Should I upsample in software (HQPlayer, PGGB or my player) before a Chord DAC, and can it match an M Scaler?

The advice, 2014 to 2019: let the DAC do it

His earliest relevant statement in the archive, on 26 June 2014 for Hugo, concerned a PC converting 24/192 files down in sample rate. He said "you don't know how well they do that", described a track that sounded awful because the PC was converting everything to 48 kHz, and concluded "So on balance probably best to stick with native, don't let the PC mess it up." He hedged that he would be intrigued if someone tried simple binary decimation, 192 kHz to 48 kHz Head-Fi, Jun 2014. On 30 August 2014, again for Hugo and now about upsampling: "The standard algorithms are poor, with insufficient tap length, just let Hugo do it properly." He made one exception, the Red Reference, which he described as an 18K WTA up-sampler that would improve on Hugo, adding "But this is the only exception to the rule!" Head-Fi, Aug 2014.

On 22 November 2015, in the DAVE thread, he told users not to use a computer to upsample or change the data with any of his DACs, and wrote "Computers are poor devices to use for manipulating data in real time" and set the parallel FPGA processing in DAVE against a PC's sequential processors, while adding that the filter algorithm matters as well as raw power Head-Fi, Nov 2015. He repeated the processing argument for DAVE in August 2016 Head-Fi, Aug 2016. In February 2017 he told Mojo, Hugo and DAVE owners "So keep the data to Mojo/Hugo/Dave bit perfect." because PC tap lengths are limited and the algorithms conventional Head-Fi, Feb 2017.

For Hugo 2 in September 2017 he said "it's vital that you ensure the source does not up-sample" Head-Fi, Sep 2017. In November 2017 he said PC-based upsamplers meet neither of his two conditions (about 1 million real taps at 16FS, and coefficients matching an ideal sinc to 16 bits) and "they don't even come close to Mojo's WTA performance let alone an M scaler" Head-Fi, Nov 2017. For Qutest he gave the same advice in April and September 2018: "You will get poor results if a 44.1k file is upsampled" and "make sure the source is bit perfect, with no up-sampling" Head-Fi, Apr 2018Head-Fi, Sep 2018. In March 2018 he wrote "Never use up-sampling from your source into one of my DAC's, even a Mojo" Head-Fi, Mar 2018.

For the Hugo M Scaler itself, on 26 July 2018: "You need to send the M scaler bit perfect data." He gave two reasons, limited processing power and not using the WTA algorithm, and said even 528 cores' worth of taps would bring little benefit with a conventional algorithm Head-Fi, Jul 2018. The next day, in the Hugo TT 2 thread and discussing vendors' tap-length claims (the post cites Audirvana's user guide), he said "I have very limited experience with this" and that he is "very sceptical about the tap length claims made by some vendors" Head-Fi, Jul 2018. On 2 August 2018 he wrote that those who claim it can be done in a PC "don't actually deliver the real tap length that is needed, nor the algorithm that is sinc like in it's coefficients" Head-Fi, Aug 2018.

He did not say a PC could never reach M Scaler tap counts. On 3 August 2018 he wrote that "a standard PC would not have the CPU power to do a full 1M taps. But a GPU equipped PC could do it." and then that "it still would not be a WTA algorithm, and that's more important than anything else" Head-Fi, Aug 2018. Earlier, at CanJam RMAF in October 2017, an audience member asked whether he had considered GPUs for his own designs; he answered that his goal was a pocket-sized M Scaler, possible only with an FPGA, and that a GPU route "is a bit complicated and it means dedicated users" miceblue425 43:39.

On whether other filters tell you what an M Scaler sounds like, December 2018 (Qutest thread): "You can't get an assessment on the sound of the M scaler based upon other implementations, as they are not WTA." Head-Fi, Dec 2018. In January 2019: "All the up-samplers I have seen either have insufficient taps, or do not have WTA like algorithms" Head-Fi, Jan 2019. In December 2019, answering a question about digital level changes in the M Scaler thread, he called keeping the source bit perfect "my standard advice", adding that any upsampling filter is not the same as a WTA filter Head-Fi, Dec 2019.

On offline processing, in September 2019 he said "Yes converting it to 16FS offline is doable but a massive concern is file size." He estimated his library would grow about 30 times and lose metadata, "so it's not a practical way forward to actually enjoy music" Head-Fi, Sep 2019.

Late 2019 to 2020: the same advice, an admitted gap, and his own 1M-tap test

In a Passion for Sound interview published 2 January 2020 (recording date not stated; Part 1 of the series was published in December 2019, so he spoke in or before December 2019), asked whether to upsample in Roon or JRiver, he said "they should let the DAC take care of all of that", citing non-WTA algorithms and limited PC processing "unless you're very clever and use the the graphics card" Passion for Sound 15:20. He hedged the listening side: "I haven't bothered listening to these over sampling filters, so this is not based on my listening test." He said owners he respects reported it sounds a lot worse, and singled out Roon and JRiver as "a very simple interpolation filter" Passion for Sound 15:20. He then said "It's just you'd be crazy to use it. But you know, if that's something you want to try, then by all means" Passion for Sound 16:07, and that owners who dislike the WTA filter have the option of these upscale filters, or of not buying Chord products Passion for Sound 17:23.

On 11 November 2020 he described his advice as long-standing and named a change in its basis: "Today of course there are solutions that claim long tap lengths too, so the recommendation on not using up-samplers is losing out on one benefit." Head-Fi, Nov 2020. To test whether algorithm still mattered at equal length, he built 1M-tap rectangular-sinc and Kaiser filters on his M Scaler test rig and compared them with the 1M WTA. These were his own hardware filters, not commercial software. His listening notes rated both below the 1M WTA; he called the rectangular sinc "not fit for purpose", said the 1M Kaiser lost to DAVE's own 164k WTA, and said "the 1M Kaiser sounds similar to what I expect a 20k tap WTA filter would sound like", and concluded "I ended up being more convinced that algorithm is vastly more important than tap length", while adding "I am certain that more musical benefits are to be had by further increasing the tap length." Head-Fi, Nov 2020.

On 27 November 2020 he summarized: "My recommendation has always been to never use conventional up-samplers on my DACs as the WTA filter built into the DAC will outperform any up-sampler." He said the test showed he had underestimated algorithm's importance, and gave his "very strong advice is never to use up-sampling of any type on my DACs" Head-Fi, Nov 2020. Asked which upsamplers that covered: "Absolutely all up-samplers... unless it contains a WTA filter." Head-Fi, Nov 2020. Asked about putting the WTA on a PC, he wrote "As too putting the WTA onto a PC - that process would degrade SQ" and said the project did not interest him Head-Fi, Nov 2020.

2021 to 2023: PGGB and HQPlayer by name

His PGGB statements concern the version he heard. On 23 May 2021 he said PGGB uses an apodizing filter, not sinc, so long taps bring no advantage: "If you like PGGB then fair enough; but you are simply preferring distortions." He added "I have heard offline true sinc filters, and these sound much more like the M scaler; PGGB on the other hand, sounds nothing like ideal sinc or the M scaler." Head-Fi, May 2021. In a post dated about 2 June 2021 (estimated from post ID) he said "I was sent some samples in December" and described them as sounding typical of apodizing filters Head-Fi, Jun 2021.

On 4 June 2021 he allowed that offline processing done appropriately "means you can in practice do a pretty much good approximation to true Whittaker-Shannon interpolation", while naming 64-bit floating point as a further problem: "Floating point creates noise floor modulation" Head-Fi, Jun 2021. On 7 December 2021 he said "64 bit floating point is not accurate enough for transparency" and that "An FPGA based up-sampler has innate advantages - lower RF noise than a PC" Head-Fi, Dec 2021.

His most direct answer to whether software can stand in for an M Scaler is from 6 December 2021 in the Qutest thread: "Using software up-sampling like HQPlayer and PGGB will give you no idea at all of the performance of the M scaler. Millions, or billions of taps using the wrong algorithm simply gives you the wrong result." Head-Fi, Dec 2021. In July 2022: "It doesn't matter if it's 10 taps or 10 billion taps - apodizing is the wrong approach." Head-Fi, Jul 2022.

At CanJam Singapore in March 2023 he said "it's not actually about the taps, it's about the algorithm itself" Eric's Hi-Fi 9:13. He said a drawback of software, "particularly on PGGB, is you can't do it real time", which matters for streaming and video Eric's Hi-Fi 10:28. He limited his PGGB verdict to the original files, said "I've not heard the latest version", and said of HQPlayer "An HQ player I I've got no experience on." Eric's Hi-Fi 11:13Eric's Hi-Fi 12:01.

In August 2023, responding to PGGB's newer claims of no windowing and pure sinc coefficients, he wrote "so I can't state if there is anything wrong with their algorithm because we don't know what it is". He said published PGGB plots were poorer than the M Scaler in transient-timing aspects, and that true pure sinc would mean "a transition band of very much less than 1Hz not the 240Hz shown" Head-Fi, Aug 2023. He said "the early versions sounded poor by comparison to the Hugo M scaler" Head-Fi, Aug 2023. He also stated his constraints as a trade member: "If I were to publicly state that something was better, then the blowback I would get from Chord would be absolutely immense." and "I would never, ever state that a filter sounded worse than mine, if that was not the case." He added "if I were doing an off-line upscaling filter, I would take a very different approach" Head-Fi, Aug 2023.

2023 to 2024: same advice for Mojo 2 and all his DACs

For Mojo 2 in September 2023: "I strongly recommend that Mojo 2 is fed bit perfect signals - do not use up-sampling". He added that an M Scaler on Mojo 2 gives better sound through longer taps, "But the algorithm is very much more important than the number of taps employed." Head-Fi, Sep 2023. In January 2024 he wrote "When using my DACs do not use SRC or upsampling, always ensure bit perfect data is given to the DAC" Head-Fi, Jan 2024. In November 2024, in the Mojo 2 thread and for his DACs generally (he said he had not used Roon): "certainly not let the source mess with PCM via EQ, volume or upsampling" Head-Fi, Nov 2024.

October 2026: Quartet interviews and FFT convolution

Asked on the Distorsiuni Interioare podcast (published 11 October 2026) whether HQPlayer on a powerful PC could match Quartet, he began "obviously I can't talk about other people's solutions because I don't know what they're doing" Distorsiuni Interioare 35:02. He then described FFT convolution's batch-size frequency uncertainty, which he said translates to transient timing error, and 64-bit floating point noise floor modulation Distorsiuni Interioare 35:49Distorsiuni Interioare 36:27. He said "the numbers don't seem to add up in terms of what the computing is capable of doing and what they're actually delivering" Distorsiuni Interioare 37:16, and that the bigger issue is the algorithm and listening-test tuning Distorsiuni Interioare 38:03.

In Soundnews (17 October 2026), when the interviewer said a powerful PC might go above four million taps, he said those claiming more taps use FFT convolution and do not publish their processing sizes: "So I'm very, very skeptical when people say this is millions and millions of taps, because they're not using taps." article, Oct 2026article, Oct 2026. Asked whether he could offer his own filters as software on a powerful computer, he said creating coefficients on his own PC "takes forever", and "So this is why I'm skeptical." article, Oct 2026article, Oct 2026. He set the FPGA's direct FIR filter against FFT convolution, said "The algorithm itself is much, much more important than the number of taps", and added "I don't think you can do a WTA-type solution directly with convolution. Well, actually, you could do it, but it would be really, really difficult." article, Oct 2026.

What the archive does not contain: no listening comparison by Watts of HQPlayer (in March 2023 he said he had no experience of it) or of any PGGB version after the early files, and no statement found that a particular software upsampler or setting is acceptable before a Chord DAC Eric's Hi-Fi 12:01Head-Fi, Aug 2023. His favorable remarks about offline processing are general and name no product: offline 16FS conversion is "doable" but impractical for file size and metadata Head-Fi, Sep 2019, offline true sinc filters he heard "sound much more like the M scaler" Head-Fi, May 2021, and offline processing done appropriately can approximate Whittaker-Shannon interpolation, within the limits of 64-bit floating point Head-Fi, Jun 2021.

Quotes

“my very strong advice is never to use up-sampling of any type on my DACs - just feed them bit perfect data”2020-11-27 Head-Fi, Nov 2020

“Using software up-sampling like HQPlayer and PGGB will give you no idea at all of the performance of the M scaler.”2021-12-06 Head-Fi, Dec 2021

“Today of course there are solutions that claim long tap lengths too, so the recommendation on not using up-samplers is losing out on one benefit.”2020-11-11 Head-Fi, Nov 2020

“The algorithm itself is much, much more important than the number of taps, as the Blackbird illustrates and as the WTA illustrates.”2026-10-17 article, Oct 2026

Dated positions

Caveats and open points

• His PGGB verdicts rest on early sample files (sent December 2020) and published plots; by March 2023 he said he had not heard the latest version.
• He said in March 2023 he had no experience of HQPlayer; his 2021 remark lumps HQPlayer with PGGB on algorithm grounds, not listening.
• His Passion for Sound advice on Roon and JRiver upsampling (published January 2020, recording date not stated) was, in his words, not based on his own listening test but on reports from owners.
• His November 2020 1M-tap Kaiser and rectangular-sinc comparisons used filters he built on his M Scaler test rig, not commercial software.
• In August 2023 he said that as a trade member he would stay publicly silent if a rival filter were better, while never calling one worse unless it was.
• Head-Fi, Jun 2021 carries an estimated date (between 31 May and 4 June 2021).
• The October 2017 RMAF GPU remark answered a question about his own design tools, not about owners running software upsamplers.

Topics: Software and source-side upsampling, Pre-ringing, phase and rival filter designs, Tap length and diminishing returns