Back in March, I wrote a detailed editorial on how hardware-accelerated audio is possible in Windows 11, and it's possible for sound card manufacturers to put native audio stack processing on the card to overcome issues with DPC latency spikes. Today, I'll talk a bit about USB-based onboard audio codecs, and how the Intel platform is better suited to handle them than the AMD platform, particularly given how motherboard designers handle the latter. The Intel "Azalia" high-definition audio (HDA) bus, introduced in the early-2000s, laid the foundation for high-quality onboard audio, offering consumers high-resolution 24-bit/192 kHz audio. This is a low-latency interface designed to ensure the quickest possible movement of audio data between the system and the DAC. On Intel platforms, the HDA bus tends to be wired out from the PCH (the chipset), whereas on the AMD platform, the HDA bus is located on the SoC (the processor), and not the FCH (the chipset). This was a deliberate choice by AMD, and has to do with the chipset bus the company uses.

Unlike Intel, AMD's platform designers adopted a "SoC first" strategy. Modern Ryzen client-segment processors are fully-fledged SoCs in their own right; while the chipset only serves to increase downstream platform connectivity. Ryzen SoCs put out not just PCIe lanes, but also certain platform connectivity, such as USB and SATA; while the FCH puts out additional USB and SATA connectivity. Traditionally, on the Intel platform, the processor only puts out PCIe lanes, while the PCH handles all other forms of connectivity, including USB and SATA; and of course the HDA bus. Intel is able to place the HDA bus on the chipset, while AMD avoids this. The HDA bus is instead put out by the SoC.Before we get into why the two companies laid their platforms out this way, it's important to understand DPC latency spikes. DPC (Deferred Procedure Call) latency spikes occur when a poorly optimized driver or hardware component monopolizes the CPU, delaying other critical system processes. Because Windows handles DPCs in a queued priority system, a single slow driver—often related to networking, graphics, or power management—creates a system-wide bottleneck. These micro-delays directly translate into dropped frames, input lag, and noticeable audio crackling during real-time tasks. So imagine you have a machine with an NVMe SSD and a Wi-Fi 7 connection wired to the chipset; you've just initiated a data transfer from this SSD, while the WLAN adapter is handling a large download. If the HDA interface was located on the chipset, it would run into DPC latency spikes, which would be audible during music playback as crackling or popping artifacts.

Intel is able to place the HDA bus on the chipset, and AMD isn't; because of the different chipset bus types the two companies use. Intel uses DMI (direct media interface), while AMD uses standard PCI-Express as the chipset bus. At the physical layer, DMI is PCIe, however, this physical layer is wrapped with a complex protocol layer which offers hardware QoS (quality of service) features for isochronous data transfers. These allow the DMI link to guarantee bandwidth and prioritize critical, low-latency traffic over bulk storage traffic; favoring latency-sensitive devices such as audio codecs. For example, DMI ensures that audio streams via the HDA bus, USB controller interrupts, and network packets, bypass the traffic jam caused by bandwidth-heavy devices such as NVMe SSDs or SATA RAID controllers. AMD doesn't have this, and instead placed the HDA interface on the SoC, where it talks directly to Infinity Fabric, the internal switching fabric of the processor, which has a sophisticated protocol layer, including QoS controls.

The Azalia HDA specification endured nearly 15 years of hardware primacy, offering sufficient bandwidth for 24-bit/192 kHz audio streams, however, it started to show its age with the introduction of newer audio standards in the professional space, such as 32-bit/384 kHz, and the need for DSD playback. Intel hence was able to guide audio codec manufacturers to switch to USB. Contemporary USB audio codecs such as the Realtek ALC4082 and ALC4080 offer format support up to 32-bit/384 kHz, bypassing the bandwidth limitations of older HDA codecs like the ALC1220S. Intel was able to seamlessly integrate USB codecs onto its platforms by wiring them to the USB 2.0 ports put out by its chipsets, and letting DMI QoS features take care of the rest. Things get tricky with AMD.

On Socket AM5 motherboards with USB audio codecs, the plumbing falls to the shoulders of the FCH (chipset), where most motherboard designers tend to connect codecs to the USB 2.0 ports of the chipset. Since the chipset talks to the SoC over standard PCIe, there is a slightly higher susceptibility of DPC latency spikes causing audio artifacts when the chipset bus is clogged.

It doesn't have to be this way, the AM5 SoC does put out a small number of USB ports from its on-silicon USB host controllers that are connected to the fabric, however, motherboard designers prefer to wire out these ports as USB 3.2 Gen 2 ports on the rear I/O rather than spend them on an audio codec. You can't entirely fault them—why would you waste 10 Gbps of prime bandwidth from the SoC on a device that only needs 480 Mbps?

The high-end Socket AM5 motherboard scene offers a good spread of choice. There are plenty of enthusiast-segment AM5 boards that skip USB audio codecs altogether, instead sticking to the highest quality HDA codecs such as the ALC1220S or ALC1220P, sometimes even giving them secondary high-SNR DACs from the likes of ESS, for specific output channels, such as Front Out. They lose out on native 32-bit or DSD256 support from the codec, but by pairing the ALC1220 with an enthusiast-grade ESS Sabre DAC for the front channels, these boards can still deliver exceptional SNR, as high as 130 dB, within the 24-bit/192 kHz class.

All this said, the problem of DPC latency spikes affecting audio quality requires a specific set of circumstances, such as the chipset bus being severely clogged, or a poorly-designed device driver causing systemwide DPC stalls. Most users will not be severely affected by them, and most users should be perfectly happy with an ALC1220 implementation; but then the whole point of bringing in high-resolution USB audio codecs to PC motherboards was to cater to the discerning audiophile or music composer. The plumbing better be right.