Re: Lowering latency
--------------------------------------------------------------------------------
SAC is certainly written to use multiple processors... this has been mis-quoted many times... but as has been stated in many of my messages, there are reasons why multiple processors can actually cause a loss in performance for a realtime engine such as SAC... in reality, the maximum performance can currently be achieved with 2 cores... but using multiple cores can also cause more slipped buffers, depending on many factors.
So... a single highest speed core would give best overall results... with little or no slipped buffers even when loaded to 99%... but in today's world, multiple cores became the norm in the attempt to improve performance and advertise faster and faster computers, since they have pretty much maxed out the speed they can achieve on a single core.
The issue, once again, has to do with a lack of thread priority control once multiple cores are involved as well as a common memory pool across all cores... this is what allows lower priority threads from one core to stomp all over the main realtime priority thread of the engine core, which causes the main thread to not have enough time to finish assembling all of its buffers... overall causing a loss of performance and many slipped buffers.
Bob L
I can challange the entire last paragraph with one word: AMP. It seems to handle all of these problems quite well, and faster than SAC (by far).
From measurements take today:
PC – RME Card – running at 48K in all tests.
Using TotalMix loopback right on the card: 0.79 ms
Using AMP:
(48K / 64 Samples) 0 buff = 4.12 ms 1 buff = 5.46 ms 2 buff = 6.79 ms
(48K / 32 Samples) 0 buff = 2.79 ms 1 buff = 3.46 ms 2 buff = 4.12 ms
Using SAC:
(48K / 64 Samples) 1 buff = 5.46 ms 2 buff = 6.79 ms
(48K / 32 Samples) 1 buff = 3.46 ms 2 buff = 4.12 ms
All measurements were taken using SMAART 7.0 and a SMAART I/O and Pink generated by same.
This system is set to run at 64x2 under SAC (the most stable with only an occasional slipped buffer).
This system runs perfectly at 64x0 under AMP (no slipped buffers)
The latency difference is 6.79 down to 4.12 - a difference of 2.67 ms - It is a significanly noticible difference, even on wedges.