Hi folks, Just curious if anyone here has tried PGGB with their dCS rig. This has been the subject of a long and winding thread over at AudiophileStyle which has piqued my interest.
If you haven’t seen it, it’s a software tool that its author describes as “offline remastering”. In simple terms it’s a sort of “offline HQP”. You process your files offline and then stream the upsampled/EQed version instead. The potential benefits are clear:
the ability to perform transformations too computationally complex to perform in real time with rational/commercially viable amounts of compute capacity at current prices
the ability to easily scale compute resources (principally RAM but also CPU) according to the amount of processing required (32fS vs 16fS etc)
the avoidance of potential detrimental side effects of the real-time processing
etc.
I have seen comments in other threads from folks who are skeptical to off-platform processing. Wondering if that skepticism extends to offline processing and if anyone here has tried it?
Seems that “zettelsm” on AudioPhileStyle has tried it with a Vivaldi-based system. @oldmustang, that wouldn’t happen to be you by any chance?
As you mentioned, I’ve tried it with my dCS Vivaldi “short stack”. I’ve had mixed results. Some tracks and albums it has improved, sometimes moving the needle from “unlistenable” to “very pleasant”. Other times not so much.
I’ve upsampled to 8fs (the limit of dCS equipment) 24 bit (the native bit depth of dCS equipment), IOW 352.8 or 384/24. I’ve compared directly into my DAC as well as through the Upsampler set to DXD Lock (max rate 384/24).
I think ones’ assessment of PGGB will strongly be influenced by ones’ sonic priorities. The transparency over everything crowd tend to give PGGB unqualified raves, however in my own listening I’ve found this transparency can veer over the line into thinness and lack of body.
I’ve also found it difficult to predict which music files will benefit and which will not. It seems that the better the original engineering and mastering the better the result, or at least, the better the chance that PGGB will work well. Which is a pity because in my music library it is usually the early rock 1fs CD rips that need the most help.
Overall, I’ve found my interest in PGGB has waned over time. The advantage for me of the Upsampler is that it works in real time on streaming content, which makes up a significant percentage of my listening. Also, while a minimum specification PC or Mac will allow PGGB to upsample 1fs content to 8fs pretty easily, it takes a progressively more powerful platform as the original file sample rate increases.
I guess my bottom line is that when it works (for me) it can make a striking improvement. But it is not a universal panacea in my experience and can make some content sound thin(ner) and somewhat overly sanitized.
It is worth the free trial to hear if it will work for you.
While this sounds interesting, I had a look at their licensing and was a bit put off by the fact they seem to tie the license to one specific machine. So, if you upgrade your PC, or (potentially) your hard drive fails, your investment goes up in a puff of smoke.
On something like rooExtend costing much smaller amounts of money, I could understand this, but at up to ~$1000, this seems pretty harsh.
Anyone had any experience persuading them to move your license from one machine to another?
The dCS family (except for the Lina) use I believe a pair of Freescale DSP56309 chips for its primary upsampling compute tasks - thats a specialised Digital Signal Processors that can perform 100 million multiply-accumulates per second!
What these DSPs can do in real-time would take a general purpose CPU many minutes, if not more.
The Lina on the otherhand, IINM, uses parts of the Xilinx Artix-7 chip for its DSP function for upsampling. Coincidentally, that’s the same chip that Chord Electronics uses in their M Scaler upsampler which famously touts “1 Million Taps”, which is able to Upsample to 2 x DXD rates in just 0.6 seconds.
By the way, if you’re interested in a technical analysis of PGGB, check out;
Thanks for the link Anup, that was a very interesting read!
Not quite sure what you mean by your “Not quite actually” comment though, please could you explain? I still believe removing the constraints that the requirement to process in real time imposes is a potential benefit if, for instance (and of course I am not asserting this), it means the job of a $27k Vivaldi Upsampler can be done by a $2k PC.
Maybe you mean that the MAC headroom of the DSP in the Vivaldi Upsampler is so high that it has the horsepower to meet all foreseeable future needs, so even if it is a potential benefit it is moot. Yes?
One thing is clear however. Regardless of any sound quality improvement or advantage in absolute performance or price performance vs the Vivaldi Upsampler, to @Chocky 's point it is a lot of hassle!
HQplayer relies on NVIDIA GPUs for acceleration if I’m not mistaken. Personally I wouldn’t want a pc with a discrete graphics board anywhere near a hi if system because of the noise and heat output. The chips dCS uses mentioned above will do just as good a job or better, more quickly.
I’ve just checked the specs and it looks like for the GPU offload/acceleration to work, you do need NVIDIA GPU’s, but I don’t think HQPlayer needs them to function.
I’m wondering if your memory may be playing tricks on you there @Urbanluthier? For SRC (if we temporarily ignore the subordinate use cases), which is essentially a sequential process, the parallel processing strengths of a GPU would not seem to me to be particularly relevant. Strike me down if I am wrong here, but this is certainly what the theory suggests.
Regardless, I think there is a fairly strong consensus that general purpose computers aren’t the optimum solution for any real time processing in an audio chain for a host of reasons. My intention at the start of this thread was to focus on offline solutions, but maybe understandably the focus has broadened to also encompass real time applications.
Sorry Jeremy, you’ve lost me there. Per my post above I don’t see the relevance of parallel processing horsepower for DSP (e.g. SRC) which is entirely sequential. Please enlighten me!
Sorry Jeremy, I’m being a bit thick here. I’ve trawled the Signalyst website for said specs but can’t find them. Please could you post a link? Hopefully something there will provide the necessary enlightenment and the penny will drop!
Oh just that dedicated DSPs can run circles round any general purpose CPU for specialised mathematical tasks. So, the supposed benefits of offline processing with a general purpose compute platform isn’t as significant as PGGB might make it out to be.
An embedded system like dCS might seem limited in resources, but it doesn’t actually take tens of gigabytes of RAM for well designed interpolation filters to do a good job. Plus, no offence to folks like Audiowise (PGGB), but dCS has been doing this for a very long time
@jandersonhill and @Anupc arecorrect an NVIDIA gup is not required for hqplayler but the software will use it for intensive tasks like DSD 512 transcoding upsampling. Some people want to do this for some reason.
