can cause this? is there a way I can always let it keep at P0?
and can I delete this endpoint if I notice the perf is bad? Any API can do this besides go to the dashboard and delete that node?
Unknown User•2mo ago
Message Not Public
Sign In & Join Server To View
can you reference me to it?
I only see how to stop/cancel the current job, not seeing directly delete endpoint
Unknown User•2mo ago
Message Not Public
Sign In & Join Server To View
btw this is a start time graph and avg frame time graph.
the lower the better, for both case.
you can easily see that the quick start machine has a lower frame time (higher fps).
Those are running the exactly same task
The start up time is mainly python loading those ML models.
Unknown User•2mo ago
Message Not Public
Sign In & Join Server To View
it's the machine that can load my ml model quick.
they are loading the ml models into gpu
These are TensorRT (.trt) engine files being loaded from disk into GPU memory.
And in the worst case, those loading can take up to 2 mins, and left some extremely bad frame time (200ms)
Unknown User•2mo ago
Message Not Public
Sign In & Join Server To View