Hi-fi stands for high-fidelity. Specifically, the high fidelity of a reproduced audio signal compared to its original source. Recording and reproducing sound introduces artifacts, and your listening environment has an effect as well. So, playing back recorded audio never sounds exactly the same as the original. You can get pretty close, though.
McIntosh MC275 50th Anniversary: A $6500 amplifier with no DAC capabilities
Hi-fi often is often associated with exotic (and expensive) equipment. Tube amplifiers. Silver cables. Gold-plated interconnects.
Yet, unless you own a dozen shelves of 180-gram vinyl records, most of your hi-fi audio is probably stored in an affordable digital format, either on optical media (like CDs, DVDs, SACDs, and LaserDiscs) or magnetic storage, in the form of files on your hard drive.
In order to play back that content, you need a few different components. You need something to access the information (a CD reader, perhaps). You need to convert the digital signal back into an analog one using something called a Digital to Analog Converter, or DAC. You need an amplifier. And finally, you need something to create sound pressure waves in the air around you at the right frequencies set by the signal (speakers or headphones).
As long as the content arrives to the DAC in a bit-perfect state, the source really doesn't matter. A quality CD-player or a PC playing a bit-perfect stream over USB should sound the same.
The challenge for us as PC enthusiasts is that uncompressed audio takes up a lot of disk space. A CD-quality stereo stream uses two (channels) x 16 (bits per sample) x 44,100 (samples per second) = 1411.2 Kb/s A 60-minute CD, uncompressed, ties up 635 MB of storage. That was a ton back in the days of gigabyte drives, 1 Mb/s Internet connections, and slow Wi-Fi. The solution was lossy compression in the form of MP3 (first) and AAC (later), which addressed the capacity issue with a quality compromise deemed acceptable by most consumers. But audiophiles balked at the idea.
Creative Labs' Sound Blaster 2.0 from 1991, the first PC audio card capable of 44.1 kHz playback
Of course, today we enjoy multi-terabyte drives, very fast broadband connections, and almost gigabit-class wireless data rates, and advanced lossless compression schemes like FLAC and ALAC, the latter of which can halve the size of an audio file with no quality loss whatsoever. And so, the story changes.
Suddenly, a $60 1 TB hard drive can store 3000 CDs at their native quality. That's a lot of shelf space saved. Buying and downloading a new disc takes minutes, at most. Finding an album or track in your collection happens quickly. Online stores are never out of stock. And if you back up your library, it will never get lost or degrade. What's not to like?
One part of the pipeline that remains constant, and where PCs traditionally lag, is the translation from digital source to actual sound. Thankfully, hi-fi devices natively supporting PCs are becoming increasingly common. And if the quality of more traditional hi-fi equipment can be matched, then a case can be made (given overwhelming convenience) for our PCs becoming the ultimate audio source.
But what are the options for hi-fi audio on a PC today, and at what price points? In today's story, we're looking at the differences in sound quality, features, and value of a few pieces of hardware able to turn your system in the ultimate hi-fi machine. In the process, we'll introduce you to blind listening tests done right (at least in our view), and why that's so important.
Four different devices are on the bench, ranging from $2000 all the way down to $2: the Benchmark Media Systems DAC2 HGC, JDS Labs' O2+ODAC, Asus' Xonar Essence STX, and Realtek's ALC889 multi-channel codec. That's a 1000x factor in cost.
All of our tests were run on a PC with Windows 7 x64. For the most part, the machine's specs aren't really relevant, except for the fact that the motherboard hosting our Realtek ALC889 codec is Asus' Rampage III Formula.
The following devices are included, covering price points from $2000 down to $2.
Foobar2000 sources, with the WASAPI sources circled in red
| Benchmark Media DAC2 HGC | JDSLabs O2+ODAC | Asus Xonar Essence STX | Realtek ALC889 | |
|---|---|---|---|---|
| Price | ~$2000 | ~$290 (including AC adapter) | $190 | ~$2 (OEM in volume) |
| Format | External USB Device | External USB Device | Internal PCIe card | On-board codec |
| Driver version | 1.61 | Native USB Audio | 7.12.8.1794 | 6.0.1.7023 |
| DSP Chip | Custom FPGA | N/A | Asus AV100 (C-Media CMI8788) | Realtek ALC889 |
| DAC Chip | ESS SABRE32 ES9018 | ESS SABRE32 ES9023 | TI PCM1792A | Realtek ALC889 |
| I/V conversion | 2 x TI LME49860/LME4562 | 2 x JRC NJM4556AD | 2 x TI LME49860/LME4562 (1) | Realtek ALC889 |
| Buffer | TI LME49600TS | Custom 4 x NJM4556D stage | TI TPA6120A2 | Realtek ALC889 |
| PCM audio support | Up to 176.4/192 kHz sampling Up to 24-bit word length | Up to 96 kHz sampling (2) Up to 24-bit word length | Up to 192 kHz sampling (3) Up to 24-bit word length | Up to 176.4/192 kHz sampling (4) Up to 24-bit word length |
| DSD audio support | Natively supported via DoP | Not supported | Not supported (5) | Formally supported, but couldn't get it to work |
Notes:
- The STX has swappable socketed op-amps. We replaced the native 2 x JRC2114D (which are also very good) with 2 x TI LME49860/LME4562, the same ones used in the DAC2.
- Does not support 88.2 kHz
- Does not support 88.2/176.4 kHz
- Did not support 88.2 or 176.4 kHz in our test at 16- or 24-bit depth
- The PCM1792A DAC does support DSD, but unfortunately, the C-Media CMI8788 does not, hence the lack of this capability on the card
Despite the higher product series number, the ESS Sabre ES9023 DAC is, on paper, inferior to the ES9018. That doesn't imply an audible difference, though.
Few devices are consistently praised in the audiophile community. The Benchmark DAC1 is one of the chosen few. You'll have a hard time finding someone with critical feedback about that device. And, although personal preferences and arguments over value are rife, it really is the reference for a high-end DAC and headphone amplifier combination.

In October of 2012, Benchmark Media released its DAC2 HGC to weighty expectations. At $2000, it's certainly not affordable (a DAC1 HDR, comparable in features, still goes for $1600). But to Benchmark's credit, aside from the headphone amplifier, which is the same HPA2 found on the DAC1, the DAC2 is an entirely new device. It leverages what is one of the world's highest-end DACs, the ESS ES9018, adding to it, among other things, custom jitter-reduction logic. It lives up to its Hybrid Gain Control name by implementing separate volume controls: digital for digital inputs; analog for analog inputs.

The Benchmark DAC2 HGC is the Cadillac of this round-up. It includes many features I'd imagine are generally useful to PC enthusiasts:
- Asynchronous USB input: up to 172.4/192 kHz at 24-bit PCM, plus native DSD64 support
- Four S/PDIF digital inputs (two coax, two optical)
- Two RCA stereo single-ended analog inputs
- Two RCA stereo single-ended analog outputs
- One XLR stereo balanced analog output
- Two front-panel stereo TRS (1/4") headphone jacks
- Input selection, word-length, and word-clock display on front panel (finally!)
- A remote control commanding a motor-driven actuator attached to the master volume control
- Polarity and dim/mute buttons, and a 12 V trigger (less commonly used)
Both front-panel headphone jacks can be active concurrently, without any signal degradation. The left headphone output jack mutes the back-panel analog output, while the right headphone output jack does not. This is a simple (but incredibly useful) feature that lets you mute (or not) your speakers by picking the appropriate jack for your headphones.
The DAC2 HGC operates as a USB Audio Class 1 device by default, which means that it doesn't require driver support for Windows and Mac compatibility. It can be manually switched to operate as a Class 2 device, necessitating a driver in Windows, which is included. The main reason to switch to UAC 2 is to play PCM files above 24-bit/96-kHz, DSD files, or if you need an ASIO driver for any reason. If none of those apply, there's no reason to change modes.
More affordable versions of the DAC2 HGC do exist. There's a DAC2 D without analog inputs and a DAC2 L with analog inputs, but without the headphone amplifiers. Both models are $200 cheaper at $1800. PC enthusiasts may look favorably at the DAC2 D, since it's unlikely that you'd need the analog inputs and that device supports concurrently multiple sets of both speakers and headphones, whereas the DAC2 L does not support headphones.
In case you're wondering, the DAC2 reflects exceptional build quality. That's something you'd no doubt expect at this price point, but it's still an important point to confirm. There's also a bundled remote that, while not terribly useful in a PC environment, is still a nice touch.
The O2+ODAC that JDS Labs (among others) manufactures, based on an open source design, is the most innovative concept in our round-up.
In 2011, a mysterious blogger who used the handle "NorthWest Audio/Video Guy" (NwAvGuy) began ranting about the snake oil he believed was being sold to the audiophile community. He set out on a personal crusade to design and build a low-cost headphone amplifier that, through blind tests, could not be distinguished from the Benchmark DAC1.
That design eventually became what's known as the "O2" (from Objective2) headphone amplifier. NwAvGuy went on to create an implementation of ESS' ES9023 chip to function as a DAC feeding the O2, using similarly objective criteria. That latter device came to be known as the "ODAC" (from ObjectiveDAC). Both devices can easily be connected and integrated into a single enclosure.

The O2+ODAC is an uncommon design by most standards. The headphone amplifier, in particular, was originally designed with battery-powered operation in mind (for portability). While the JDS Labs implementation does away with the batteries, it retains many of the portability-related design choices. It employs an external AC transformer, not the internal one you'd find on most comparable DAC/amps. There are no RCA stereo outputs at all. And the O2 does not use an amplifier chip (like the TPA6120A2), but rather a custom design on the output stage.

Feature-wise, the O2+ODAC is really barebones. It has a 3.5 mm line-in connection up front, a mini-USB port in the back, and a 3.5 mm headphone-out jack. That's it as far as I/O goes. Controls are limited to a gain switch and an analog (high-quality) volume control.
Based on NwAvGuy's open-source license (which, paraphrasing, states: anyone and everyone can manufacture one of these without paying me a dime, as long as they don't change the design), a variety of manufacturers are now selling O2+ODAC devices.
In 2012, after being banned from the headfi.org forum for, according to his version of the story, criticizing one of that forum's sponsor's products, NwAvGuy started building a desktop-oriented version of the O2+ODAC, called the Objective Desktop Amplifier. The twist is that NwAvGuy mysteriously disappeared without a trace in mid-2012 before completing the ODA's design. Nobody seems to know why he stopped blogging or what happened to him. So we're left with the O2+ODAC to test, and a lingering dream of what the ODA could have been.
The manifestation we're testing today is provided by JDS Labs. It is sold fully assembled for roughly $290, including the required AC transformer. If you're nimble enough with a soldering iron, you can pick up the O2 do-it-yourself kit for $69, add the ODAC board for $99, and buy the transformer for $11. That'd get you going for about $180, not including the enclosure.
Aside from the performance commentary you'll find through the following pages, I encountered one specific problem with the O2+ODAC. After receiving and unpacking it, I clicked into high-gain mode, plugged in my Sennheiser HD 800s (300 Ω impedance), turned up the volume, and noticed that the sound was terrible, affected by massive distortion. It turns out that the supplied transformer isn't powerful enough to drive high-impedance phones using the high gain setting. Unfortunately, JDS Labs doesn't stock higher-power transformers, so a replacement wasn't an option. The only solution was to use the low-gain setting at much higher volume. That did successfully solve the distortion issue. But JDS Labs should consider, in my opinion, stocking AC transformers that better-support the high gain setting. Going one step further, transformers should really be included with the assembled product.
So, did NwAvGuy end up winning his crusade? Can the O2+ODAC be distinguished from the 7x-more expensive and 5x-larger Benchmark DAC2? Read on...
Update: We received the following response from JDS Labs:
"AC adapters for O2 are packaged separately because JDS Labs ships worldwide. There's little incentive to bundle AC adapters since each customer requires a unique plug. Thus, the items are presented separately for customer's selection. Our shopping cart reminds customers to choose an appropriate model. As of late December, we now stock a higher-power model for U.S. customers who need additional current; this represents less than 2% of customers, though. All European, Australian, and British AC adapters stocked by JDS Labs are high-power models."
In May of 2009, Asus launched its Xonar Essence ST. The STX, which is almost identical except for its PCI Express interface (and a few other minor differences, notably the lack of daughterboard support for eight-channel audio), followed shortly thereafter.
The Xonar Essence ST(X) was one of the first cards designed specifically to support high-end headphones and bridge the gap between PCs and the high-end audiophile world. Most notably, it included 1/4" TRS connectors, which are rare on add-in cards. The Xonar ST(X) totally came out of left field; nobody imagined a motherboard manufacturer taking that route.

The STX is still sold today at close to its full launch price. How many other sound cards from 2009 can boast such a claim? Asus' design was presumably so effective that when the company introduced its Xonar Essence STU in October of 2013 (almost five years later), it pretty much mirrored the implementation, adding an external enclosure, two volume controls, and some other minor stuff (including 49720 op amps, which aren't popular on the STX, but might have been implemented differently). The Essence One uses somewhat higher-quality components and includes a few more features. But the STX must have been something special, since the Essence One is no longer for sale and the Essence STU doesn't appear to have much momentum.
So, what makes the Xonar Essence STX so successful?

Red: LME48960 op amps (used and spare), Green: stock JRC2114D + LM 4562 (unused), Blue: LME49720 (unused), Pink: TI PCM1792A DAC, Orange: PLX PCIe-to-PCI bridge, Purple: AV100/C-Media AV8788 DSP, Yellow: TI TPA6120A2 headphone amp, Gray: ADC section
One of its perceived strengths is the flexibility to manually swap out the operational amplifiers. Although many would argue that the real-world benefit of doing this is close to non-existent (in fact, you're more likely to make the card operate at a lower fidelity if you swap out the stock JRC2114D op amps which the card was designed for), tinkerers love the option. Tweaking a sound card, since overclocking it doesn't make much sense, resonated with the enthusiast community.
More important is the Essence STX's reportedly well-reviewed sound quality paired to high-end headphones. Asus' implementation of the once-top-of-the-line Texas Instruments PC1792A DAC and TPA6120A2 amplifier was well-received.
And then there's pricing. The Essence One ($600-$1000) and Essence STU ($400), while likely sounding similar, lacked the STX's value proposition at $190.
Because it used PCI Express, Asus' Xonar Essence STX didn't need a dedicated (and expensive) power supply. Except for a three-way gain setting, it also didn't need volume controls. It didn't need an external enclosure or the cables to connect it. The company did choose to add analog inputs and an ADC, though I doubt many of the enthusiasts who bought an STX valued that feature.
If Asus decided to create cheaper and higher-quality sound cards, rather than pricey external DACs, I think it'd win over more converts. A second version, made more affordable by cutting the ADC stages and equipped with a newer-generation DAC, would be great. It's also conceivable that enthusiasts would pay for an optional front-panel display with a quality 32-bit volume control and LCD.
In addition to its work in the networking space, Realtek has a significant share of the integrated audio market, too. The company sells a variety of codecs with different feature sets. The Rampage III Formula motherboard I'm using comes equipped with an ALC889, so that's the multi-channel codec I'm testing alongside the discrete solutions.

Except for the very similar ALC898 and technically better ALC1150, neither of which is listed in the table below, Realtek's ALC889 is pretty much top-of-the-line. Beyond sporting the most advanced specifications, it's also Realtek's only codec with support for DSD (though we couldn't find a suitable ASIO driver to get it working with foobar2000).
You can purchase the ALC889 in volume as an OEM for an indicative price of ~$2 per chip (or less, depending on the volume ordered), which means that the cost it adds to your motherboard is probably less than $10. Talk about an indicator of how commoditized the integrated audio market is.

Before you lean on integrated audio, be sure to do a little research into the codec your motherboard includes. Specifically, higher-sounding part numbers aren't always indicative of a better component. For example, the popular ALC892's specifications are inferior to the ALC889.
According to its datasheet, the ALC889 sports headphone amplifiers integrated at six output ports. They drive the Sennheiser HD 800s at 93.6 dB(A), and as such have more than enough power for anything at or below 300 Ω.
As you'll see, the ALC889 appears to be the least hi-fi of the devices we're testing, with a 1.4 dB(A) difference at 100 Hz. It is quite easily distinguishable in a pure-tone comparison at that frequency, although it is much harder to detect in regular music-listening scenarios (as at 1 kHz and 10 kHz the volume difference is much smaller).
We also want to explore this codec's output impedance. At 77 Ω for the recommended implementation, it is by far the highest (almost by an order of magnitude over the second-highest) in our round-up. Is that a factor in the real world?
We're testing using two set of headphones. Because this is a story about high fidelity, they're expensive. If you don't listen to quality music recordings, or don't care about the upper echelon of audio equipment, they're almost certainly overkill. You'd get better value out of more affordable models.
Beyond their price tags, these two headphones are different in many ways.

Sennheiser's HD 800 is the company's flagship headphone. At $1500, it's prohibitively expensive. Quality-wise, however, it's considered one of the best in the world (along with the Audeze LCD-3 at $2000, Grado's PS1000 at $1700, the Fostex TH-900 at $1500, Ultrasone Edition 12 for $1700, HiFiMan's $1300 HE-6, and the $1400 Beyerdynamic T1 Tesla). The $4450 STAX SR-009 also deserves a mention as the best-known commercially available electrostatic headphone. We're sure that the circle of headphone royalty could be argued to include others, but it's safe to say that this list generally almost always qualify for the distinction.
The HD 800 is an open back and circumaural design. In plain English, that means these headphones don't block outside sounds and envelop your ears, resting on the sides of your head. Not surprisingly, then, they're also big, though surprisingly light for their size. Sennheiser rates the HD 800s for 300 Ω impedance, placing them in the high-impedance category of electrically-inefficient headphones that amplifiers have a hard time driving. This can also be a boon too, making them less sensitive to the output impedance of the amplifier itself. Even at 300 Ω, the HD 800's impedance is still half of some Beyerdynamic headphones, which can reach 600 Ω and are some of the hardest headphones to drive.
Sennheiser's top-end headphones connect through a Y-cable with a fixed 1/4" TRS plug. The cable is removable, so if you happen to break it, replacements are available. You can also swap out the stock cable with a more expensive one, though we haven't seen the benefits proven in a reliable blind test.
Besides their audio quality, the one characteristic I personally appreciate about the HD 800s is that their circumaural design and light weight mean they're extremely comfortable over long listening sessions.

We also used the AKG K 550 headphones for a few tests. They're marketed as reference. At $200 (and sounding very nice), they certainly deserve a place in the ranks of headphone nobility. The K 550s employ a circumaural design too, though they're smaller than the HD 800s and don't clear my ears as well. With less padding on the head band, the AKGs are far less comfortable overall, at least for me. AKG employs a single non-detachable cable instead of a Y-cable, which is intended to help with clutter.
These are traditional 32 Ω headphones. They come equipped with a 3.5 mm connector and 1/4" screw-on adapter. So, the K 550 can connect to your portable music player or phone, while the HD 800 is simply too difficult for a mobile device to drive. Another major difference is that the K 550s are based on a closed-back design, yielding certain privacy advantages. You can't really get away with using open-back headphones, say, at the office. Closed-back headphones also dampen outside noise; that can be a boon in relatively noisy environments.
The merits of open- and closed-based designs when it comes to sound quality are much more open to debate. Most top-of-the-line headphones are at least semi-open. Only the Fostex TH-900 is closed-back, and arguably not as popular as some of the other options on our list. Personally, I own affordable closed-back (Sony MDR7506, $85) and open-back (Grado SR80, $100) headphones, in addition to the high-end HD 800. Regardless of price point, when I have the choice, I use open-back headphones; I simply prefer their sound.
The Listening Environment
All of our tests were conducted in a room with a background noise level of 36.5 dB(A) ±0.2. Of course, we had a PC in the room, and the noise we measured was primarily a result of the system's cooling fans. When my machine dropped to standby, the background noise fell to 32.2 dB(A) ±0.2. In other words, we listened in a very quiet room.
With all of the talk about signal-to-noise (SNR) ratios, total harmonic distortion + noise (THD+N) and dynamic range (DR), it's easy to forget that regular listening environments are inevitably subject to quite a bit of background noise. Beyond a certain threshold, increasingly high SNRs and the "N" component in THD+N become audibly irrelevant when the noise floor of your environment is meaningfully higher than the hardware being tested. That's particularly true for open-back headphones, which, unlike closed-back designs, provide practically no attenuation of ambient noise. Check out some of the (non-scientific) tests in the conclusions page to do a bit of related tests directly on your own.
Imagine trying to listen to your favorite CD on the deck of an aircraft carrier. You can't; the background noise level is so high that you actually need hearing protection. That's an extreme of course, but background noise in any environment still affects what we can hear and what we cannot.
Volume Matching and its Importance
Volume-matching sources when blind listening is important for two reasons. First, if sources are at different levels, they're easy to tell apart. From there, the test is no longer blind. Second, us humans tend to prefer (all other factors being equal) louder sources. Again, that's something we want to control.
It's rudimentary but effective. In this image, we're calibrating headphones using a standard SPL monitor.
Using Sennheiser's HD 800, we accurately volume-matched the individual devices using the 100% digital volume and minimum gain setting of the Asus Xonar Essence STX (which, as an add-in sound card, lacks an analog volume control) and a 1 kHz test tone.
Three test tones at 100 Hz, 1 kHz, and 10 kHz were used from mediacollege.com. The 1 kHz reference level is most important; that's the frequency at which human hearing is most sensitive. The devices we're using are rated to be fully linear in the specified range, so calibration values should match across all three tones.
At 1 kHz, all sound sensor weightings, such as dB(A), dB(C), and dB(Z), are exactly the same with a 0 dB gain. Meanwhile, at 100 Hz and 10 kHz, the weightings yield different values. We're using the common A-weighting, which approximates human hearing best in terms of relative loudness of sounds at different frequencies. This goes a long way in explaining why 100 Hz and, to a lesser extent, 10 kHz, measure consistently lower than 1 kHz. The remaining "drop" comes from the HD 800's own frequency response, which is far from linear above 1 kHz.
| Calibration Tone Frequency | Benchmark DAC2 HGC | JDS Labs O2+ODAC | Asus Xonar Essence STX | Realtek ALC889 |
|---|---|---|---|---|
| 100 Hz | 57.0 dB(A) ±0.1 | 57.4 dB(A) ±0.1 | 56.9 dB(A) ±0.1 | 58.3 dB(A) ±0.1 |
| 1 kHz | 93.9 dB(A) ±0.1 | 94.0 dB(A) ±0.1 | 94.0 dB(A) ±0.1 | 93.6 dB(A) ±0.1 |
| 10 kHz | 80.5 dB(A) ±0.1 | 81.0 dB(A) ±0.1 | 80.3 dB(A) ±0.1 | 80.2 dB(A) ±0.1 |
As you can see, the calibration is very good, though not absolutely perfect. The Benchmark DAC2 is not perfectly aligned because it uses a digital gain control to affect the volume of its digital input. This control has roughly 0.5 dB(A) "steps" at the level we tested, compared to the analog potentiometer in JDS Labs' O2+ODAC. Given the DAC2 HGC's higher price tag, I'm giving it a minor handicap and setting it at the rounded-down closest setting to the other devices. Realtek's codec is slightly softer at 1 kHz and significantly louder (1.4 dB[A]) at 100 Hz. In this sense, it's simply the least-linear or least-transparent of the devices we're testing.
Audiophiles might argue that a listening difference of 0.2 dB is notable, and might impact our test results. This might hold true for a small minority of humans. For us, it does not matter. This isn't just claimed; we'll prove it shortly. Furthermore, 0.2 dB approaches our equipment's margin of error. Realtek's 1.4 dB(A) difference at 100 Hz is the one measurement that might be noticeable.
Of course, listening at >90 dB(A) for extended periods of time can cause hearing loss. You'll be fine a few minutes at a time. But maintaining high volume should be avoided.
The Most Important Instrument to Calibrate: You
Because everyone's ear is morphologically different, we each hear sound uniquely. There are some general truths, though. For example, we become progressively incapable of hearing higher frequencies as we age. The typical human hearing range is conventionally referred to as 20 Hz to 20 kHz (sometimes 22 kHz).
Our tests involve two listeners: a moderate enthusiast, Listener A, accustomed to ~$3000 in audio gear, and a more serious enthusiast, Listener B, used to ~$70,000 in audio gear.
| Measurement | Listener A | Listener B |
|---|---|---|
| Highest Frequency Heard | 17 kHz | 20 kHz |
| Lowest Frequency Heard | 12 Hz | 14 Hz |
| Volume Sensitivity (95% Confidence) | ±1 dB | ±1 dB |
At the high end, Listener A can hear a 17 kHz tone using the DAC2. Tones at 18 kHz and above are absolutely silent. Listener B, despite being a few years older, can hear up to 20 kHz.
On the other end of the spectrum, Listener A can faintly hear 12 Hz. Anything lower is total silence. Listener B's hearing starts roughly at 14 Hz. This is uncommon; typically, the threshold is around 20 Hz. Some say such low frequencies are felt, rather than heard. Another possible explanation is harmonic distortion in the headphones or audio equipment. If that was the case, the tone heard at 12 Hz should sound the same as 24 Hz, but softer. But it doesn't. It sounds far lower than the 24 Hz tone.
Using these calibration settings, a blind A/B test of a difference in ±0.5 dB volume levels at 440 Hz results in a score of 5/10 for both listeners, essentially equivalent to a random guess. That means neither participant can tell 0.5 dB levels apart. To reach a 95% confidence level that listeners can tell volume levels apart, we have to move to ±1 dB, where they score 9/10 or 10/10 consistently.
Thus, the "calibration range" of your listeners today is 12 Hz to 17 kHz and 14 Hz to 20 kHz, with a 1 dB volume sensitivity.
Given that the devices we're testing are calibrated well below the level where either listener can hear the volume difference, we consider them accurately volume-matched (except for Realtek's codec at 100 Hz).
For reference, here is the hardware both listeners use:
| Component | Listener A | Listener B |
|---|---|---|
| Primary source / DAC | Asus Xonar Essence STX $190 | Burmester 061 CD Player ~€9000 |
| Power conditioner | None | Burmester 038 (no longer in production) ~€4000 |
| Integrated amplifier | Built into powered speakers | Burmester 032 ~€12,000 |
| Secondary power amplifier (For horizontal bi-amping) | None | Burmester 036 ~€7000 |
| Speakers | Yamaha HS80M + HS10W $900 | Ascendo Z-F3 ~€21,000 |
| Headphone amplifier | Built into Asus Xonar Essence STX | Lehmann Audio Linear SE ~€1400 |
| Headphones | Sennheiser HD 800 $1500 | Sennheiser HD 800 ~€1500 |
| Cables | Budget RCA cables $5 | Burmester/Ascendo cables ~€4000 |
As you can see, Listener A is accustomed to an audio setup worth around $3000. Listener B is in another category altogether, with a configuration well into five figures. Listener A's setup is also a 2.1-channel near-field active-monitor setup, while Listener B's setup relies on high-end full-range speakers. Both listeners are well-acquainted with Sennheiser's HD 800 headphones though, which are what we'll primarily be using for our tests.
Cables and Connections Used
For the O2+ODAC and the DAC2, we used vendor-supplied USB cables.
The Sennheiser and AKG headphones both relied on stock cables.
Four identical Hosa 10-foot TRS female-to-TRS male extension cables were used. For the O2+ODAC and Realtek ALC889 codec, which do not have TRS jacks, a Hosa TRS female-to-3.5 mm RCA male adapter was used. These are not exotic components; the cables are $7 each and the adapters are $3 each.
Here's an important question, though. Why use extension cables at all? Why not plug headphones directly into device jacks?
The reason has to do with our blind listening process. Extension cables essentially render interconnect noise the same (when the headphones are switched from one device to another) and require no moving around on the part of the test assistant performing the switch. It'd be hard to perform a truly blind test if the assistant had to get up, walk around to the back of the PC to plug into a sound card, and so on.
I believe (until I'm otherwise proven wrong) that the cables and adapters are audibly transparent. If you'd like to read a serious study on the effects of cables on audio frequency response, we recommend this article.
Music Formats
We classified sources into three categories:
Compact Disc Digital Audio (CD-DA)
These tracks are encoded with the common pulse-code modulation (PCM) scheme used in standard audio CDs. The format is a two-channel signed 16-bit linear PCM at a 44.1 kHz sampling frequency.
As a bit of trivia, the main reason why CD audio is sampled at 44.1 kHz is that the corresponding Nyquist frequency (the highest frequency that can be captured using the sampling rate) is 22,050 Hz. Hence, CD audio is designed to capture and represent all frequencies humans can hear.
What's more, sometimes this format is referred to as "Red Book" from the color of the binding book containing its technical specification.
The reason why CDs are designed to hold 74 minutes of audio is said to be less technical and more "human"; reportedly, Sony's president Norio Ohga wanted to listen to Beethoven's entire 9th symphony in his car.
Benchmark DAC2 HGC in operation. The LCD tells us it's configured for 16-bit/192 kHz. The smaller O2+ODAC is on top.
DVD-Audio (DVD-A) and DVD-Recordables (DVD-R), including PCM files (typically .WAV)
These are tracks that have higher word length and sampling frequency than "Red Book" CD audio.
While 24-bit recordings are relatively common in the recording studio, they tend to be used more for headroom in editing than an actual audible benefit. We've never seen a properly-conducted blind test where individuals could reliably tell between 24- and 16-bit audio. With that said, "audiophile" recordings made available at the native 24-bit, though not very popular, do exist.
Higher sampling frequencies than 44.1 kHz again provide more headroom for editing; the actual benefit in terms of audio quality is the subject of much debate. The ultrasonic frequencies that such high sampling frequencies collect are inaudible, unless you happen to be a dog (which can hear up to 60 kHz), cat (79 kHz), or a bat (up to 200 kHz). And that's assuming that your speakers/headphones can play those higher frequencies correctly. They certainly aren't designed to. In fact, the harmonics from those frequencies can actually result in lower fidelity.
We sourced our DVD-A files as uncompressed WAVs from www.hdtracks.com.
Super Audio CD (SACD)
Direct-Stream Digital-encoded audio takes a very different approach than multi-bit PCM. DSD has a word length of only one bit and a sampling frequency of 2.8224 MHz. It is a pulse-density modulation scheme. The benefits and limitations of DSD versus PCM are, not surprisingly, the subject of much debate and there is really no consensus, even among academics, on which is superior.
Although DSD content is rare (SACD never really took off), so-called "native DSD" digital-to-analog-converters appear to be popular in the audiophile community. Mytek, in particular, is pushing the format.
Some SACDs contain multi-channel tracks. The most famous one is likely Pink Flyod's The Dark Side of the Moon, which contains six channels on a "hybrid" disc.
Of the devices we're testing, only Benchmark's DAC2 supports DSD natively. So, we'll reserve a separate section of this article to delve into it.
Test tracks
In our choice of test tracks, we wanted to represent a wide variety of genres and musical format options. Obviously, how well a particular track was recorded influences the listening experience massively. So, our selections were picked because, in addition to being enjoyable, they were also recorded well.
| Album / Track / Artist | Format | Comments |
|---|---|---|
| Skyrim Official ST / Dragonborn / Jeremy Soule | CD-DA (16/44.1) | Nothing screams Dovahkiin! like Jeremy Soule's signature soundtrack. Sung by 90 voices at the same time (30 people singing three times then mixed together), this dark, gloomy track is the ultimate test of deep bass. |
| Delta Machine / Soothe My Soul / Depeche Mode | CD-DA (16/44.1) | Electronic music and vocals at their finest. This is one well-recorded track. Also available in DVD-A 24/44.1 format. |
| Inhuman Rampage / Through the Fire and Flames / DragonForce | CD-DA (16/44.1) | Two-hundred beats per minute and rapid twin guitar solos by Herman Li and Sam Totman of power metal band DragonForce. One broken guitar string in the process. |
| Random Access Memories / Get Lucky / Daft Funk feat. Pharrell Williams | DVD-A (24/88) | Smash pop hit. A well-recorded track available in DVD-A format. |
| Symphonic Dances / Andante con Moto / Rachmaninoff [Eiji Oue w/ Minnesota Orchestra] | DVD-R (24/176.4) | A beautiful classical piece recorded in 24-bit at 176.4 kHz, edited and mastered at 88.2 kHz, and re-transferred to 176.4 kHz |
| Thriller / Billie Jean / Michael Jackson | SACD (DSD64) and DVD-A (24/176.4) | Thriller is the most-sold album ever (100 million copies sold worldwide). If you haven't been living under a rock, you know the track Billie Jean from this album. This particular SACD version is reportedly created by Gus Skinas from the original SACD cutting masters. |
Player Software and Configuration
We used foobar2000 v1.3 beta 7. It's free, it works, and it doesn't mess things up. It's doesn't look cool, but that doesn't deter us. Benchmark wrote a pretty good article on how to configure it.
In short, configure outputs as WASAPI devices, set the output data format to 24-bit (the maximum supported by the devices we're testing), set all volume levels to max (0.00 dB), turn off Replaygain, and bypass all audio plug-ins. Note that devices in WASAPI mode will default to 16-bit operation if 16-bit content is played, ignoring the 24-bit setting. That's a good thing. It means that 16-bit tracks don't get padded to 24-bit, which would be undesirable.
It doesn't really matter if you configure outputs as WASAPI, KS (Kernel Streaming), or ASIO devices, as long as you are consistent. Each mode bypasses the Windows mixer, resulting in a bit-perfect stream to the DAC. The only mode you want to avoid is Direct Sound, which doesn't bypass the mixer.
Although it takes an extremely convoluted process to get working, foobar2000 is one of the few players in the world that can natively play DSD files on supported hardware. A notable alternative is JRiver Media Center, which has a free trial, but isn't free per se.
Objective or Subjective?
Nowhere in this article do we talk about technical specifications or benchmarking individual components. If you want that information, it's available for all three discrete devices. Realtek doesn't provide measurements, only specs, but those are published online too. The point we are making is that, they should all be completely transparent. Realtek's codec shouldn't follow far behind, at least on paper.
If that's true, then we shouldn't be able to tell them apart in a sequence of blind listening tests. That's the angle we're setting out to explore, hence our subjective approach.
A Properly-Blind Subjective Methodology
It's easy to be influenced in a listening test by what you expect to hear. If you feel like you can be objective without a blind test, then great. But we know we cannot. So, we went to every length possible to remove expectations, correcting for any factor that provided unwarranted information.
Typical A/B tests let you hear A, then B, then a random sequence of As and Bs, testing to see if you can correctly tell them apart. If you can guess correctly with a 95% confidence interval, then it's fairly certain that you can tell them apart. If not, you must concede you can't. It's really that simple.
We've deliberately complicated the event in that this is essentially a blind tailored A/B/C/D test. We have four devices. We test one track at a time. We test each track eight times. The only guarantee is that each device will be presented twice in the sequence, though that could be in any order (even consecutively). A proper blind test would not guarantee equal distribution in the sequence, since that creates some form of expectation. But that was a compromise we had to make to generate sufficient data samples for each device.
The tests are conducted with a partner helping us by selecting the sources randomly. During each test, we write down our subjective thoughts. At the end of each run (lasting the first few minutes of each track), if we feel comfortable doing so, we make a guess on the device we just heard. After the eight runs, we compare our impressions and guesses to the actual device list, which our partner wrote down separately.
As you already know, every device is carefully volume-matched, demonstrating good matching across three representative test tones. Only Realtek's ALC889 codec could not quite get there due to its technical limitations.
Color-coding sources using identical extension cables is important for true blind listening
Furthermore, we used identical color-coded extension cables from each devices, so the partner didn't need to move from the test bench at all and the connecting noise for each device was the same. We went one step further and removed the headphones in between runs as the partner was switching connections to avoid hearing any distinct connection-related click or pop.
Due to time constraints, not all listeners tested all content. We also had some issues with volume-matching the Realtek ALC889, and those are called out where they're relevant.
Our precautions worked well; we could not tell the devices apart from each other in any way except their sound.
Challenging the Methodology
A few of the listeners who tried replicating the process above challenged our methodology. This is good, fair, and needs to be openly discussed. So, we present those challenges below.
We were questioned on:
- The process of listening to the same track multiple times using the same or a different device (versus switching across devices seamlessly)
- Using four devices (versus doing A/B testing of individual device pairs)
- The applicability/extensibility of these tests from headphones to full-sized speakers
On the first point, we agree that there is some merit to this. Human acoustic memory happens to be short-lived. Not only that, but rarely are individuals conscious of it. So, trying to "remember" and "compare" how a given track sounds over time (even after multiple seconds) is really, really difficult. With that said, because we were testing on familiar hardware using our favorite tracks, we felt we should have been able to identify differences, with at least directional reliability, if we could hear them. But yes, ideally, we would have liked to try seamlessly switching as well. Unfortunately, we could not find any 1/4" TRS stereo rapid-switching boxes and, even if they exist, foobar2000 won't output over more than a single device at a time (and, we may be wrong on this one, but neither does Windows). Running multiple instances of foobar2000 at the same time is possible, though it creates temporal alignment issues. The idea is nice; it's just technically problematic.
On the second point, our purpose here wasn't telling pairs of devices apart, but rather trying to gauge whether any one component sounded significantly better or worse than the others. Based on what we were trying to achieve, I think our methodology is even better than A/B pairs. This is one challenge thus we'd like to directly rebuke.
Finally, regarding the last point: we agree. These tests, as they were conducted, only apply to headphones. More specifically, they apply to high-impedance headphones. Hopefully, we'll get the opportunity to extend our experimentation to low-impedance headphones in the near future. Full-sized speakers are more challenging for a variety of reasons, and we can't promise that'll happen any time soon.
Wrapping Up
If you've read through the last four dense pages of setup background, then you can appreciate the complexity of arranging proper blind tests. We did our best with the equipment, knowledge, and time we had available to create the best possible experiment, documenting each and every step so that you can judge for yourself how relevant these tests are to you.
The tests aren't perfect, and we don't claim they are. They cannot be generalized beyond the specific cases we tested, and we don't claim they can be. Nevertheless, we hope you'll find them interesting within the scope of their applicability.
We also would have liked to test more devices. If there's enough reader interest, you can bet we'll follow up with a wider range of products.

Track notes
If you're reading this page and you've never played Skyrim, then stop reading. Now. Go play Skyrim. One hundred or 200 game-hours later, you'll appreciate the remainder of this page...
...or will you? In the game's encoded format (a lossy, compressed, WMA-related format with an .xwm extension), Skyrim's soundtrack is pretty outstanding. But the official soundtrack on CD has a special degree of immersiveness that goes beyond the in-game music (or so my brain seems to think).
So, after you put in your game time, go buy Skyrim's soundtrack, spend some more time just listening to that, and after you catch yourself singing out Dovahkiin! with those 90 voices, you will really appreciate the rest of this page.
The Dragonborn track itself is a Red Book-standard 16-bit, 44.1 kHz file. It is remarkable because of its deep vocals and bass-heaviness, intermixed with high treble female vocals that create an extreme sonic contrast.
Test results (Listener A)
| Run | Actual Device | Guess device | Correct / Incorrect |
|---|---|---|---|
| 1 | Benchmark DAC2 HGC | Benchmark DAC2 HGC or JDS Labs O2+ODAC (uncertain) | (Directionally) Correct |
| 2 | Realtek ALC889* | Realtek ALC889 (uncertain) | Correct* |
| 3 | Realtek ALC889* | Benchmark DAC2 HGC or JDS Labs O2+ODAC (uncertain) | Not Correct* |
| 4 | Asus Xonar Essence STX | Asus Xonar Essence STX (uncertain) | Correct |
| 5 | Benchmark DAC2 HGC | Benchmark DAC2 HGC or JDS Labs O2+ODAC (uncertain) | (Directionally) Correct |
| 6 | JDS Labs O2+ODAC | Asus Xonar Essence STX (uncertain) | Not Correct |
| 7 | JDS Labs O2+ODAC | Realtek ALC889 (highly uncertain) | Not Correct |
| 8 | Asus Xonar Essence STX | Benchmark DAC2 HGC or JDS Labs O2+ODAC (uncertain) | Not Correct |
Listener A's comments:
As we start our blind tests, I'm not really sure what exactly to listen for. Based on the results above, obviously for an untrained ear it is hard to tell most devices apart. I'm really curious to see if I improve as I listen to more content. From my notes, it seems that the device I subjectively preferred was what turned out to be Asus' Xonar Essence STX (runs four and eight). That is surprising to me. I'm curious to see if that continues to hold true.
*: Tests of the Realtek ALC889 codec marked with an asterisk had a volume level calibration issue that was corrected later. We kept the results in for the sake of transparency, although they should not be considered representative of an actual ability to distinguish the ALC889 from the other devices being tested.

Track notes
The second single released from the album Delta Machine, Soothe My Soul is a rich track with deep drums and intense vocals. We listened to the Red Book 16/44.1 version, although a 24-bit release is available.
Test results (Listener A)
| Run | Actual Device | Guess device | Correct / Incorrect |
|---|---|---|---|
| 1 | JDS Labs O2+ODAC | Asus Xonar Essence STX (uncertain) | Not Correct |
| 2 | Realtek ALC889* | Realtek ALC889 (very certain) | Correct* |
| 3 | JDS Labs O2+ODAC | JDS Labs O2+ODAC (uncertain) | Correct |
| 4 | Benchmark DAC2 HGC | JDS Labs O2+ODAC (uncertain) | Not Correct |
| 5 | Realtek ALC889* | Realtek ALC889 (very certain) | Correct* |
| 6 | Asus Xonar Essence STX | Benchmark DAC2 HGC (uncertain) | Not Correct |
| 7 | Benchmark DAC2 HGC | Asus Xonar Essence STX (uncertain) | Not Correct |
| 8 | Asus Xonar Essence STX | Benchmark DAC2 HGC (uncertain) | Not Correct |
Listener A's comments:
This was an interesting second test. I can now clearly distinguish the Realtek ALC889 and detect it immediately with certainty. I was highly uncertain of the others. Note my double error in runs six and eight. Twice I mistook Asus' sound card for the DAC2. That's because, subjectively, those were the devices that sounded the best to me. They should be the most expensive, right? It turns out that I show a patter of personal preference for the Xonar Essence STX. It doesn't sound as "reference" as the others, but it seems to be what I favor. The bottom line, either way, is that (excluding the ALC889) I got almost all of the others wrong.
*: Tests of the Realtek ALC889 codec marked with an asterisk had a volume level calibration issue that was corrected later. We kept the results in for the sake of transparency, although they should not be considered representative of an actual ability to distinguish the ALC889 from the other devices being tested.

Track notes
If you have any experience with Guitar Hero III: Legends of Rock, or if you like power metal in general, you almost certainly know this song. It is widely regarded as the game's hardest track.
At 200 BPM, it is extraordinarily fast, and with Hernan Li and Sam Totman playing guitars together in the foreground, this is one track that can very quickly degenerate into a muddled mess with poor headphones or other equipment.
Of the tracks we're using to test, this is probably the one that received the least amount of attention in the studio recording and mixing rooms. It's nowhere near as polished as the others. Since the music itself is amazing, however, that shouldn't deter listeners excessively.
Test results (Listener A)
| Run | Actual Device | Guess device | Correct / Incorrect |
|---|---|---|---|
| 1 | Benchmark DAC2 HGC | JDS Labs O2+ODAC (uncertain) | Not Correct |
| 2 | Asus Xonar Essence STX | Benchmark DAC2 HGC (uncertain) | Not Correct |
| 3 | JDS Labs O2+ODAC | JDS Labs O2+ODAC (uncertain) | Correct |
| 4 | Asus Xonar Essence STX | Asus Xonar Essence STX (uncertain) | Correct |
| 5 | Benchmark DAC2 HGC | Asus Xonar Essence STX (uncertain) | Not Correct |
| 6 | Realtek ALC889* | Realtek ALC889 (very certain) | Correct* |
| 7 | JDS Labs O2+ODAC | Benchmark DAC2 HGC (uncertain) | Not Correct |
| 8 | Realtek ALC889* | Realtek ALC889 (very certain) | Correct* |
Listener A's comments:
By this third test, Realtek's ALC889 codec became easy to tell apart. The others, however, are much more difficult. I have notes riddled with comments like "sounds the same as before" with, for instance, round five (DAC2) compared to round four (the Xonar).
Interestingly, I wrote "sounds familiar" in run four (Xonar), which was actually true, since I've owned Asus' sound card for a long time. Evidently, though, it wasn't familiar enough to become apparent in run two. I'm now starting to focus on whether I can tell the Xonar apart from the O2+ODAC and DAC2. I'm not yet close to telling those two latter components apart.
*: Tests of the Realtek ALC889 codec marked with an asterisk had a volume level calibration issue that was corrected later. We kept the results in for the sake of transparency, although they should not be considered representative of an actual ability to distinguish the ALC889 from the other devices being tested.

Track notes
HDTracks' version of Daft Punk's Random Access Memory album, inclusive of the hit single Get Lucky featuring Pharrell Williams, is made available in a DVD-A format of 24-bit/88.2 kHz. That created issues for us. While the Benchmark DAC2 supports such sampling frequency, the JDS Labs O2+ODAC and Asus Xonar Essence STX do not. Realtek's ALC889 codec does support that frequency in its specs, but in WASAPI mode using foobar2000, neither 88.2 nor 176.4 kHz worked.
Our only choice left was resampling. We could upsample up to 96 kHz, which was closer in absolute value and supported by all devices, or downsample to 44.1 kHz. Though it may sound counter-intuitive, we chose to downsample because converting from 88.2 to 44.1 kHz is a very simple and precise operation. All that is required is a sum-and-divide-by-two. Upsampling from 88.2 to 96 kHz is far more more likely to cause artifacts. Given that the differences between 88.2 and 44.1 kHz are generally considered inaudible anyway, we went the route more likely to preserve fidelity. Also, we used foobar2000's PPHS resampler in Ultra mode, which is considered a very high-quality resampler.
For the sake of integrity, Listener A did some (non-blind due to time restrictions) tests on the DAC2 to see if he could immediately tell 88.2 from resampled 44.1 kHz. The (unscientific) answer is that he could not. He also tried reducing the word length from 24- to 16-bit, with and without dither, and he still couldn't tell any difference.
Test results (Listener A, take one)
| Run | Actual Device | Guess device | Correct / Incorrect |
|---|---|---|---|
| 1 | Asus Xonar Essence STX | Asus Xonar Essence STX (relatively certain) | Correct |
| 2 | Realtek ALC889* | Realtek ALC889 (absolutely certain) | Correct* |
| 3 | Benchmark DAC2 HGC | Benchmark DAC2 HGC (uncertain) | Correct |
| 4 | JDS Labs O2+ODAC | JDS Labs O2+ODAC (uncertain) | Correct |
| 5 | Asus Xonar Essence STX | Asus Xonar Essence STX (relatively certain) | Correct |
| 6 | Realtek ALC889* | Realtek ALC889 (absolutely certain) | Correct* |
| 7 | Benchmark DAC2 HGC | Benchmark DAC2 HGC (uncertain) | Correct |
| 8 | JDS Labs O2+ODAC | JDS Labs O2+ODAC (uncertain) | Correct |
Listener A's comments:
In all honesty, I was really surprised to guess each device correctly. I believe luck played a role, although there is certainly a learning factor involved. I can definitely tell the Realtek ALC889 apart by its lower volume. But this was the first time I felt I could (and did) reliably identify the Xonar sound card from the O2+ODAC and DAC2. I did so through the bass' particular definition. It's hard to say if this track's 24-bit resolution was a factor.
Asus' Xonar Essence STX seems to have a more lively and emotional bass that I subjectively appreciate. It probably isn't as "reference" as the O2+ODAC or DAC2. In that specific sense, it can be argued that those two devices both appear to be marginally superior to the Xonar.
The Benchmark DAC2 HGC and JDS Labs O2+ODAC I simply could not tell apart. I was frankly guessing and happened to be correct. Don't take the outcome as definitive; I believe I could equally have guessed incorrectly.
With that said, all three discrete sources sound absolutely amazing on a track that is absolutely masterfully recorded.
*: Tests of the Realtek ALC889 codec marked with an asterisk had a volume level calibration issue that was corrected later. We kept the results in for the sake of transparency, although they should not be considered representative of an actual ability to distinguish the ALC889 from the other devices being tested.
Test results (Listener A, take two)
| Run | Actual Device | Guess device | Correct / Incorrect |
|---|---|---|---|
| 1 | JDS Labs O2+ODAC | Realtek (uncertain) | Not Correct |
| 2 | JDS Labs O2+ODAC | O2? DAC2? (uncertain) | (Directionally) Correct |
| 3 | Realtek ALC889 | O2? DAC2? (uncertain) | Not Correct |
| 4 | Realtek ALC889 | Asus Xonar Essence STX (uncertain) | Not Correct |
| 5 | Benchmark DAC2 HGC | Realtek (uncertain) | Not Correct |
| 6 | Benchmark DAC2 HGC | O2? DAC2? (uncertain) | (Directionally) Correct |
| 7 | Asus Xonar Essence STX | O2? DAC2? (uncertain) | Not Correct |
| 8 | Asus Xonar Essence STX | Asus Xonar Essence STX (uncertain) | Correct |
Listener A's comments:
My second take, run a few days after the first, demonstrates how hard it is to tell these devices apart. The Realtek codec's volume issues are addressed and it easily blends in with the pack, making identification even harder. I still felt like I could tell the Xonar Essence STX apart, but the results proved me wrong. More interestingly, I could not even tell that a given device was the same when my partner smartly decided to run each device twice in a row.

Track notes
HRX's version of Rachmaninoff's Symphonic Dances is made available in a DVD-R format of 24-bit/176.4 kHz (after a 176.4 to 88.2 to 176.4 kHz process). That created similar issues as Daft Punk's 88.2 kHz album, and the reason is the same. Only the Benchmark DAC2 supports this sampling frequency; the JDS Labs O2+ODAC and Asus Xonar Essence STX do not. Realtek's ALC889 codec does support that frequency in its specs, but in WASAPI mode using foobar2000, neither 88.2 nor 176.4 kHz worked.
Since we were going from 176.4 kHz all the way down to 44.1, we wanted to show that resampling using the same foobar2000 PPHS resampler in Ultra mode introduced no audible artifacts. Both listeners tried multiple times, and neither could tell any difference. In our subjective opinions, 176.4 and 44.1 kHz are exactly the same.
Test results (Listener B)
| Run | Actual Device | Guess device | Correct / Incorrect |
|---|---|---|---|
| 1 | JDS Labs O2+ODAC | N/A - "Less preferred" | N/A |
| 2 | Realtek ALC889 | N/A | N/A |
| 3 | Asus Xonar Essence STX | N/A - "More preferred" | N/A |
| 4 | Realtek ALC889 | N/A - "Less preferred" | N/A |
| 5 | Benchmark DAC2 HGC | N/A | N/A |
| 6 | JDS Labs O2+ODAC | N/A - "More preferred" | N/A |
| 7 | Benchmark DAC2 HGC | N/A - "Less preferred" | N/A |
| 8 | Asus Xonar Essence STX | N/A - "Less preferred" | N/A |
Listener B's comments:
I was surprised at the results, struggling to decide whether the validity of our blind tests could be called into question, or rather if these tests indeed show our auditioned devices cannot be reliably told apart. I do believe that the story would be different if we used full-sized speakers, rather than headphones.
Sampling into Megahertz
Having tried 24-bit/88.2 kHz (and 96 kHz) tracks without being able to tell the difference from Red Book audio, we were starting to get skeptical about high-def audio. Still, we wanted to test the pinnacle of digital audio formats. We picked the most-sold album in history, Michael Jackson's Thriller.

We chose two formats: the exotic DVD-Audio (PCM at 24-bit/176.4 kHz) and the equally (if not more) exotic SACD (DSD64 at 1-bit/2.8224 MHz). The former is available from hdtracks.com for no less than $25. The latter is available from acousticsounds.com, also for $25.
To give you a better idea of the amount of data we're talking about, the uncompressed PCM version (24-bit/176.4 kHz, or 8467 Kb/s) is 2.5 GB, while the DSD64 version (1-bit/2.8224 MHz, or 5645 Kb/s) is 1.66 GB. That's just one album. From a bit rate perspective, DSD64 is essentially equivalent to 16-bit/176.4 kHz PCM, although that in and of itself says nothing about perceivable sound quality.
On paper, Realtek's ALC889 codec supports DSD. But we weren't able to get it working with foobar2000 due to a lack of an ASIO driver. Asus' Xonar Essence STX's DAC chip does support DSD, and Asus does supply a quality ASIO driver. However, Asus' DSP choice, the C-Media CMI8788, does not support DSD, breaking the chain. By design, the O2+ODAC does not support DSD. It's a driverless device that tops out at 24-bit/96 kHz PCM. That left us with Benchmark's DAC2 as the only device supporting DSD through foobar2000.
The catalog of SACDs is small, so a lack of support for this format is hardly a deal-breaker in any circumstance.
A Difficult Comparison

The DSD version of the album plays louder in foobar2000, creating a nightmare as we tried to volume-match. Unfortunately, foobar2000 does not support digital volume control of DSD files, and manually adjusting the DAC2 is an imprecise exercise at best. With that said, the tracks we listened to sounded extraordinarily similar, and we'd guess that they're from the same set of master tapes, though we don't know if the mixing is the same.
Given issues with volume matching and questions about mixing, we hesitate to generalize about DSD versus PCM, so please consider our observations specific to just these two recordings.
Both listeners felt that, while the two versions were enjoyable, the DSD-based copy was better overall. Listener B observed "greater musicality", while Listener A noticed a difference but had a harder time putting it into words, eventually concluding that the DSD version felt more natural.
So far, all of our tests employed Sennheiser's HD 800 headphones. As a reminder, they're relatively high-impedance (300 Ω) cans.
HD 800s and K 550s in their respective housings
As we talked about testing the O2+ODAC, JDS Labs asked that we also try using low-impedance headphones, and the company sent a set of of AKG K 550s, rated at 32 Ω. Its point is that one of the O2+ODAC's main advantages is a vanishingly small output impedance (close to 0 Ω), which is supposed to be great with low-impedance headphones.
Now, the Benchmark DAC2 also has vanishingly small output impedance; its HPA2 headphone amp is rated close to 0 Ω, too. Naturally, then, testing against the DAC2 again was fairly redundant. But what about Asus's Xonar Essence STX, which implements the TI TPA6120A2 datasheet-recommended 10 Ω output impedance level, or Realtek's codec, rated at 2 Ω but subject to a suggested 75 Ω resistor in series on the output path, yielding a typical total of 77 Ω?
Output and Load Impedance
In order to understand why output and load impedance might matter, we need to introduce a concept known as Damping Factor.
As speaker (or headphone) drivers oscillate, they generate a voltage difference of their own that affects all directly-connected electrical components. Without going into too much detail, if an amplifier's output impedance is high compared to the load's impedance, speaker motion and control are impeded. This is particularly true at low (<500 Hz) frequencies, and extremely so at the driver's resonance frequency (resonance, as you can imagine, is very bad for hi-fi audio). The ratio between an amplifier's output impedance and a load (headphones, in this case) impedance is called Damping Factor.
Impedance is a concept that applies exclusively to alternating-current circuits. Furthermore, impedance is not a set figure. It varies based on the frequency of the electrical signal. The 300 Ω-rated HD 800s, for example, typically measure in the 600 Ω impedance range below 1 kHz frequencies.
| Headphone / Amplifier | Headphone Actual Load Impedance at <500 Hz | Amplifier Output Impedance | Damping Factor |
|---|---|---|---|
| HD 800 / Benchmark DAC2 HGC | 600 Ω | 0.1 Ω | 6000 |
| HD 800 / JDS Labs O2+ODAC | 600 Ω | 0.1 Ω | 6000 |
| HD 800 / Asus Xonar Essence STX | 600 Ω | 10 Ω | 60 |
| HD 800 / Realtek ALC899 | 600 Ω | 77 Ω | 7.8 |
| K 550 / Benchmark DAC2 HGC | 35 Ω | 0.1 Ω | 350 |
| K 550 / JDS Labs O2+ODAC | 35 Ω | 0.1 Ω | 350 |
| K 550 / Asus Xonar Essence STX | 35 Ω | 10 Ω | 3.5 |
| K 550 / Realtek ALC899 | 35 Ω | 77 Ω | 0.4 |
A DF of 50 or more is typically considered excellent. That means amplifiers rated for up to 12 Ω output impedance should encounter little trouble driving the HD 800s (600/12 = 50 DF at <500 Hz), even through deep bass. The benefits of even higher damping factors, which you see in the chart above can approach 6000, are debatable.
But using a 32 Ω headphone as a load yields quite different numbers. According to InnerFidelity, the K 550 measures between 34-37 Ω impedance below 500 Hz. With that load on the 77 Ω Realtek ALC899 codec, the Damping Factor is a fairly poor 0.4, and on the 10 Ω Xonar Essence STX, it's a not-so-stellar 3.5.
Technically, you also need to add cable impedance to the amplifier's impedance. But given that the 10-foot cables and connectors we're using have an impedance of <0.1 Ω, I felt that could be excluded. If you're using long or thin cables, the same might not be true.
Coming Soon: Low-Impedance Headphone Testing
We would have loved to test AKG's K 550 as part of this article. But at over 12,000 words, this piece was already a behemoth before going down that road. Still, we know that testing the AKG K 550 (or another low-impedance headphone, for the matter) is important since high-impedance headphones tend to be exotic, expensive stuff. A vast majority of PC users own headphones rated at or around 32 Ω.
Because we haven't yet covered 32 Ω headphones, the results discussed throughout this article only apply to 300 Ω headphones.
We plan to explore whether the same conclusions can be drawn about lower-impedance headphones in a future article.
Musical records vary enormously in their recording and mixing quality. Albums like Daft Punk's Random Access Memories, most of Lady Gaga's pop work, Robin Thicke's Blurred Lines, and many others are simply masterfully recorded/mixed.
Obtaining an audiophile-quality 24-bit/192 kHz version of a poorly put-together track does nothing to make it sound better. As a matter of fact, in all of our blind tests, we couldn't tell the difference between 44.1 and 176.4 kHz, or 16- and 24-bit recordings. While those formats do have a place (namely, in the recording studio where the mixing headroom is a real advantage), they don't seem to add anything to consumer audio. Based on our experience, 16-bit and 44.1 kHz provides the best audio quality you're able to experience. Everything beyond that format tends to be a waste of drive capacity and, since the high-def recordings are more expensive, money as well.
Downsampling a 176.4 kHz track to 44.1 kHz using a high-quality resampler should prevent clipping and yield an output that you can't distinguish from the original. So, if you somehow find a 176.4 kHz recording in your hands and your hardware doesn't natively support it, don't worry. Resample it to 44.1 kHz and know that you're, in practice, not losing any of its fidelity.
Pretty much all modern DACs oversample inputs before the analog conversion (sometimes to a fixed rate [Benchmark converts everything to 211 kHz internally] and sometimes to a input-dependent rate, which is still usually pretty high). Besides the loss of audio frequencies above 22 kHz, which are inaudible, there should be very little difference between a native signal at 176.4 kHz converted to 211 kHz and a native signal at 44 kHz converted to 211 kHz.

The main advantage of 24-bit versus 16- is greater dynamic range (144 dB compared to 98), but that's practically irrelevant. Many of today's records succumb to the loudness war, where dynamic range is artificially compressed in the production stages. Michael Jackson's "Black or White", shown in the figure above, is a great example of this phenomenon. Even if the dynamic range of records wasn't becoming smaller, you'd be hard pressed to meaningfully experience a larger dynamic range in music. As a test, try some of the unscientific but directionally interesting tests on our conclusions page.
Monty at xiph.org has a separate and far more exhaustive discussion of this topic, and I encourage you to read it if you find the subject matter interesting. For our part, we're limiting ourselves to what we hear and understand, which tells us there is no difference between 16- and 24-bit, or 44.1 and 176.4 kHz.
DSD is a bit of a different story. SACDs are vanishingly rare. Any suggestions that the DSD64 format will pick up steam seems highly unlikely to us; even the academic world cannot decide if it is better than multi-bit PCM. The technicalities differentiating DSD and PCM are very complex, though high-quality DSD and PCM recordings shouldn't sound all that different. Both DSD64 and Red Book PCM are, in essence, hi-fi formats. Moreover, it's extraordinarily challenging to compare them in any objective way since DSD and PCM encodings are often obtained through separate masters. When they are not, the DSD encoding comes from a PCM master (what's the point, then?) or an eight-bit DSD called DSD-Wide, which has more similarities than differences from regular PCM. We just have to rely on subjective opinions on this one.
Although we don't typically review audio gear, we believe that we have a few advantages over some hi-fi reviewers. First, we have no financial interest in the products we review. Second, we're PC enthusiasts, not self-proclaimed audiophiles. Consequently, we're not afraid to talk about our strengths and weaknesses. In the audio field, an inability to hear differences among devices spanning a large range of price points is self-defeating. But here, we can comfortably suggest that those products might simply perform similarly.
One thing we know we're good at is designing objective tests, learning from them, and drawing fact-basing conclusions based on the analysis. The integrity of our methodologies is everything, and we can't help but believe that approach is rare in audio equipment testing. We hope our readers will find our experience in testing valuable.
Of course, we also have to acknowledge our own shortcomings and the limits of these tests; neither is perfect. We are audio amateurs, not audio professionals. However, we've tried to create the best possible tests, documenting each and every step along the way so that others can conduct their own experiment and form their own opinion. If you see a way to improve upon our process, we welcome this and look forward to seeing your results, too.
If some of the conclusions we drew sounded implausible, don't worry; they did to us as well.
Try A Few Things For Yourself
Although there is no quick and easy way to replicate the tests in this article at home on your own, here are a few tests we hope you'll have fun with. They should be far more enlightening than our technical explanations of some of the concepts we discussed.
You probably can tell the difference in 1 dB volume levels, but can you reliably tell the difference in 0.5 dB volume levels?
Can you hear all the way up to 22 kHz? What about at or below 20 Hz?
You can probably hear an absolute 54 dB of dynamic range in your environment, but can you reliably hear 78 dB? For reference 16-bit audio has roughly 96 dB of dynamic range. Twenty-four-bit manages a theoretical 144 dB, although it's almost impossible to achieve more than 120-130 using real-world ADCs. Eight-bit audio has a dynamic range of "only" 48 dB; can you reliably tell the difference between 8-bit and 16-bit audio?
Play all of these tests at maximum digital volume. Just be aware that they're not designed to be scientific, but rather to give you some perspective. Try them out for yourself and feel free to post your results in the comments section below!
A $2 Codec Sounds (to us) like a $2000 Device
| Benchmark Media DAC2 HGC | JDSLabs O2+ODAC | Asus Xonar Essence STX | Realtek ALC889 | |
|---|---|---|---|---|
| Price | ~$2000 | ~$290 (including AC adapter) | $190 | ~$2 (OEM in volume) |
| Pros | -Great sound quality -Outstanding build quality -Only device to support 88.2/176.4/DSD64 in practice -Dual headphone out -Greatest number of analog/digital I/O and features (remote control, LCD display) -Free 30-day trial | -Great sound quality -Open-source design that can be self-assembled at lower price point -High-quality volume control -Semi-portable | -Great sound quality -Does not take up desk space -Has both RCA and 1/4" TRS output -Has ADC stage | -Great sound quality -Outstanding value -Does not take up desk space -Supports eight-channel audio -Doesn't require PCIe or USB connectivity |
| Cons | -Very expensive -You pay for features; sound quality is matched at lower price points -Adds desk clutter | -No RCA output -No TRS 1/4" jack -Power transformer not included -Adds desk clutter | -RCA and 1/4" TRS output cannot be concurrently active; switch is software-only -Requires free PCIe slot -No external volume control -Essentially no portability | -Not as linear or hi-fi as the other devices (-1.4 dB @ 100 Hz) -No TRS 1/4" jack -Fixed gain setting -No external volume control -Essentially no portability |
| Application | Extreme PC-driven DAC / headphone amplifier and natural interconnect point with any high-end hi-fi system | Dedicated DAC and headphone amplifier with a convenient volume control and option for limited portability | Budget hi-fi solution that allows switching between 2(.1) stereo speakers and high-end headphones | "Near-Fi" solution that fits almost all major use cases and dominates from a value perspective |
I sank $2000 of my own money into the DAC2 HGC last December, so I subjectively wanted it to sound better than everything else. Tests have shown that it doesn't. I was surprised, but, having been personally involved in the evaluation and believing in the integrity of what we set up, I rationally accept the findings.
Of course, we're ready for the audiophile community to rise up in arms about the statement you'll read next, but it's true that neither an intermediate enthusiast nor a serious one with ~$70,000 in gear at home were able to reliably tell apart any of the four devices once we properly set up a blind test with accurate volume-matching. We actually enjoyed them all as great audio experiences.
Using world-class headphones, a $2 Realtek integrated audio codec could not be reliably distinguished from the $2000 Benchmark DAC2 HGC in a four-device round-up. Again, all four devices sounded great. The same might not apply to full-sized speakers; we can't say, since we didn't test them. But as far as some of the best headphones in the world go, we stand by these test results.
While calibration does show that Realtek's ALC889 is less linear, and thus less hi-fi than the other devices we're looking at, the 1.4 dB difference at 100 Hz apparently isn't enough to reliably differentiate the experience it delivers from others in real-world scenarios. Isn't 1.4 dB a pretty big difference? In a "pure tone", it would be quite noticeable. That's less the case when you're listening to regular music though, especially if the more sensitive 1 to 4 kHz tones are more accurately matched.
But $2 Buys A Smaller Subset of Features
If we halted our exploration at perceived audio quality, we'd only be telling half of the story. There's just so much more to a DAC/amplifier.
Neither the Realtek codec nor Asus' Xonar provide volume control, aside from what you get in Windows. Realtek does support DSD, but without an ASIO driver, we couldn't get it working in foobar2000. Neither lower-end solution can drive headphones and speakers concurrently, let alone automatically mute speakers when headphones are connected. They don't support amplifying an external source, either. Not surprisingly, they're strictly tied to a single device with no real portability. Realtek does facilitate eight-channel output, and the integrated codec and discrete sound card help prevent clutter on your desk.
The DAC2 and O2, being USB-based audio devices, can be plugged into and rapidly switched between any USB source. Want to connect your laptop to your audio system rather than your desktop? That's easy. The O2 has a very high-quality analog volume control, which provided the finest calibration in our round-up. The DAC2 has a motorized volume control with remote control. Want to listen to your headphones in bed and adjust the volume without getting up? Only the DAC2 can do that.
Ultimately, music is about entertainment and personal enjoyment. Hi-fi is meaningful insofar as it heightens the experience of music; it is not necessarily helpful beyond that. Some audiophiles even prefer the low-fi distortion that tube amplifiers introduce. The DAC2, O2+DAC, Xonar Essence STX, and ALC889 are all outstanding solutions. Each delivers a beautiful experience that you'd certainly enjoy. They're similar when it comes to sound quality. Where they differ is mainly in their feature sets and price points.
Value Considerations
I think that money spent on quality recordings, whether they're digital recordings, CDs, DVDs, or SACDs, is the money best-spent. They'll simply never become obsolete.
From there, speakers and headphones are the most important components in your sound system. Headphones generally give you better bang for your buck and are usually more convenient. Obviously, though, they can't replicate the experience of full-range speakers. You feel bass from a subwoofer in ways a headphone can't match. Also, listening to high-end speakers well-separated provides a more immersive experience.
If headphones are the way you go, then our tests show that quality integrated audio codecs are sufficient for driving some of the best in the world. You simply have to live with the fact that a motherboard with built-in audio is going to give you fewer features. That's the point where you have to decide what you're willing to pay for.

We debated whether to recognize a codec for its achievements, rather than a specific product hosting it, but decided that Realtek deserved credit for its work. The ALC889 is found on premium motherboards, as are the newer ALC898 and ALC1150. It's somewhat lacking in the features department; you don't get external volume control, RCA (or 1/4" TRS) outputs, or an ASIO driver. You could argue that eight-channel output partly compensates, though it's not a factor in hi-fi audio. The ALC889 is the least-linear of the devices we tested, though its worst performance of -1.4 dB(A) at 100 Hz is still acceptable overall.
With all of that said, we simply could not tell the little codec's sound quality apart from other, much more expensive devices. For a component that costs 1/1000 of our highest-end contender, that's an impressive-enough feat to earn Smart Buy honors. It also encourages a broader re-evaluation of how integrated audio is reviewed.

The Benchmark DAC2 HGC, on the other hand, easily gets the money-is-no-object Tom's Hardware Elite award. It is a wonderful device with a wide array of features that aren't just mashed together, but rather designed for intuitive usability. Further, the DAC2 is built like a rock. Of course, you'll have to decide if the spec sheet and build quality are worth $2000, particularly since our ears couldn't tell it apart from much cheaper products at a fraction of the price. If you're shopping in this price range for a DAC/amp, also consider the Mytek Stereo192-DSD ($1600), Bel Canto C5i ($1900), and Violectric HPA 200 ($1000), all of which sport similar features and are generally well-reviewed. Remember that our Elite recognition is not a reflection of the DAC2's performance compared to similarly-priced contenders in its class; we haven't had a chance to test them yet, after all.
The PC As The Future of Hi-Fi
As optical storage fades away, we believe PCs will increasingly become the center of the hi-fi listening experience. Nothing can match the accuracy (bit-perfect sourcing and streaming, and no degradation over time) and convenience (thousands of losslessly-compressed albums a mouse-click away) of PCs. Today we even demonstrated that a $2 codec is sufficient for driving some of the most expensive headphones in the world. We haven't tested this yet, so we can't say with certainty, but a DAC hooked up to a PC should also drive amplifiers and associated full-sized speakers as well as the DACs built into, say, high-end CD players. For PC enthusiasts, that convergence is just one more reason to love our versatile systems.
