Many luxury technology hallmarks are highly subjective: elegant design aesthetics, visceral sensory experiences, and frictionless usability. The specific elements that achieve these results will vary from customer to customer, and it takes an artist’s touch to create the experience your client is seeking. However, some building blocks of the luxury experience are objective, quantifiable, and undeniable. When it comes to immersive audio, processing power translates directly into quality and capability.
Processing power and chip architecture can account for a 10x difference in the price of an AV processor, so it’s important for CI pros to understand what they’re asking clients to pay for. This article is going to dive into silicon design, computer memory, giga floating point operations per second (GFLOPS), and some other fairly esoteric mathematical and computational considerations, but never forget what it’s all for: an utterly unforgettable immersive audio experience, for music, movies, or anything else, from any position in the listening area. That goal is worthy of just about any amount of math.
Why Processing Power Matters
With the introduction of immersive sound and high channel count, processing power has emerged as the most critical feature of a high-performance AV processor. Today’s high-end processors must simultaneously support high-resolution audio (96k and 192k) and high-channel-count rendering and upmixing (up to 34 channels), as well as advanced room correction and active acoustics features.
Without sufficient processing power, these extreme processing demands can degrade audio quality. Depending on the AV processor’s chip architecture, the device may have certain implementation trade-offs baked in to allow all these intensive requirements to run simultaneously. Typical limitations designed to mitigate insufficient processing power include:
- Limiting the number of rendered immersive sound channels to 24 or fewer, reducing the spatial resolution.
- Limiting the FIR frequency resolution, degrading low-frequency resolution and time domain control.
- Giving up the ability to process high-resolution audio at native 96 kHz or 192 kHz audio frequencies.
- Using 32-bit floating point processing operations instead of 64-bit, potentially introducing audio artifacts.
- Excluding some loudspeakers from digital room correction or active acoustics processing.
When determining audio system design and budget, ask the client which of these sacrifices they’re willing to make. Exceptional audio isn’t the priority for every project, but if the customer wants a true no-compromise system, they’re going to need a lot of computational horsepower.
Chip Architecture Options
Raw computational power is measured in giga floating point operations per second (GFLOPS). A device’s theoretical processing power is determined by the type and number of chips within — though, in practice, architecture and implementation are a massive part of the equation.
It’s worth exploring some specific examples here for an apples-to-apples comparison. The widely used ADSP-214xx Sharc series (4th generation) from Analog Devices, such as the ADSP-21489, runs at 450 MHz and can achieve a sustained processing power of up to 1.8 GFLOPs on average, and up to 2.7 GFLOPs peak. This chip is frequently deployed on its own in cost-effective DSP-based AV processors.
Higher-end processors rely on arrays of two or four DSPs, offering a processing power of up to 6 to 12 GFLOPs peak. These include the single-core ADSP-21569 and the ADSP-21593, a system-on-chip (SoC) composed of an ARM processor (cortex-A5) and a dual-core Sharc+ DSP. These DSPs and SoCs are a common choice for AV processors with audio post-processing features like digital room correction.
At the next tier, the Texas Instruments K2G (66AK2G12) is among the most powerful embedded audio DSPs. It includes two ARM Cortex-A15 cores and a C66x DSP core operating at up to 1.2 GHz, offering up to 19.2 GFLOPs of audio processing. You can find this chipset in high-end processors designed for high channel count, immersive sound decoding, and upmixing.
DSP-based arrays do face limitations compared to a single-chip solution, however. A DSP array requires complex interconnections, communication protocols, and synchronization between two or four separate DSP chips. Communication between multiple DSPs can introduce latency and overhead, impacting audio processing performance. There is another approach: implementing a single, high-performance CPU chip instead of an array of DSPs.
To be transparent: A high-powered CPU typically costs ten times more than a DSP chip, but it also has ten times the processing power. Conservatively, the Intel Core i3-12100 CPU (12th Generation i3), a 4-core, 8-thread processor, delivers a peak performance of around 211 GFLOPs. Its big brother, the Intel Core i5-12400 CPU (12th Generation i5), a 6-core, 12-thread processor based on the Alder Lake architecture, has 240 GFLOPS of available processing power, and a peak processing power of 422 GFLOPs at 4.4 GHz with Intel turbo boost. These chipsets have more than enough processing power to handle an array of demanding simultaneous tasks — and with the single processor architecture, the internal memory is shared among all the internal processing cores so that every piece of data is instantly available to all the cores, eliminating any latency and processing overhead.
Though precise MSRPs vary over time and by manufacturer, Charts 1 and 2 offer a rough estimate of processing cost for common chip architectures. If you look at audio processor costs in terms of price per GLFOPS, single-chip CPU architectures will always measure as a massive bargain.
| Chip/SoC | Single ADSP-214xx (Cost-Effective AV Receivers) | Dual ADSP 2156x (AV Processor) | Quad ADSP 2156x (High-End AV Processor) | Intel Core i3 (CPU-Based High-End AV Processor) |
Intel Core i5 (CPU-Based High-End AV Processor) |
| Typical AV Processor Cost | $1K – $2K | $5K – $10K | $15K – $20K | $15K – $20K | $20K – $40K |
| Expected GFLOPS | 2.7 | 5.4 | 10.8 | 211.2 | 240 |
| Cost Per GFLOPS | $555.56 | $1388.89 | $1620.37 | $82.86 | $125.00 |
Memory: Internal Versus External
Internal and external chip memory also have a significant impact on processor performance. Internal memory is key to effective real-time processing speed. Put simply, it doesn’t matter how much raw processing power a chip has if it’s waiting around for data and program instructions.
The chipset’s internal memory, also known as cache memory, provides a high-speed storage layer that temporarily holds frequently accessed data and instructions, thereby reducing the time needed to fetch this information from the main memory. In real-time audio processing, cache memory ensures smooth and uninterrupted signal processing. This is especially crucial in the specific case of long FIR filter processing. All the FIR filter coefficients, all the audio samples to be processed, and all the processing instructions must be entirely available in internal memory for maximum performance. If there is not enough internal memory, the processor will waste a lot of processing cycles waiting for the missing data to be fetched from external memory.
External memory allows the processor to execute complex algorithms with many instructions. Often, processors also use external memory to implement computation accelerators — for instance, by creating look-up tables of precalculated results, minimizing computation time.
Investing in the Future
No matter how you calculate it, a high-end audio processor is a significant portion of your client’s AV investment. It’s therefore extremely important that the processor you select has the memory and computational headroom to accommodate future processing-intensive technologies. Sound reproduction is not settled science. We’ve made incredible breakthroughs in active acoustics technology in the past two decades, and we anticipate many more to come in the next 20 years. Planned obsolescence is categorically not part of the luxury experience. Your processor recommendation must leave room for future upgrades, which, not incidentally, will drive future room-tuning service and upgrade opportunities for you.
