Best practices for —pools?

[DotJun]

When encoding multiple instances, is it better to just let x265 determine the number of threads on its own or would it be better to specify a number?

For example: When encoding 2 instances on 12 cores x265 will spawn 12 threads for each of the 2 encodes. Would it be better to set —pools 6 for each encode (12 cores / 2 encodes = 6 cores each)?

I’m assuming that if left to default values x265 would use any/all cores for each encode at its whim. What exactly is it doing when setting —pools then? Does one encode get cores 0-5 and the second encode gets cores 6-10?

[Selur]

For me, using pools only seems to help when running multiple parallel encodes on systems with multi cpus,...

[DotJun]

For me, using pools only seems to help when running multiple parallel encodes on systems with multi cpus,...

Yes I agree that using pools on multi socket is helpful, but my question is more to do with a single node that has an abundance of cpus. I really just want to understand what —pools is actually doing under the hood.

[benwaggoner]

There is a long and pretty exhaustive description here: https://x265.readthedocs.io/en/master/cli.html#performance-options
And a good overview of threading in x265 here: https://x265.readthedocs.io/en/master/threading.html

Did you have a question beyond what's covered in those?

[DotJun]

There is a long and pretty exhaustive description here: https://x265.readthedocs.io/en/master/cli.html#performance-options
And a good overview of threading in x265 here: https://x265.readthedocs.io/en/master/threading.html

Did you have a question beyond what's covered in those?

Thank you for the reading material. It was informative, but some of the wording used left me with questions. For example, when setting pools +,+ the encoder should be free to use all available threads on nodes 0 and 1, but it does not. Is this because a single pool cannot jump nodes? I’m assuming only a single pool is created, even though my log says that two pools are created with one on each node, both pools are labeled as pool 0.

[benwaggoner]

Thank you for the reading material. It was informative, but some of the wording used left me with questions. For example, when setting pools +,+ the encoder should be free to use all available threads on nodes 0 and 1, but it does not. Is this because a single pool cannot jump nodes? I’m assuming only a single pool is created, even though my log says that two pools are created with one on each node, both pools are labeled as pool 0.
Encoding everything on a single pool is more efficient, so unless x265 can use more threads than will fit on a single pool, it'll stick to one. I've been able to get two to work when doing 8K content, for example.