Optimum # of cores vs speed - Xeon E5 etc.

[Frank]

All other things being equal, I presume that the more cores you have, the better speed you will get with Handbrake, especially with the newer/nightly builds.

But all things aren't equal, more cores, more $, unless we downshift in GHz…

So of the configuration (assuming RAM, Disk I/O etc remain the same), what would be more effective as a handbrake encoding/crunching machine?

Intel Xeon E5 with 10MB L3 cache and Turbo Boost up to 3.9GHz
• 3.5GHz 6-core processor with 12MB L3 cache
• 3.0GHz 8-core processor with 25MB L3 cache
• 2.7GHz 12-core processor with 30MB L3 cache

or

Intel Xeon E5 with 12MB L3 cache and Turbo Boost up to 3.9GHz
•3.0GHz 8-core processor with 25MB L3 cache
• 2.7GHz 12-core processor with 30MB L3 cache

[Smithcraft]

It depends on certain functions which might not be able to use multiple threads, and if you use those functions.

SC

[Frank]

Interesting… which are the thread-limited options in Handbrake? The resizing / scaling filters or the noise filters? Are there others to consider?

My basic workflow involves encoding incoming 1080i ts files to a 720p and passing the AC3 audio through untouched.

[Deleted User 13735]

Encode log please?
When you know exactly what filters are in the chain, it should be fairly easy for you to identify which ones are bottlenecks by a process of elimination, or by simply trying various x264 presets, which are ordered by relative speed.

The number of CPU threads is by far not the only limiting factor to encoding speed. Some bit-intensive operations are just slow, period. Certainly resizing and denoise, two that you mentioned, are among them.

libx264 scales well up to ~6 cores and plateaus after that. Using any filters impacts on this as many are not well threaded.

The filters all come from different external libraries, some dating back years; no one maintains a comprehensive list of what you want to know, afaik. See this thread for additional discussion wrt the handbrake engine.
viewtopic.php?f=6&t=21278&p=98072

[Deleted User 11865]

IIRC encoders, decomb/deinterlace and some video decoders are threaded - everything else only uses a single thread.

But you can run more than one encode at once.

Assuming you're comparing apples to apples (i.e. the same underlying microarchitecture; basically, the same generation of Xeon), just do cores * frequency to get an estimate of performance. If you don't want to run multiple encodes concurrently, then 4 to 6 cores is your best bet.