Perhaps the biggest surprise—and the best kept secret of AMD’s new R7 and R9 graphics chips--was the inclusion of a new advanced audio technology dubbed TrueAudio. That’s right, a video card with audio support. With advanced PC audio considered a long forgotten technology, Maximum PC played 23 questions with AMD’s Carl Wakeland. Wakeland is a Fellow Design Engineer and considered the “author” of TrueAudio.
Maximum PC: In one sentence TrueAudio is:
AMD: A programmable audio core built into the GPU, representing our effort to breathe life into game audio environments as the programmable graphics pipeline breathed life into the diversity of PC graphics.
MPC: I didn't get to attend the briefing but if I get this; AMD has integrated Tensilica HiFi2 EP DSPs directly into the die or on the package or on the PCB? It says multiple but how many exactly?
AMD: The entire AMD TrueAudio hardware block is built directly into the die of the AMD Radeon R9 290X, R9 290 and R7 260X GPUs. While I cannot disclose the quantity, I can say that the game developers we’ve been speaking to have been overjoyed by the hardware capabilities we’re providing.
(AMD has since disclosed to us that it has three Tensilica DSPs on the die).
MPC: This is in addition to a separate Tensilica Xtensa SP is integrated directly onto the die or package or PCB?
AMD: Inclusive of the Xtensa HiFi EP core and multiple Xtensa HiFi2 EP cores, the AMD TrueAudio block is 100 percent integrated into the die of the graphics chip.
MPC: I'm not up to speed with Tensilica hardware but why the need for the Xtensa SP? Isn’t it an FPU? I thought the calculations were done in the GPU?
AMD: The Tensilica Xtensa HiFi EP provides single and double precision floating point assistance for calculating accurate simulations of the intended audio environment. Compute resources from the GPU or CPU pipelines are not required, and that’s the intention of AMD TrueAudio: 100 percent offloading to preserve or even improve system performance, even with superior audio.
MPC: How much of the calculations are GPU-based, how much are host-based?
AMD: Any calculation AMD TrueAudio is harnessed to perform is done entirely on the AMD TrueAudio silicon.
MPC: Could TrueAudio be implemented in an APU-setup?
AMD: AMD TrueAudio could be integrated into any Graphics Core Next-based graphics chip, but it’s too early to discuss where the audio engine might appear next.
MPC: Could TrueAudio be implemented with a discrete card with the right hardware (in other words, without the need for an AMD GPU?)
AMD: AMD TrueAudio depends upon the greater GPU. TrueAudio leverages the significant bandwidth and low access latency of a Radeon’s memory pools, and that bandwidth and latency is critically important when offloading audio tasks from the CPU to the TrueAudio engine.
MPC: I'm guessing Xbox One will not have TrueAudio as Microsoft has its SHAPE audio engine but what about PS4?
AMD: Sony and Microsoft would be in a better position to comment on the functionality of their audio hardware. I wouldn’t want to answer on their behalf.
MPC: Those who have followed PC audio for a long time have long heard that an X-Fi or EAX took so-and-so performance that the CPU can't do or that the reflections in A3D 2.0 are too much for the CPU, with most gamers running quad-core and up today, and most of the graphics calculations offloaded to the GPU today, is there really much work for a 4/6/8-core CPU to do?
AMD: Oh gosh, there’s plenty for a CPU to do. There’s so much to do that today’s PC gaming audio engines are capped to 8-10 percent (at best) of the average CPU’s performance budget just so the more “important” bits like AI, physics or game simulations are able to function unhindered. More often than not, the CPU is so busy that audio engineers only have access to an uncertain amount of “leftover” CPU resources, rather than a fixed target. That encourages very conservative audio production, of course, which we hope to alleviate with AMD TrueAudio. CPUs are also responsible for feeding the beast—the graphics card—with all the data it requires. More powerful graphics cards require correspondingly more powerful CPUs.
MPC: You guys are using GenAudio and McDSP. GenAudio’s says AstoundSound modeling is based on how the brain reacts to sound rather than the older model of putting microphones into head dummies to create the algorithms. GenAudio calls it HRBF. How is HRBF an improvement on Head Related Transfer Functions (HRTFs)?
AMD: A key point I want to make here is that we are working to integrate AMD TrueAudio technology into game audio libraries. The very libraries that game developers would license to incorporate an audio engine in their game—that’s where AudioKinetic’s Wwise and Firelight Technologies’ FMOD come into the equation. GenAudio and McDSP are an even earlier stage in that process, creating plugins that interface with Wwise or FMOD to provide diverse audio effects that are compatible with the AMD TrueAudio chip. Now, the library guys themselves could adopt these plugins, or the game developers could go get the GenAudio (or McDSP) plugin and snap it into the library they have. The latter approach is what Xaviant Games has done. Their game, Lichdom, runs on the Wwise audio engine, which they’ve augmented by utilizing the Wwise-compatible GenAudio AstoundSound plugin.
To answer the second part of your question, GenAudio’s BRTF is based on the way the brain actually interprets audio, as analyzed through EEG-fMRI and MEG data. Traditional HRTFs assume, as you implied, that your head is the same size and shape as the dummy head used to create the psychoacoustic model. But the critical failing of an HRTF is that nobody’s head is the same size or shape as the binaural dummy head, so you have to go deeper. Right into the brain. Learn how the brain itself receives and processes a 3D soundfield, and then you work back from there to create the algorithms that work on that level. The GenAudio team has done that, and I’ve not met one person who hasn’t been blown away by the quality of their methods as demonstrated in Lichdom.
Jerry and his team at GenAudio are a team of mad audio geniuses, and I’m sure they would love the opportunity to give you the full breakdown of how they are different.
MPC: Even in the 1990s HRTF's were pretty impressive for headphone users, but the results were often mixed depending on a person's lumpy potato head. HRTF'ed audio always seemed particular week at placing audio behind you with headphones, will TrueAudio offer an improved performance?
AMD: Remember that AMD TrueAudio technology is ultimately a programmable DSP core, so the quality of the simulation will depend on the game developer and their chosen combination of libraries and plugins. I’m 100% confident, however, that AMD TrueAudio already provides a significantly better behind-the-head spatialization than any two-channel positional audio solution I’ve previously encountered (primarily EAX and A3D, as with most gamers).
MPC: AstoundSound offers enhanced positional such as Distance and elevation/azimuth? Does it offer occlusion and reflections or wave tracing capability that Aureal 2.0 offered?
AMD: Distance and elevation are certainly within AstoundSound’s purview. For the other effects, you would have to reach out to them for clarification.
MPC: Is there room to add other advanced audio techniques or is there just too much licensing involved to get it?
AMD: That’s the beauty of having a programmable DSP core: we can accommodate plugins and libraries that utilize whatever IP the industry has on tap. We’re unveiling a blank canvas and inviting the audio artists to come play, but it’s up to them to bring the right palette. We see this as the same opportunity presented to graphics artists at the advent of the programmable graphics pipeline.
MPC: I understand you guys are also working with McDSP too? Most of their IP seems centered around professional plugins. Where do they fit into the TrueAudio picture?
AMD: As an audio plugin provider, they’re one of many solutions a library vendor or game developer can call on to achieve AMD TrueAudio-accelerated environmental effects. Their “mastering limiter” plugin is especially well-suited for gaming audio as it allows for a greater number of sounds to be dynamically mixed at high volume without “clipping” artifacts in the audio stream. This is a great example of a pro audio plugin finding a relevant niche within the realm of gaming audio.
MPC: Didn’t Microsoft pretty much put a stake through the heart of advanced audio with Microsoft Vista when DirectSound was demoted and games could no longer directly access the hardware audio layer? How does TrueAudio get around this limitation?
AMD: Ah, this is some of the real magic behind AMD TrueAudio. By operating through the graphics driver, we aren’t held accountable to the limitations imposed on the hardware audio layer. We do not touch the Windows audio stack at all. TrueAudio offloads directly from the game via Wwise or FMOD, right at the point of conception, before the audio is sent through the rest of the audio pipeline to the user’s endpoint device (e.g. headsets). It’s real-time, it’s an earnest revival of advanced gaming audio, and it sounds bloody wonderful.
MPC: One of the problems with gaming today is a lot of gamers run USB headsets. The DSP's are external to the PC – but with TrueAudio, you’re saying you’re getting to the audio before it even gets pushed out to the audio devices? How does this work if you have an existing sound card already such as an X-Fi, Xonar or advanced onboard audio already?
AMD: AMD TrueAudio comes into the audio chain at the application level, long before sound ever reaches the user’s audio chip or audio endpoint. Whether you have integrated audio on the motherboard, a discrete sound card, or a standalone USB headset, AMD TrueAudio is already part and parcel of the audio stream that’s being fed to these devices by the game’s audio engine. That’s the beauty of operating at the level of the audio library: it’s the first stop in the audio process! And because it’s the first stop, only AMD TrueAudio is fully aware of the game’s positional and environmental data. We are alone in our ability to provide audio data that fully reflects the game’s goings on.
MPC: So a person could keep their existing discrete X-Fi/Xonar/Recon for the superior DACs/ADC, and switch off their effects to use TrueAudio but then switch them back for games that use OpenAL or don’t support TrueAudio?
AMD: That’s 100 percent correct. We designed AMD TrueAudio in the manner that we did precisely because we know users have great audio hardware with high-quality OPAMPS that they don’t want or need to forfeit. Whatever audio device(s) a user has right now, that hardware is ready to go with AMD TrueAudio.
MPC: For TrueAudio to work, developers must use and support TrueAudio’s API correct?
AMD: The game developers themselves don’t necessarily have to use or support the TrueAudio API. Because we’re working with library and plugin developers directly, game developers only have to choose between a software-accelerated plugin or the TrueAudio-enabled plugin when they’re working with their library of choice. Devs are making plugin/SFX choices for their games anyhow, so TrueAudio just becomes another choice on the menu, meaning it’s very simple and unchallenging to implement.
MPC: Isn't that the classic uphill battle that Aureal and Creative Labs faced time and time again? Or in a less polite way, isn’t this just in ability to get anyone to actually adopt Advanced EAX and A3D 2.0 all over again?
AMD: I would start by saying that AMD TrueAudio is genuinely different from the audio algorithms that have come before. For starters, we’re totally programmable. You’re not getting an audio environment the way AMD thinks you should get an audio environment; you’re getting an audio environment designed by a professional sound engineer, specifically tasked with breathing life into the soundscape of a game. For the first time, his or her game can sound exactly like they dreamed of, which was not the case with the canned audio algorithms of previous generations.
This way was paved by the growth of middleware audio libraries like AudioKinetic and FMOD, which abstract audio APIs like TrueAudio. Game developers no longer have to work directly at the API layer, and can instead get on with picking or programming the right effects from the menu provided by the audio library.
MPC: Why do you guys think developers will really adopt the TrueAudio API in any meaningful number beyond a handful of showcase titles?
AMD: Based on the initial reactions of the developers we’ve communicated the technology to, we’re very optimistic.
MPC: Getting back to the USB headset thing, since many USB headsets feature faux 3D that often can't be turned off, will there be a problem with audio that is has already been processed to have the effect ruined by being processed again?
AMD: I’ve not yet seen a USB headset whose inline DSP can’t be disabled. It doesn’t mean they don’t exist, and the plural of anecdote is not data, but I’d say they’re probably the exception rather than the rule. That said, AMD TrueAudio does have the basic expectation that another DSP-driven audio effect isn’t being inserted into the audio pipe.
MPC: So the cards that will get TrueAudio are: R9 290X, R9 280X and R9 270X and R7 260X? Why not implement TrueAudio across the lineup and why only on the R7 260X?
AMD: The AMD Radeon R9 290 Series and the R7 260X are the flagship models of the R9 and R7 Series, respectively. We wanted to enable the very best features on our very best products.
MPC: TrueAudio is fully implemented and ready to go in games already so people can expect to be playing TrueAudio-enabled games by this Christmas?
AMD: Lichdom and Thief are 2014 games, whereas Murdered: Soul Suspect doesn’t yet have a release date. Further out, Chris Roberts of Cloud Imperium announced that Star Citizen would eventually support AMD TrueAudio as well. While none of these games are imminent in 2013, there is a clear cadence of supporting titles, with even more in the wings that we can’t yet announce.
MPC: Can we possibly build a PC that support's Intel's QuickSync, Nvidia Physx and AMD TrueAudio?
AMD: Not until it’s possible to simultaneously run multiple brands of GPUs in one system without your operating system of choice having a meltdown.