Dante AVIO-USB Review
Latency can be a showstopper
This article is somewhat harder to write than others. I wanted it to be a love letter to Audinate. I wanted to pledge my undying support and future client recommendations to the Audinate product line because of the wonderful set of capabilities within the DANTE world. Please excuse the future lack of capitalization of this acronym. Okay, let's go.
For the impatient, here are the essentials:
The AVIO-USB is powered via USB or PoE (802.3af)
AVIO-USB macOS Sierra: 17.3-milliseconds round trip latency
AVIO-USB Win-10: 20-millisecond round trip latency (probably)
AVIO-USB in Win-10 is not ASIO, or WDM; compatibility issues
Dante combines the best features of what computer audio connectivity needs to be. The distance limitations of Thunderbolt, USB, and the sunsetting firewire are a constant source of troubles. Networked audio overcomes that problem. Good old MADI remains as a contender, but it’s point-to-point, and therefore flexible topologies are not convenient. Networked audio overcomes that problem too. Channel count has only been a problem for the largest installations, but in general it’s not a problem for any of the competing connectivity solutions.
That leaves latency to be considered. Where Thunderbolt, and MADI shine, there are only a couple of solutions in USB that keep round-trip latency (RTL) down in the small single digits where it needs to be for several solutions that I will outline later in this piece. The PCIe Dante hardware and the Dante adapters that fit into various mixers have impressively low latency, and simply disappear as a consideration in audio network deployment.
Hundreds of other Dante products, including the other three Dante AVIO edge solutions for line level stereo and AES3 connectivity also appear to be free from any onerous latency considerations. (I haven’t tested the three, so it should be noted as a speculation. ) However, audio networks still need additional computer-based-audio edge feeder solutions.
The Dante Virtual Soundcard (DVS), and Dante Via software solutions have been solid edge feeder solutions. They combine low cost ($30) with convenient user interfaces. As software, the has been a sacrifice in latency specs that makes it untenable for a couple of solutions. That’s where I had pinned my high hopes for the AVIO-USB Dante adapter. I hoped it would fill the void of edge solutions with a low-cost, low-latency way of getting a couple of channels onto a Dante network fabric. It didn’t happen. So, enough foreshadowing; let’s get to the data.
What It Does
As it stands, the AVIO-USB serves as an alternative solution to the new connectivity issues created by laptops that no longer have native Ethernet ports. In the case of Apple products, the generic Fast Ethernet and Gigabit Ethernet adapters built on top of USB adapters are not supported with Dante networks, while the genuine Apple are supported. Sometimes 'not supported' means it won't work. Other times it means 'not tested'. In the PC world, I've seen testimony of users with generic USB-to-Ethernet that's working fine.
However, the AVIO-USB sets out on its mission in a limited way. It’s 48kHz-only, and it’s a 2x2 Dante interface. It’s good to see that the USB connection checkbox is filled by the AVIO-USB, but it’s sad to see its latency profile.
The first set of tests were done in macOS Sierra. After that, Win10 was used as a host. In both cases, the round-trip flow is as follows:
• AVIO-USB-01-out thru LAN switch to X-Dante in Behringer X32 Core
• X-Dante-in to Card-In 01 of X32 to Input 01 Direct Out
• Direct Out of Input 01 to Ultranet 09
• Ultranet 09 to Card-Out 01
• Card-Out 01 to X-Dante-out through LAN switch to AVIO-USB-01-in
This topology achieved a 0dB Return Loss and a -250dB Noise Floor in the RTL Utility (beta) offered by Oblique Audio that runs in macOS. The Windows version of the tool has been out for quite a while, and is not a beta level product. Note that there is no AD or DA conversion in this flow. That explains the incredible noise floor number. In addition to the RTL images, screen shots of the Dante Controller and X32-Edit configuration are visible at the left side of this article.
The Oblique RTL (ORTL) numbers are higher (slower) than the reported RTL figures calculated by Logic Pro X, and Studio One 3.5. As you can see in the chart, the DAW figures for the DN32-USB (Midas USB) and the DVS figures make sense in this context.
There is a crying need to explain why the AVIO-USB figures are so different between what the DAWs calculate and what the ORTL measures. I need someone to explain to me why these differences are occurring. In other "soundcard" tests that I have done, the ORTL tool very closely corroborated the RTL figures calculated by the DAWs.
The bottom line is that in macOS Sierra, the AVIO-USB has its best ORTL at 17.354 milliseconds; at buffer size of 160 samples. In Windows 10, the AVIO-USB produced an error message at every buffer size from 512 downward. While noodling around in Studio One 3.5 in Windows 10, a 20-millisecond RTL value is reported by the Studio One DAW, accomplished with a 480-sample buffer.
As for the ORTL tool in Windows-10, the AVIO-USB presents as a Windows Audio device, and for all values sample buffer size, the error message is, “The measured figure is more than 50% different than what the driver reports.” In Windows, this AVIO-USB simply didn't cooperate with the ORTL tool at all. By the way, in Win10. the lowest allowed value for audio buffer size in ORTL is 144 samples.
Soft Synths, Guitar Cabinets, and Artists
The computer edge requirement for a soft synth connection to a Dante fabric requires a low latency link with the computer. Keyboard players need a MIDI-out latency of 5 milliseconds (mS, which I will generally write as ms) or less to be completely comfortable with playing a soft synth. That 5ms one-way figure is the lower edge of what the DVS can achieve.
An even more stringent requirement arises with guitar cabinet simulator software. The deep and cost-effective capabilities of this software is lost to guitar players who need sub-5ms RTL to retain the ‘feel’ of the cabinet responses. The DVS has no chance at offering this with its current range of configurable parameters, and the AVIO-USB would/could have been an absolutely great answer to this requirement if it had the same RTL as the PCIe Dante adapter.
Electronic artists traveling small and light, that use a laptop for sound generation as part of the show could use an AVIO-USB Dante connection through a LAN switch to an Audinate AVIO 2-channel output adapter driving the main speakers. (Note the 2-channel AVIO devices will need a PoE LAN switch. ) Visiting artists at a studio could easily plug in to create and interact with a Dante infrastructure, and take advantage of ad-hoc I/O opportunities on a campus. All of these possibilities need a low-latency AVIO-USB.
Potential Solutions and Target Market
I think there are a couple of ways forward from here. The first one is an obvious upgrade to the firmware of the AVIO-USB device. If it's able to run with a 64-sample buffer at 3-milliseconds output latency, then it's a winner. The guitar sim capability of 3ms RTL is not necessary for this to be an enormously successful product. The next one that comes to mind is to have a higher-performance version of the DVS utility. Right now, the ‘Dante Latency’ value can be set to 10ms, 6ms, and 4ms. I have observed that the one-way latency changes by twice the change in the ‘Dante Latency’ value used in the DVS tool. For example, going from the 6ms setting to 10ms, changed the one-way latency by 8ms; twice the value of the offset.
To me, that implies that if the 4ms setting was allowed a value of 2ms, then that might allow much better one-way latency by 4ms, and RTL by 8ms. If that’s true, then a 3.6ms RTL becomes possible in software. That would be tremendous. Perhaps that setting would have to be put in red bold letters to indicated danger, but the performance opportunity is excellent.
Another possibility is an exclusive relationship with RME, that as a company has famously outperformed all other comers in the USB latency competition.
Even if both or neither of those things are done, I see it as imperative that the selling price of the Dante PCIe card be reduced to a ‘street price’ of $400. I believe that there is an enormous pent-up demand for Dante solutions that are held back by the $1000 selling price of the PCIe card. Although I can’t offer market research proof, I have the following ‘demographic’ observations:
There is a large population of studio owners and audio/video post-production houses of all sizes that run on desktop machines. These are generally the Mac Pro tower and Windows workstation content creators, who have either not embraced the limited professional products from Apple or have successfully braved their way into the Windows high performance hardware jungle. Video post production and Dante seem to be natural partners in workgroup LAN environments.
Availability of proper Thunderbolt solutions has been weak from the audio interface vendors, and weaker still from the Windows support side while retaining all the unsatisfying attributes noted at the top of this piece. Dante is perfectly positioned to take full advantage of the potential market if the pricing structure had a better alignment between the perceived value of PCIe cards, and the function provided.
I have observed that top-end video cards with very high chip populations, and multiple voltage regulators on them sell for less than a Dante PCIe. In addition, I believe that MADI users are prime for the ability to leverage Cat5E and Cat6 wiring conduits, but cannot move from their $700 MADI cards to the $1000 Dante cards with all the concomitant costs. MADI users also have the inertia of the networked audio learning curve holding them back. Attractive pricing helps to overcome inertia. A close cousin sharing the price of the PCIe Dante card is the Lynx AES16e-50. It connects to an almost closed system, since AES50 is limited to Midas, Behringer and just a few other points of light. I doubt that the Lynx card is experiencing sales growth.
At a personal level, I would buy the $1000 Dante PCIe card if the $99 AVIO-USB worked for my needs. Alternatively I'd buy three Dante PCIe cards at $500 each. As things stand, I may simply get the DVS for the MainStage computer, abandon hope for the guitar sim capability, and wait for better times. So from my little piece of the world, Audinate's chain of developers and vendors sees $30 rather than $1100 or $1500. Latency can be a showstopper.
I have used Dante as one of two remote recording setups. In Dante, two laptops, a LAN switch/router, my Dante-enabled Behringer X32 along with the Midas DL151, combine together as a reliable redundant recording setup. To me, more Dante stuff is good from every angle.
While the AVIO-USB is a good idea, the low performance of the device leaves me unable to recommend it to anyone. Audinate can do better, and I trust that this article will be taken as a prompt to offer an improved solution.
- Ted Gary of TedLand
August 10, 2018