Blu Mk2
What it is
In November 2016 he wrote that it "is CD only - it uses the same mech as the original Blu" Head-Fi, Nov 2016. In an October 2017 talk he said "the Blu 2 is the first product That we've got that's using this million tap length filter", which he named the M Scaler miceblue425 24:38. In May 2017 he wrote that Blu 2s were "now getting shipped" Head-Fi, May 2017. In June 2017, describing how he gets his carry-on through airport security, he said he tells a screener it is "a high end CD player" Head-Fi, Jun 2017.
In 2017 "M scaler" names the filter in Blu 2 Head-Fi, Jan 2017Head-Fi, May 2017; from July 2018 it also names the Hugo M Scaler c w 23:19. He says the Blu 2 design came from his Davina ADC project, which started first, and that the code is being used with Davina in M scaler mode Head-Fi, Jan 2017Head-Fi, Oct 2017. Blu got the scaler first because Chord still made it and "Dave had a longer tap length than Blu 1" Head-Fi, Jan 2017.
He first heard a 512,000-tap version in late July 2016, the full million in late November, and the final version on Christmas Eve Head-Fi, Jan 2017; his 2017 review gives August and December 2016 Head-Fi, Dec 2017.
Design
The filter has 1,015,808 taps Head-Fi, Aug 2017Head-Fi, Oct 2017 in a single 16FS stage Head-Fi, Nov 2017. It takes 44.1 or 48 kHz to 705.6 or 768 kHz Head-Fi, Aug 2017; in January 2017 he wrote that it can output all multiples of the incoming rate Head-Fi, Jan 2017. Its coefficients match an ideal sinc to better than 16 bits (August 2017) Head-Fi, Aug 2017, or 16.6 bits in March 2018, "assuming my spreadsheet is right" Head-Fi, Mar 2018. The window forces coefficients beyond 1.4 seconds to zero Head-Fi, Oct 2017.
The FPGA is the Xilinx 7A200T, running 528 DSP cores in parallel Head-Fi, Jun 2017Head-Fi, May 2017Hifistatement 0. A clip Hifistatement published in October 2017 (recorded earlier at Chord's Pump House, recording date not stated) has 208 MHz, 740 cores in the device, one coefficient ROM and data memory per eight cores, and "97 percent or ninety odd percent" memory efficiency Hifistatement 0Hifistatement 45.
Designing Blu 2's output truncators, he compared 768 kHz noise shapers "guaranteeing 350 dB performance" with Gaussian and TPDF dither; the noise shaper sounded best Head-Fi, Feb 2018Head-Fi, Oct 2016. In December 2019, in the Hugo M Scaler thread, he called the output "very aggressively 11th order noise shaped" and said he arrived at it in listening tests "When the M scaler was first put together with the Blu 2" Head-Fi, Dec 2019.
A video mode cuts delay from 0.66 seconds to 100 mS Head-Fi, Jan 2017, or from 630 mS to 104 mS with an asymmetric filter of 2/3 million taps, and "it's surprising how much degradation you get" Head-Fi, Aug 2017; in June 2017 he wrote that against 2/3 million taps the full million made the sound-stage open up tremendously Head-Fi, Jun 2017.
He kept the FPGA out of DAVE, as its ground current is "too signal dependent and large at 10 A" Head-Fi, May 2017Head-Fi, Jan 2017; a clip Hifistatement published in October 2017 (recording date not stated) has 10 A processing against 2 A idle Hifistatement 0. His first prototype had only RF filtering, the second added isolation, and the production unit better RF filtering on the isolation supply, each audible Head-Fi, May 2017.
Inputs, outputs and power
It has USB and BNC coax inputs (May 2017) Head-Fi, May 2017. In 2018 to 2020 he said its USB uses VBUS only as an RF-filtered logic flag, never needed isolation, and has correct TX and RX termination Head-Fi, Sep 2018Head-Fi, Jul 2018Head-Fi, May 2020. In 2018 he said DSD goes through the Hugo 2 DSD filter, greater than 200 dB at 88.2 kHz Head-Fi, May 2018Head-Fi, Aug 2018.
In November 2016 he wrote that optical reaches only 176.4 kHz, so 705.6 kHz needs dual BNC, RF isolated with options for full galvanic isolation Head-Fi, Nov 2016. By December 2018 he called the BNC output galvanically isolated, the grounds joined by 5pF Head-Fi, Dec 2018. DAVE takes it on BNC 1 and 2 or 3 and 4 Head-Fi, Feb 2017; Hugo 2 accepts 768 kHz dual BNC Head-Fi, Aug 2017.
In May 2017 the FPGA drew 10 A normally and 2 A in a low power mode switching DSP cores off at digital silence Head-Fi, May 2017. In February 2018 he gave a 12 W peak at 1 V and about 15 W from the mains for "the M scaler" Head-Fi, Feb 2018. The first Blu 2, with 512,000 taps, had 6 A regulators; the million taps needed 12 A Head-Fi, Jan 2017. In March 2018 the die reached 60 deg C at full output, mounted to the thick aluminum chassis Head-Fi, Mar 2018. Standby affects only the CD mechanism Head-Fi, Oct 2018. Insertion loss is -2.766 dB, so he raised DAVE 3 dB Head-Fi, May 2017Head-Fi, May 2017.
How he says to use it
Connect DAVE by dual BNC Head-Fi, Nov 2016Head-Fi, Feb 2017 (see How do I connect the M Scaler to DAVE, and why does dual BNC only…, Which Chord inputs are galvanically isolated, and which setups…). On Blu to DAVE, ferrites with higher impedance at 2.5 GHz sounded marginally better, and he used HF ferrites at the DAVE end and lower-frequency ones toward Blu 2 Head-Fi, Nov 2017Head-Fi, Dec 2017. In December 2018 he said clip-on ferrites work with Blu 2, longer cables help, and beyond 2m gives no significant benefit Head-Fi, Dec 2018 (see Which BNC cables, and what length, should I use between an M…). USB ferrites did nothing for him Head-Fi, Dec 2017 (see Do digital cables (USB, coax, glass or plastic optical) make a…). In February 2018 he wrote that the BNC input is "just another input" to "the M scaler", without naming the product Head-Fi, Feb 2018. For Blu-ray he sets a negative JRiver delay to keep the full million taps Head-Fi, Jan 2017Head-Fi, Aug 2017.
Compared with other Chord models
In April 2018 he listed taps: DAC64 (1999) 1,024, Hugo (2013) 26,000, DAVE (2015) 164,000, Blu MKII (2017) 1,015,808 article, Apr 2018, against 49,152 on Hugo 2 and Qutest, "a 20 times capacity advantage" article, Apr 2018. In February 2018 he wrote that "Dave is completely unbeatable in terms of transparency", while Blu2 with Hugo2 "wins in musical performance" Head-Fi, Feb 2018.
Of the Hugo M Scaler he said in 2018 that the filter code is identical, with improved isolation on the BNC outputs removing the need for ferrites Head-Fi, Aug 2018c w 30:37; that it "started as a copied project from the blu 2" Head-Fi, Aug 2018; and that Blu 2's benefit is playing CDs Head-Fi, Aug 2018.
Release and official documents
- Unveiled at CES, January 2017. Source: Poly, Hugo 2, and Blu Mk. II Announced at CES 2017, Chord Electronics, 2017-01-06.
- Shipping from May 2017. Source: BLU MKII upscaling digital/CD transport now shipping, Chord Electronics, 2017-05-04.
From Chord Electronics. Links open Chord's own copies; reviews hosted on Chord's site are left out. Chord's Blu Mk2 page.
Manuals and guides
Drivers (Windows)
Key figures
His own figures, dated. Where he gave different figures at different times, each appears.
Show all 36 figures
| Figure | Value | When | Source |
|---|---|---|---|
| WTA filter taps (production) | 1,015,808 ("just over a million" in an October 2017 talk) | 2017-08 | Head-Fi, Aug 2017Head-Fi, Oct 2017miceblue425 24:38 |
| WTA filter taps (first version) | 512,000, first heard late July 2016 (January 2017 account; in December 2017 he said August 2016) | 2017-01 | Head-Fi, Jan 2017Head-Fi, Jan 2017Head-Fi, Dec 2017 |
| Filter stage | single 16FS stage with around 1 million real FIR taps | 2017-11 | Head-Fi, Nov 2017 |
| Input to output rate | 44.1 or 48 kHz in, 705.6 or 768 kHz out | 2017-08 | Head-Fi, Aug 2017 |
| Output rates | all multiples of the incoming sample rate, selected for the DAC | 2017-01 | Head-Fi, Jan 2017 |
| Coefficient match to ideal sinc | better than 16 bits (2017-08); 16.6 bits, "assuming my spreadsheet is right" (2018-03) | 2018-03 | Head-Fi, Aug 2017Head-Fi, Mar 2018 |
| Reconstruction accuracy | better than 16 bit, or 99.9985% (not 100%), stated for "the M scaler" | 2018-01 | Head-Fi, Jan 2018 |
| Filter window | coefficients beyond 1.4 seconds forced to zero | 2017-10 | Head-Fi, Oct 2017 |
| Filter delay at 48 kHz | 0.661 s delay while the filter processes 1.323 s of music | 2017-10 | Head-Fi, Oct 2017 |
| Delay, video mode off | 0.66 s | 2017-01 | Head-Fi, Jan 2017 |
| Delay, video mode off | 630 mS | 2017-08 | Head-Fi, Aug 2017 |
| Delay, video mode on | 100 mS | 2017-01 | Head-Fi, Jan 2017 |
| Delay, video mode on | 104 mS (asymmetric min phase filter, 2/3 M taps) | 2017-08 | Head-Fi, Aug 2017 |
| FPGA | Xilinx 7A200T (XC7A200T in a clip published October 2017, recording date not stated) | 2017-06 | Head-Fi, Jun 2017Hifistatement 0 |
| DSP cores used | 528 running in parallel (May and June 2017); 528 processors at 208 MHz out of 740 DSP cores (Hifistatement clip published October 2017, recording date not stated) | 2017-06 | Head-Fi, May 2017Head-Fi, Jun 2017Hifistatement 0 |
| Memory efficiency | "97 percent or ninety odd percent" (Hifistatement clip published October 2017, recording date not stated) | 2017-10 | Hifistatement 45 |
| FPGA current, normal and low power mode | 10 A in usual operation, 2 A in low power mode (DSP cores off at digital silence) | 2017-05 | Head-Fi, May 2017 |
| FPGA current, processing and idle | 10 A processing, 2 A doing nothing (Hifistatement clip published October 2017, recording date not stated) | 2017-10 | Hifistatement 0 |
| FPGA power | 10 A peak at 1 V, 10 W maximum | 2017-01 | Head-Fi, Jan 2017 |
| FPGA power | 10 or 12 A at 1 V, a peak of 12 W | 2018-02 | Head-Fi, Feb 2018 |
| Mains draw | about 15 W, stated for "the M scaler" | 2018-02 | Head-Fi, Feb 2018 |
| Core regulator current | 6 A with 512,000 taps; 12 A for the 1M design | 2017-01 | Head-Fi, Jan 2017 |
| Die temperature at full output | 60 deg C, with 12 W dissipated in a 12 mm by 12 mm die | 2018-03 | Head-Fi, Mar 2018 |
| Optical output limit | 176.4 kHz | 2016-11 | Head-Fi, Nov 2016 |
| Filter insertion loss | -2.766 dB (DAVE set +3 dB, net 0.234 dB louder) | 2017-05 | Head-Fi, May 2017 |
| Insertion loss | 2.76 dB | 2017-05 | Head-Fi, May 2017 |
| DAVE volume headroom with Blu 2 | no clipping even at +7 dB; clipping possible only at +8 dB | 2017-08 | Head-Fi, Aug 2017Head-Fi, Jan 2018 |
| Output truncator noise shaper | 768 kHz noise shapers "guaranteeing 350 dB performance", preferred over Gaussian and TPDF dither | 2018-02 | Head-Fi, Feb 2018 |
| Output noise shaper test | reproduces a -301 dB signal in simulation; better than 80 bit up to 15 kHz | 2017-06 | Head-Fi, Jun 2017 |
| Output noise shaper order | 11th order truncation noise shaping at 768k, -301 dB accuracy, devised when the M scaler was first put with Blu 2 (stated in the Hugo M Scaler thread) | 2019-12 | Head-Fi, Dec 2019 |
| Prototype truncation test | 54 bits to 24 bits at 768 kHz; noise shaping sounded best against Gaussian and triangular dither | 2016-10 | Head-Fi, Oct 2016 |
| DSD filter rejection | greater than 200 dB at 88.2 kHz (Hugo 2 DSD filter, used on Blu 2) | 2018-05 | Head-Fi, May 2018Head-Fi, Aug 2018 |
| BNC output ground coupling | 5 pF | 2018-12 | Head-Fi, Dec 2018 |
| Code size | well over half a million lines | 2017-01 | Head-Fi, Jan 2017 |
| Tap length comparison | Blu MKII 1,015,808 against 49,152 on Hugo 2 and Qutest, "a 20 times capacity advantage" | 2018-04 | article, Apr 2018 |
| Sinc match, Hugo 2 / Dave / M scaler | 12.2 bits / 13.9 bits / 16.6 bits | 2018-03 | Head-Fi, Mar 2018 |
Quotes
“I have put a low power mode into Blu 2 - so when there is digital silence, the DSP cores are switched off.”2017-05-06 Head-Fi, May 2017
“Chord still make the Blu (it's very popular in China) and it absolutely needed to be upgraded ASAP as Dave had a longer tap length than Blu 1”2017-01-08 Head-Fi, Jan 2017
“the Hugo M scaler project started as a copied project from the blu 2. It was only control code that was changed, the core audio signal processing is identical.”2018-08-13 Head-Fi, Aug 2018
“I guess if I had to put it in a simple sentence it would be it just simply sounds real - and everything else sounds wrong.”2017-05-01 Head-Fi, May 2017
Caveats and open points
- Name drift: in 2017 posts "the M scaler" means the filter in Blu 2 (for example Head-Fi, Jan 2017, Head-Fi, May 2017); from July 2018 the same words also name the separate Hugo M Scaler (c w 23:19). Posts from before mid-2018 that say only "M scaler" (Head-Fi, Mar 2018, Head-Fi, Feb 2018, Head-Fi, Jan 2018, Head-Fi, Jun 2017) are read here as Blu 2, which this archive does not state outright for each one. Head-Fi, Feb 2018 (February 2018) also says only "the M scaler"; the product it means is unclear, so the BNC remark is given with that hedge.
- Tap count: 1,015,808 in Head-Fi, Aug 2017 and Head-Fi, Oct 2017; "just over a million" in the October 2017 transcript; 512,000 for the first version (Head-Fi, Jan 2017).
- Development dates differ: the 512,000-tap version first heard late July 2016 (Head-Fi, Jan 2017), "over the summer" (Head-Fi, Jan 2017), August 2016 (Head-Fi, Dec 2017; Head-Fi, Nov 2016 says the prototype worked in August), and on his home theater since September (Head-Fi, Jan 2017); the full million heard late November with a final version on Christmas Eve 2016 (Head-Fi, Jan 2017), finished in December (Head-Fi, Jan 2017, Head-Fi, Dec 2017), or working on Christmas Eve (MajorHiFi 22:30).
- Time the filter spans: 1.4 seconds (Head-Fi, Oct 2017, Head-Fi, Oct 2017, miceblue425 34:01), 1.323 seconds processed with 0.661 seconds delay at 48 kHz (Head-Fi, Oct 2017), 1.37 seconds (Head-Fi, Mar 2018), and 1.25 seconds or "one and a half seconds" in a transcript of a clip published October 2017, recording date not stated (Hifistatement 4:30). He calls his RT60 explanation for the depth change a suggestion he is skeptical of (Head-Fi, Oct 2017).
- Delay and video mode: 0.66 s and 100 mS (Head-Fi, Jan 2017) against 630 mS and 104 mS (Head-Fi, Aug 2017). His JRiver offset is -0.56 s in January 2017 (Head-Fi, Jan 2017) and -630 mS in August 2017 (Head-Fi, Aug 2017). Head-Fi, Jan 2017 ties the video switch to the SPDIF input; the clip published October 2017 (recording date not stated) ties it to the USB input (Hifistatement 2:57).
- Sinc-match figures changed: DAVE about 13 and 13.5 bits and Hugo about 12 (transcript of clips published October 2017, recording date not stated, Hifistatement 24, Hifistatement 1:12), Hugo 2 about 13 bits (Head-Fi, Aug 2017), against 12.2 for Hugo 2 and 13.9 for Dave in March 2018 (Head-Fi, Mar 2018), which he hedged "assuming my spreadsheet is right".
- Isolation prototypes: Head-Fi, May 2017 describes three units (no isolation, then isolation, then an improved production unit); the February 2018 seminar says two prototypes (MajorHiFi 20:49). In November 2016 the BNC outputs were "RF isolated" with options for full galvanic isolation (Head-Fi, Nov 2016); from March 2017 he describes the production unit's BNC outputs as galvanically isolated (Head-Fi, Mar 2017, Head-Fi, Feb 2018, Head-Fi, Dec 2018).
- Inputs: in November 2016 he wrote that Blu mk2 is CD only (Head-Fi, Nov 2016); in January and May 2017 he described BNC and USB inputs (Head-Fi, Jan 2017, Head-Fi, May 2017). The CD-only remark appears to concern the disc mechanism; the post does not show the question.
- Power: the 15 W mains draw is stated for "the M scaler", with the isolation advice said to apply to Blu 2 and Dave (Head-Fi, Feb 2018). The 6 A regulator belongs to the first Blu 2 with 512,000 taps (Head-Fi, Jan 2017).
- Headroom: in November 2018 he wrote of the Hugo M Scaler "it's exactly the same as blu 2, with the 3db headroom" (Head-Fi, Nov 2018), consistent with the 2017 insertion-loss figures (Head-Fi, May 2017, Head-Fi, May 2017). The 220 dB DSD stop-band figure (Head-Fi, Aug 2018) describes his DACs' DSD-to-PCM conversion in general and is not used as a Blu 2 figure; the Blu 2 figure is Head-Fi, May 2018.
- Head-Fi, May 2018 also says the M scaler always outputs 705.6 kHz PCM from DSD64 to DSD512; that sentence does not name which M scaler, so it is not used as a Blu 2 figure.
- Prototype figures: the 54-bit truncation test is from an October 2016 prototype (Head-Fi, Oct 2016), not a production specification. The 11th-order noise shaper figure comes from a December 2019 post in the Hugo M Scaler thread (Head-Fi, Dec 2019).
- The DAVE comparison (Head-Fi, Feb 2018) is a February 2018 listening judgment about Blu2 with Hugo2 against DAVE.
Questions about this product
Owning and setting up
- Do lossy or streamed files sound worse than local files on a Chord DAC?
- Does an M Scaler help Mojo, Mojo 2 or a non-Chord DAC?
- How many taps and pulse array elements does each Chord DAC have?
- Is M Scaler pass-through bit perfect, and why is the scaled output quieter?
- Should I put ferrites on my BNC or USB cables, and which kind?
- What does the M Scaler actually do, and is it worth adding to my DAC?
- Why does my M Scaler give no output with native DSD?
In depth
- Is the filter algorithm more important than the tap count, and would more than a million taps keep improving the sound?
- Should audio processing dither or noise-shape, and is 64-bit floating point transparent?
- Does hi-res audio sound better than CD, and should I buy the 96k or the 44.1k release?
- How do Chord DACs handle DSD, and does Rob consider DSD better or worse than PCM?
- How were Chord's current DACs and the M Scaler developed, and how long did each take?
- What does Rob mean by depth, and what in a DAC improves or collapses it?
- What is Davina, and what happened to the pulse array ADC?
- Why does Rob design to -301 dB, how can that be verified, and how does small-signal accuracy relate to depth?
- Why does tap length matter, and why a million taps?