Quote:
Originally Posted by Romario
Can you please optimize your fantastic program for AMD Ryzen arhitecture ? Can you do it before final version come ? Thank you.
Is there any audio codec, at the moment, being optimized for Ryzen ?
As ilustration, look at this image in my attachment, please
|
Development of LAME is pretty much dead. So don't expect any optimizations for "new" processors any time soon
Having said that, there is
no "magic" optimization that would make LAME run a lot faster on Zen architecture all of a sudden! Yes, LAME could probably benefit from using "modern" instruction set extensions, such as AVX or AVX2 – at the moment the assembler code in LAME only uses the SSE and SSE2 extensions – but this "problem" is
not specific to the Zen architecture at all. TTBOMK, the Zen architecture did
not even introduce a new instruction set extension.
The reason why Ryzen 1800X performs
relatively "bad" with LAME mp3 encoder is because LAME – just like most audio encoders – is completely
single-threaded. And, given the current state of LAME development, I doubt this is ever going to change. Furthermore, it is well known that
single-core performance still is the "weak spot" of the Zen architecture (compared to Intel Core i7 processors), even though things have been improved quite a lot since Bulldozer architecture.
As others have pointed out: You will need to run
many LAME instances
in parallel to take full advantage of a Ryzen processor with 16 logic cores – which is exactly what LameXP does
Quote:
Originally Posted by Romario
Well I see my attachment without any problem. Strange...
|
Attachments need to be approved by a mod, before other users can see/access them. I wasn't close to a computer in the last couple of days, so I couldn't approve it until now.
Quote:
Originally Posted by Romario
How can I use optimised lame 3.99.5 builds? While LameXP doesn't allow to change which lame build I use.
|
You can make LameXP use a custom build of LAME by putting your custom binary at the proper place, just like it is described in the F.A.Q. article that
manolito quoted (not ported to new manual yet).
Don't expect any wonders, though! LameXP already ships with up-to-date binaries of LAME 3.100 Alpha – including 64-Bit build with AVX compiler-optimizations enabled. The "best" build is automatically selected, based on the CPU's capabilities.
Note: Those
compiler optimizations have a lesser effect on performance. That's because all the "hot" functions, where the
vast majority of the CPU time is spent, already are written as hand-optimized assembler code...