It answers from the manuals, any time. AI assistant, may be wrong, the manual is authoritative. A Google account is needed to open it.
What if the computer did its demanding work before the music started? bpplay and bptune are built around that idea. Music is decoded in advance and held in locked memory. The dedicated system runs from RAM. Non-essential background tasks are stopped, and the hardware is configured to minimize avoidable activity during playback. Written from scratch, using only two third-party open-source components, the environment reduces the processes running during playback from a hundred or more to as few as three or four.
We call this Class A Computing: a design philosophy borrowed from Class A amplification, accepting extra power consumption in exchange for a constant, controlled operating state. Here, the objective is to minimize the processing and system changes that accompany music playback. Deterministic operation and a steady, predictable workload have been central development goals from the outset. To verify them, we developed a dedicated measurement method of our own: several dozen measurements that map the key parameters of how the software and the hardware work together on each target platform.
Two settings take this further, on the x86_64 kiosk and the Raspberry Pi editions. Full Pure switches the network interface off at link level for the duration of playback. Once the music is in RAM, the machine is cut off from the network, the Ethernet LEDs go dark, and the connection returns when playback ends. No network traffic, no packets to process, nothing arriving from outside while the music plays.
The selectable CPU isolation mode decides how the processor behaves around the isolated playback core. In B0, the default, only the playback core is held awake, and the others may rest. C0 also keeps core 0 awake, the core that handles interrupts. A keeps every core awake: the most constant state of all, at the cost of more power and heat. Listen, compare and choose: switching is a single key press.
All of this rests on the quiet profile. For the duration of playback, the non-essential background services stop and the internal drives go to standby. On the Raspberry Pi, the SD card is kept awake instead, since there is nothing else to put to sleep. When the music ends, everything is restored. It happens automatically, with every playback.
Live-USB, boots entirely into RAM. SSH, menu-driven browser control.
User manual →Live-USB, no SSH needed. Graphical desktop, drag files onto an icon. Available on request only.
User manual →A sealed appliance, dedicated SD card, boots entirely into RAM.
User manual →Core edition, installed on your own Mac. Terminal, drag-drop or Finder Quick Action.
Download →The x86_64 kiosk and desktop editions boot entirely from a USB stick, straight into RAM. Your existing Windows installation, its disk and its boot configuration are never touched: remove the stick, and the machine boots back into Windows exactly as before. A UEFI BIOS (rather than legacy BIOS/CSM mode) is recommended for the smoothest boot.
Setup, first start, controls and troubleshooting, edition by edition.
On an ordinary app, audio does not reach the DAC directly. It passes through the system mixer, where resampling, floating-point mixing and software volume can all quietly change the signal.
hw: device, no mixer; macOS: hog mode)On the x86_64 and Raspberry Pi editions, bptune does not assume anything about the hardware it runs on; it measures it. Everything that could touch the signal or the system's steadiness is settled before the DAC is running.
bptune measures the CPU cores at boot and chooses the playback core.
The chosen core is isolated from the general scheduler, its governor fixed and boost disabled, and /dev/cpu_dma_latency held open to keep the processor out of deep sleep states. A real-time (PREEMPT_RT) kernel runs the whole system.
The track is read, decoded and the full buffer locked into RAM (mlock) before playback starts: the same principle as the macOS core edition, enforced here at the OS level.
Non-essential background services stop. Internal drives drop to standby (on the Raspberry Pi, the SD/MMC card stays awake instead). The DAC's USB interrupt is pinned to the isolated core.
Isolation of the playback core is the same in every mode. The mode, set with the c key in the music selector, decides which other cores stay awake at all times (a power-management latency of 0 µs):
| Mode | Cores kept awake during playback |
|---|---|
| A | All cores stay awake. Highest power draw and heat, for well-cooled machines only. |
| B0 | Only the playback core stays awake; the others may enter power-saving states. This is the default. |
| C0 | The playback core and core 0, which handles interrupts, stay awake; the others may enter power-saving states. |
A change takes effect when the next playback starts, and does not affect playback already running. On a machine with only two active cores, C0 and A are equivalent. The playback core itself is chosen by the boot-time measurement.
The same principle, reached differently: hog mode gives bpplay exclusive ownership of the DAC, it negotiates an integer hardware format, the track is read and locked in RAM, and the real-time IOProc does exactly one thing during playback: memcpy.
A design philosophy borrowed from Class A amplification: accepting extra power consumption in exchange for a constant, controlled operating state. Even in silence, a Class A amplifier stays ready for the next note. The same logic applies here: the four steps above finish their work before playback starts, so the goal is not to save power during playback, it is to minimize the processing and system changes that happen while the music plays. Deterministic operation and a steady, predictable workload have been development goals from the outset, measured with a dedicated method of our own, on every target platform.
Not just the track: the whole selection, a full album or an entire playlist, is loaded into RAM before playback starts. A playlist with mixed sample rates plays in segments, and each segment is loaded in full before it begins. No on-the-fly buffering, no disk reads during playback. The music sits in RAM, locked there, and reaches the DAC with a single memory copy. This also has an electrical rationale, not just a signal-path one. A conventional player continuously reads, decodes and re-buffers; each operation is a brief, spiky load on the machine's power supply. A modern NVMe SSD's read spike can jump from 15mA to 2A in microseconds, a di/dt of nearly two million amps per second, a real source of high-frequency EMI. Once the music is fully in RAM, none of that happens during playback; there is nothing left to spike.
Neither step is lossy. The float32 mantissa is 24 bits, so the 16/24-bit integer range fits inside it exactly; the extra memory is the honest cost of two lossless conversions, not waste.
DSD256 stores the signal as 1-bit samples at 11.2896 MHz per channel, too dense for USB audio to carry directly. bpplay packs it into PCM using the DoP (DSD over PCM v1.1) protocol.
The alternating marker is a self-checking handshake: a DAC that understands DoP unpacks the bits into its own DSD decoder; one that does not just plays it as plain PCM and stays passive. Either way, the DoP frame is only the transport, never the content. Validated up to DSD256, architecture scales to DSD512.
Since the music is already loaded into memory before playback starts, the machine's internal hard disks, SSDs and NVMe drives switch to a low-power standby state for the duration of playback: quieter, cooler operation, with zero effect on the sound, since nothing is being read from them anyway. Any later access, such as picking the next album, wakes them automatically.
On the Raspberry Pi edition the logic runs in reverse: there is no internal drive to sleep, so the profile instead keeps the SD or MMC card awake throughout playback.
When the quiet profile is on, the following background services are stopped for playback: cron, avahi, rpcbind, polkit, udisks2, the periodic maintenance timers, the udev control sockets, the user session manager, and Wi-Fi supplicant services when no Wi-Fi connection is active. These are never stopped: journald, ssh, NetworkManager, dbus and logind. If any of them is not running after the stop, everything is restored at once. The profile is switched off again when playback ends. On the desktop edition the profile stops a shorter list, so that the graphical session stays open.
Full Pure is an optional mode of the kiosk editions (x86_64 kiosk and Raspberry Pi). With it switched on, bpplay first loads the whole selection into RAM (for a playlist with mixed sample rates, the whole segment), then switches the network interface off at link level, and only then starts playing. The SSH and NetworkManager services keep running, but the link itself is down, so the machine is cut off from the network while the music plays. In our hardware checks the link LED goes dark on the x86_64 kiosk, and on the Raspberry Pi the Ethernet RJ45 LEDs go dark.
When does the network come back? The network comes back when playback ends; the connection is usually available again within about half a minute on the x86_64 kiosk and within about a minute on the Raspberry Pi. A safety timer, set to the playback length plus 60 seconds, is only a backup in case bpplay is killed, so that the machine does not stay cut off.
u key in the music selector. It is off by default, and the setting is remembered.| Format | Details |
|---|---|
| WAV | 16 / 24 / 32-bit integer and 32-bit float |
| FLAC | 16 / 24-bit, Vorbis metadata, MQA detection (not decoding) |
| ALAC (M4A) | 16 / 24-bit: what iTunes and Music.app produce |
| AIFF / AIFC | 16 / 24 / 32-bit, uncompressed only |
| DSF (DSD) | DSD64 to DSD256, native DSD where the DAC supports it, DoP as fallback |
| DFF (DSDIFF) | Pi and x86_64 editions. Uncompressed DSD; DST-compressed DFF files are not supported |
| SACD ISO | Pi and x86_64 editions. Stereo area, DSD64, including DST-compressed discs. From a multichannel area, one stereo pair (the first) plays. Copy-protected images and uncompressed multichannel areas are not supported |
Also checked and explicitly not supported on the macOS core edition: DFF (DSDIFF) and SACD ISO. Both are supported on the Pi and x86_64 editions. On the macOS core edition, DSD plays via DoP.
bpplay plays files from local storage, a NAS, or a Google Drive or Dropbox folder, as long as it is mounted. It is not a streaming client: no built-in connection to Spotify, Tidal, Qobuz or similar services.
Channel count comes straight from the file header, stereo, quadraphonic, 5.1, or anything else the format carries, and goes to the DAC unchanged, up to 64 channels. No downmixing, no channel remapping.
Library management, a graphical interface, streaming services built into the player, signal processing, volume control. None are oversights; each would be a place where the signal could change, or noise could arise, between the file and the DAC.
The m key opens the drive selector. It lists mounted drives, drives that are plugged in but not yet mounted, and network shares (NFS or SMB, mounted read-only); the five most recent shares are remembered. Local drives are mounted read-only. Choosing a drive that is already mounted asks The drive is already mounted. Check for changes? (y/N); answering y re-indexes it.
The / key searches all drives at once. Every word you type must appear in the name, regardless of accents and capitals. Enter opens the folder of the result, with the cursor on the file.
The search index is stored and restored, so search is ready soon after start. Hidden files and macOS ._ files are left out. While a drive is being indexed, playback waits for the first pass to finish, and the status line shows the progress.
In the drive selector, the x key excludes a mounted drive from search and indexing: it is marked (excluded from search), stays in the selector, and can still be browsed and played. The same key switches it back.
Press g in the browser for the machine survey and optimization report: machine and kernel, motherboard and BIOS, the isolated core and the CPU isolation mode, power management and scheduling, memory, the quiet profile and, where present, the Intel management engine. It is followed by DAC diagnostics: the formats the DAC advertises, the format actually used during playback, and the mixer and mute state. A short summary is shown at the first start.
bpplay is released in Hungarian (HU) and English (EN) editions. The language is fixed in the image: it sets the texts of the music selector and of bptune, the default music folder (Zenék in the HU edition, Music in the EN edition) and the time zone (Europe/Budapest in HU, UTC in EN). On the desktop edition the keyboard layout follows the language (hu or us).
This section applies to the x86_64 kiosk and the Raspberry Pi editions. The desktop edition uses its own file manager, and the macOS edition plays local files.
x86_64 kiosk is the reference; the other editions are described by how they differ from it.
| Aspect | x86_64 kiosk | x86_64 desktop | Pi kiosk | macOS |
|---|---|---|---|---|
| License / price | Trial (15 days), then paid | On request | Trial (15 days), then paid | Free, GPL-3.0, open source |
| Installation | Boots from a USB stick; PC untouched | Boots from a USB stick; PC untouched | Dedicated SD-card appliance | Installed on your own Mac |
| Control | SSH, menu-driven browser | Graphical desktop; drag files onto an icon | SSH, menu-driven browser | Terminal, drag-drop or Finder Quick Action |
| Formats | WAV, FLAC, AIFF, ALAC, DSF, DFF, SACD ISO | Same as kiosk | Same as kiosk | WAV, FLAC, AIFF, ALAC, DSF; no DFF, no SACD ISO |
| RT tuning | RT kernel and bptune: CPU held in a fixed, non-switching performance state | Same as kiosk | Same as kiosk (different SoC power characteristics) | No OS-level RT tuning; relies on Hog mode, Integer mode and direct CoreAudio access |
| Quiet profile | Internal drives go to standby during playback | Same idea, with a shorter list of stopped services | Inverse: keeps the SD/MMC card awake, there is no internal drive to sleep | No equivalent mechanism |
| Favorites | Yes | No, does not fit the drag-and-drop model | Yes | No |
| Now-playing line | Yes | Yes | Yes | No, just the file name in the terminal |
| Network shares | Yes | Yes | Yes | No, local files only |
| Full Pure | Yes (u key) | No | Yes (u key) | No |
| CPU isolation mode | Selectable (c key) | No (no music selector) | Selectable (c key) | No |
| Search and machine survey | Yes (/ and g keys) | No (file manager) | Yes (/ and g keys) | No |
| Languages | HU and EN editions | HU and EN editions | HU and EN editions | English |
Prices exclude VAT.
Webshop: coming soon.
Until the webshop opens, you can order through the contact form.
u key): the network is switched off at link level while the music plays./ key, with a stored index; playback starts only after the first indexing pass.g machine survey: report texts, the CPU isolation mode screen, a short summary at the first start, the Intel management engine runtime state, USB controllers.._ files are not indexed.q or Esc.a and n; an SSH banner on the Raspberry Pi; the HU desktop edition gets the Hungarian keyboard layout.g report has a DAC section: the formats the DAC advertises, the format used during playback, and the mixer and mute state.R key is gone from the drive selector, and an already mounted drive asks whether to check for changes.x key excludes a drive from search and indexing only; the drive stays in the selector with the mark (excluded from search), and the same key switches it back.A short, unedited screen recording of installing bpplay and playing an album, start to finish.
Bit-perfect behavior was identical across every topology and cable tried; the bit path is independent of how you connect it.
What independent reviewers, editors, and real listeners have said about bpplay and bptune, quoted with the source linked so you can read the full piece yourself.
Written by reviewers with no connection to the project.
“Clean! That's the word that comes to mind. Absolutely clean. Transparent, resolving. Nothing but the plain truth. Of course, care is needed with volume, since the signal comes through unattenuated.”
“What a phenomenally fine article. Ferenc has hit a grand slam about DSD, the like of which I've never seen, and which fundamentally advances the state of the art. Absolutely brilliant!”
“The first thing you hear is the almost limitless energy and dynamics. The soundstage is larger than I am used to, both in depth and in width. Everything comes through with remarkable clarity, free of any distortion, completely detached from the speakers. … The difference is night and day. It lifts a hi-fi system a whole class.”
From "Hallelujah for the Machine!", a listener who came to bpplay from a carefully optimized two-box Roon/Tidal streaming setup, with almost no local files beforehand. Published with the author's permission.
Self-authored technical essays, not third-party reviews, listed here for transparency.
Real setups, from hi-fi community discussions, not formal reviews, just honest technical chatter.
“This PC happens to be mine, and right now I'm listening to it with Tóth Gábor's D/A converter over USB. What I can freshly compare it against is my setup so far: a Ravenna-based, Ethernet, Merging NADAC D/A converter, with the same PC running HQPlayer on Win11. Although I'm completely satisfied with the latter too, I actually like this new setup better: high resolution, well-defined bass, good rhythm, with a warm, velvety tone as its defining trait, a full, rounded sound.”
A reply to the developer's own post in the same thread, about a dual-boot build (linear power supply, Pink Faun OCXO clock, Matrix Audio USB) running bpplay and bptune on its Debian side.
We built an experimental AI chat assistant, based on more than 100 bpplay/bptune documents and links, that can answer questions about almost any aspect of bpplay: how it works, what it supports, how to set it up.
This is an experimental feature, not a finished or official part of bpplay. Answers may be incomplete or wrong, so always check against the manual or the website itself for anything that matters.
You need to be signed in with a Google account to use it. This is a limitation of Google NotebookLM itself, not something we can configure around.
We do not collect, store, or process any data from your use of this chat ourselves. All data handling is done exclusively by Google, governed entirely by Google's own Terms of Service and Privacy Policy for NotebookLM, not by us.
The macOS edition is free and open source (GPL-3.0). The x86_64 kiosk and Raspberry Pi editions are trial editions: 15 days, then paid. The x86_64 desktop edition is available on request. Developed by Koscso Media Engineering Kft. Available for licensing and integration into partner products.
This website uses no cookies and runs no tracking scripts.
We do not collect, store or process any personal data about your visit.