Quote:
Originally Posted by crwilli
I don’t own the PS audio DAC.
Reading about the consistent software upgrades causes me to ponder what exactly happens between each upgrade that makes Ted and Paul feel ‘this is better’? What is it they are learning about programming the DAC?
|
The latest big jump which came out in RedCloud was when they upsampled internally to 20x DSD vs 10x originally.
Ted didn’t even expect the FPGA he used to have that much headroom to do this. But this came about because of code optimization which reduced the processor load and also improved sound quality each time.
Each time a new OS was released, invariably one of the goals was to improve the noise floor. And this seemed to be by reducing spikes on the processing load. Which I suspect was how they eventually have enough headroom to do 20x.
EMM Labs and Playback all started with upsampling to 10x DSD but it took them a new processor and new DAC to implement 20x so kudos to Ted and PS Audio for persevering with an old architecture.
Even when he optimizes the code, how the code is compiled apparently creates slightly different sounds. That’s when Paul, Bascom, and the late Arnie Nudell used to perform listening tests to see which build they preferred.
Reducing power spikes isn’t a new revelation. One of the best sounding releases of Pure Music (1.5? Release number escapes me now) was when the developer realized that the best optimization he could do was reduce the power spikes in his algorithms. The sad part is that Apple introduced a new OS later which broke that release and developers like PureMusic began spending resources for compatibility and breakfixes for each new (annual) OS release instead of sound quality optimizations. Incidentally, I suspect why WAV sounded better than FLAC in computer audio was also because of lower power spikes. This isn’t necessarily the same as CPU utilization because they are still relatively low. But in switching between machine states, some paths may lead to sudden short spikes.
Speaking of power spikes, one of the lead engineers in ESS (Sabre32) reached a similar conclusion in his talk at RMAF a number of years ago. He said that when he started he only measured steady state (final machine states) but he found that when he did blind tests with more established Hifi DAC designers they could consistently identify better sounding designs over his better “measured” designs. The conclusion he reached was that often, it was the machine states with lower power spikes that sounded better. Even if it took more states for the state machine to reach the end state.