NeurIPS2018 no speedup with increasing batch size

[charliedream1]

I tested demo NeurIPS2018 with stablebaseline3, I used SAC agent, and I trained with GPU. While I increase batch size from 128 to 512, I found no changing for GPU memories and utilization rate.

The version I used as below:
stable-baselines3==1.5.0
torch==1.10.0

Training time has no change with chaning batch size, what would be the problem?

[YangletLiu]

It seems that SB3 is not optimized for GPU. If you are dealing with compute-intensive cases, ElegantRL may be a good choice.

[charliedream1]

thx，I will try on it

…

---Original--- From: "Xiao-Yang ***@***.***> Date: Mon, Mar 27, 2023 12:51 PM To: ***@***.***>; Cc: "Optimus ***@***.******@***.***>; Subject: Re: [AI4Finance-Foundation/FinRL-Tutorials] NeurIPS2018 no speedupwith increasing batch size (Issue #48) It seems that SB3 is not optimized for GPU. If you are dealing with compute-intensive cases, ElegantRL may be a good choice. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: ***@***.***>