audio latency testing thread

General information

test used: https://superpowered.com/webbrowserlatency

(assumption: 3.5mm mics are probably necessary to ensure lowest possible latency overhead from mic for this testing, idk yet tho, just use usb/xlr for now and mention it in form)
NOTE: Browser hardware acceleration matters. Use Chromium-based browsers (Firefox doesn't work) Try older versions (~107) as a safe bet, as the toggle in the newest version of Chromium-based browsers can be misleading

note: HDA-compliant cards (using Microsoft's "High Definition Audio Device" drivers usually excel in regards to audio latency (invokes right buffer size) and DPC/ISR results)

what's being tested here is WASAPI shared mode, not ASIO (as many reviews online test for)

Table of contents

few useful links

https://github.com/spddl/LowAudioLatency/releases (doesn't help in browser test as it invokes right buffer size already, but still keeping it here)
https://learn.microsoft.com/en-us/windows-hardware/drivers/audio/windows-audio-architecture
https://mark-dot-net.blogspot.com/2008/06/what-up-with-wasapi.html
https://learn.microsoft.com/en-us/windows/win32/coreaudio/user-mode-audio-components?redirectedfrom=MSDN
https://learn.microsoft.com/en-us/windows-hardware/drivers/audio/low-latency-audio#Measurement_Tools
https://matthewvaneerde.wordpress.com/2018/11/12/what-api-is-being-used-to-play-or-record-audio/
blurbusters thread made in regards to this topic: https://forums.blurbusters.com/viewtopic.php?t=14331

form submission

looking for ppl who will run this test, post results w/ following info:
1.) windows driver used (check in device manager) & windows version used (if linux, mention it explicitly. we're doing windows for now as it's more common. WIN+R => winver to figure out version)
2.) sample rate & bit rate (16bit 44.1khz, 16bit 48khz...)
4.) browser & browser version
5.) mic used (3.5mm, usb or xlr, exact model)
6.) audio interface tested (onboard solution, dac thro usb or toslink, with amp or without amp...)
(run a few times if possible. to ensure its consistent enough, its short)

Results

anon n1

1.) W10 22H2 LTSC IoT, High Definition Audio Device, Driver Date: 09/05/2020, Driver Version: 10.0.19041.264
3.) Tested multiple combinations, shown in table
4.) Brave (131.)
5.) Fifine K669
6.) Tested onboard (ALC1150) and Apple DAC (only 2 I own)

(dac on respective usb controller, mic onboard usb controller)
ASMEDIA APPLE DAC
16bit, 48 kHz {67ms, 71.81ms, 76ms, 3.325335}

RENESAS APPLE DAC
16bit, 48 kHz {83ms, 88.7272ms, 94ms, 3.6702}

Note: When using USB mic & Apple DAC on ASMedia, I seem to be getting +10ms, haven't explored further.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57

DRIVER INFO FOR XG2401 
DRIVER VER: 10.0.22621.1 (06/05/2022, Microsoft, High Definition Audio Device)

**NVIDIA GPU (GTX 1070, 551.23)**
(w11 23h2 + vt2020 (), q24g2a @ DisplayPort)

**onboard enabled**

24bit 48khz
{37, 37, 41, 36, 43}

16bit 48khz
{38, 40, 34, 40, 38}

headphones 16bit 48
{39 39 40 40 34}

**onboard disabled**
16bit 48khz
{43, 38, 37, 40, 38, }

q24g2a @ hdmi
{43, 42 39 40 41}

dual monitor (q24g2a@hdmi, xg2401@displayport) - q24g2a speakers
{42, 39, 47, 40, 45, 41}

[both results below are dual monitor]
- xg2401 speakers - 24bit 48khz
{41, 34, 32, 33, 38, 32}

- xg2401 speakers - 16bit 48khz
{32, 34, 40, 41, 39}

xg2401 only (q24g2a plugged out, displayport on xg2401, 16bit 48khz)
{40, 33, 31, 33, 39, 31}

**AMD GPU (rx580, 22.8.1 driver)**
AMD High Definition Audio Device
06/12/2019
10.0.19041.1 (Microsoft)
alc1150 pc (where i got 32ms originally from)

onboard disabled, q24g2a @ displayport (16bit 48khz)
{108, 89, 84, 90, 86, 94, 97, 94, 93, 93, 91}

q24g2a @ hdmi (16bit 48khz)
{79, 83, 84, 80, 89, 84, 83, 85, 85, 83}

xg2401 @ hdmi
{79, 81, 84, 83, 79, 75, 81, 84, 80, 83, 86}

xg2401 @ 3.5mm jack
{78, 69, 85, 87, 82, 81, 80, 80}

note: i have more bit depth & sample rate options in windows with amd drivers
notenote: changing ref + res combo does not change results

anon n2

USB Audio 2.0 (for the X1), Driver Date: 22/05/06, Driver Version: 10.0.22621.1
Chromium 115.0.5769 (developer build)
W11 23H2
Creative X1, AE-5 (cpu pcie), ALC4080
Samson Go Mic Connect
DAC (X1) 24bit 48kHz
Mic: 16bit, 48kHz

(ONBOARD DISABLED)
1.1) hdmi => startech audio extractor 202a => toslink topping e30 => jds atom Amp — 42ms
1.2) gpu => monitor (xg2431 @ 16bit; 44,1khz) => amp => headphones — 48ms
1.3) Displayport (4090) => xg2431 => jds atom amp — {49ms, 49ms, 39ms ,41ms}
2.) apple dac — 78ms
3.) ifi go link — 81ms
4.) topping e30 — 85ms
5.) AE-5 — 95ms (direct mode enabled)
6.) Creative X1 — 97ms
7.) tempotec BHD pro — 99ms
8.) AE-5 — 104ms (direct mode disabled)
9.) hdmi extractor => G6 via toslink => jds Atom amp — 102ms
10.) ALC4080 — 145ms

anon n3

Chrome 131.0.6778
"High Definition Audio Device"
Driver Provider: Microsoft
Driver Date: 19.10.2023
Driver Version: 10.0.22621.2506

3.5mm speakers (ALC897)
Fifine USB mic
38, 39, 41, 39
iPhone — 69ms

anon n4

anon n5

6800xt, 24.8.1
monitor (AW2523HF) to usb mic (rode nt-2) — 44ms
usb to usb — ~72ms

https://www.amazon.com/dp/B0BKV9GH99
hdmi extractor to usb mic: 49, 45, 50, 45, 49, 49, 45, 48
dp monitor 3.5mm: 48, 45, 51, 54, 47, 54, 46, 50
pretty much the same, usb to usb is like 75

anon n6

alc4080 (downloaded from encore support page) — 144ms

anon n7

winver: W11 23H2 (OS Build 22631.3810)
mobo: msi z690-a pro (Realtek® ALC897 Codec)
topping e2x2 driver: TOPPING Pro, 3/5/2024, 5.68.0.19861, Microsoft Windows Hardware Compatibility Publisher
High Definition Audio Device driver: Microsoft, 10/19/2023, 10.0.22621.2506, Microsoft Windows
browser: Version 110.0.5481.177 (Official Build, ungoogled-chromium) (64-bit)
mic: rode nt1 signature edition xlr
headphones: dt770 pro x

topping e2x2 interface over usbc w/ mic and headphones plugged into it

44.1kHz, 32bit, 8 buffers(min), buffer safe mode off:
84ms, 81ms, 84ms, 85ms, 88ms, 85ms

44.1kHz, 16bit, 8 buffers, buffer safe mode off:
83ms, 82ms, 81ms, 81ms, 87ms

44.1kHz, 32bit, 8 buffers, buffer safe mode on:
89ms, 87ms, 91ms, 88ms, 90ms

192kHz, 32bit, 32 buffers(min), buffer safe mode off:
68ms, 88ms, 91ms, 73ms, 93ms, 71ms, 91ms, 88ms

192kHz, 32bit, 32 buffers, buffer safe mode on:
76ms, 71ms, 87ms, 77ms, 72ms

192kHz, 16bit, 32 buffers, buffer safe mode on:
70ms, 71ms, 88ms, 89ms, 88ms, 76ms

192kHz, 32bit, 64 buffers, buffer safe mode off:
68ms, 76ms, 89ms, 87ms, 73ms, 98ms

192kHz, 32bit, 64 buffers, buffer safe mode on:
71ms, 92ms, 91ms, 94ms, 71ms, 91ms

192kHz, 32bit, 8096 buffer(max), buffer safe mode on:
124ms, 117ms, 138ms

headphones into mobo, mic into interface
44.1kHz, 16bit
{42ms, 41ms, 43ms, 35ms, 39ms}

anon n8

S1220 (DEV_1168) — 33ms

“i noticed loudness correction in audio enhancements consistently cucks it, the other ones like bass boost don't seem to have much of an impact”

loudness correction on

loudness correction off

anon n9

Mobo: STRIX Z490-E (DEV_1168)
Windows 10 (version wasn't revealed)
allegedly stock MS drivers
motu m4 with Sm7b

 1
 2
 3
 4
 5
 6
 7
 8
 9
10

Mojo 2 
usb-c Result: 187ms(antivirus running a scan)
optical 
167ms(antivirus running a scan) , 181ms (antivirus running a scan), 
116ms, 125ms, 122ms and about 15 failures because it kept saying variance is too big (maybe because of the virus scan)

gustard r26
optical: 106ms 
usb3: 181ms(antivirus running a scan), 153ms, 139ms, 166ms, 
usb2: 142ms, 162ms, 163ms, 166ms 

after fixing browser hw accel (by using chrome 106)

1
2
3

with gustard r26 and a mic pointed to fiio ft1 pro going into motu m4
optical: 37ms
usb: 157ms

anon n10

B650M-HDV/M2 (lacks TOSLINK)
Windows 10, Latest Chrome, mic wasn't specified
TempoTec March III (thro USB) — 125, 125, 127ms
onboard 3.5mm (alc897) — 35, 40ms

anon n11

b650e taichi lite (ALC4082) through 3.5mm jack
unknown mic
stock MS drivers — 75ms
realtek drivers — 120, 206ms
astro mixamp pro tr (spdif) — ~120ms

anon n12

Windows 10, Thorium with hardware accel enabled
onboard ALC1200
rode nt usb mini
dt 1990 thro fiio k7 (TOSLINK) — ~50 ms (sometimes bit more, sometimes less)
(usb adds 40-50 ms)
installing realtek onboard drivers also adds latency, goes to ~100 ms

comment:

1
2

bitrate/bandwith has basically no impact on latency
so I'm just running "Digital Audio (S/PDIF)" in mmsys.cpl

anon n13

GB z790m pro ice (listed as DEV_1220)
truthear hexa
at2020+ usb
w11 24h2
brave & latest chrome gave him ~70ms (no hw accel)
old ugc (around version 106) gave him the usual 30ms from onboard controller due to working hw accel

anon n14

asus strix b550f (S1220 / DEV_1168)
3.5mm headphones
blue yeti snowball
w10 22h2 w/ chrome 106 (newer versions didn't engage buffer size, seems to be common for dev_1168 devices)
results: 45,40,31,46,40ms

anon n15

alc4080 (ms driver) — 82ms
fiio k3 (usb) — 80ms
apple usb-c dac — 76ms
startech 3.5mm pcie card — 36ms
startech thro optical — 40ms

LOWEST RESULTS

[1.] 15ms — macOS 10.15.7, Edge 112 (intel hardware very likely, everything else is unknown. results are from superpowered's website)
[2.] 26ms — Windows NT 10, Chrome 128, 48kHz; 128 frm (very likely on pcie audio, but a faster microphone, doesnt seem to differ too much compared to everyone else)

DPC & ISR testing

PCIe-based audio controllers

The test consists of running a 10s XTW trace during the playback of a Youtube video in a Chromium-based browser, with GPU hardware acceleration enabled.
With this, we're evaluating the DPC & ISR behavior of the audio driver.
The API in question is WASAPI shared mode

browser testing

Limitations of testing:

• No conclusion on whether this data can be extrapolated onto different audio APIs in-game (more testing is necessary)
• A single sample point / snapshot in time is a bit misleading to be used to extrapolate performance of a driver within a system. More statistical analysis is necessary, using visualizationsds

(note: ~100 is roughly the run2run variance)
(note: usb ones such as alc4080, xonar se are excluded)
robotic's gpu audio w hdmi extractor
360 µs dpc (elapsed time 99%ile: 1,2µs)
3086 µs isr (elapsed time 99%ile: 5,8µs)
spok's gpu audio
550 µs dpc (elapsed time 99%ile: 1,2µs)
2965 µs isr (elapsed time 99%ile: 7,9µs)
finna's gpu audio
659 µs dpc (elapsed time 99%ile: 3,3)
3168 µs isr (elapsed time 99%ile: 3,7)
via vt2021 (msi mode enabled)
833 µs dpc (elapsed time 99%ile: 3)
4068 µs isr (elapsed time 99%ile: 6,3)
dat's cm8828
585 μs dpc
8155.10 μs isr (?, odd result)
dat's audigy fx v2
504 μs dpc elapsed time 99%ile:
8150 μs isr elapsed time 99%ile:
axaion's hdmi extractor (25h2, firefox)
1207 μs (5,3) elapsed time 99%ile:
6063 μs (51,3) elapsed time 99%ile:
alc1220, msi mode enabled
848,90 μs dpc (2,4) elapsed time 99%ile:
7455,70 μs isr (12,2) elapsed time 99%ile:
(bloat) alc1150 (msi mode enabled)
2110 µs dpc(18) elapsed time 99%ile:
5321 µs isr(29) elapsed time 99%ile:
amd dgpu 3.5mm out
2898.200 µs dpc (3,6) elapsed time 99%ile:
7544.300 µs isr (2x count) (6) elapsed time 99%ile:
ax's z-se
5242 μs dpc elapsed time 99%ile:
7462 μs isr elapsed time 99%ile:
debo llevar's b580 (intel dgpu) audio data
(higher result is FF 147, lower result is Edge/Chromium)
3,2k – 4267 μs dpc; elapsed time 99%ile: 7,2
8,9k – 10977 μs isr; elapsed time 99%ile: 11,2

in-game testing

test scenario: dd2 t-spawn, ak47 in sky (infinite ammo)
x670e steel legend
rtx3080 @ hdmi (rando 60hz tv); hags off; msi enabled
21h2 ltsc
(.etl for each is available, dm if u want it)

LEFT (ALC1200) VS RIGHT (NVIDIA)
in-game (cs2?) behavior changes
large reduction in dpc elapsed time (board→nvidia); particularly max is severely reduced (~10µs);
large isr usage reduction (on nv),
lower count (on nv); on realtek → ISR_count=2∙DPC; on nvidia ISR_count=DPC_count
rest is within run2run variance (at least IMO, can't extrapolate without more data

USB-based audio controller

to be added

Solution for motherboards with the USB-based audio controller

1.) PCIe HDA-compliant add-in card

(e.g.: based on CM8828 chipset)

Make sure to find a card that has MSI (message-signaled interrupt) capabilities, as opposed to IRQ, to lower CPU driver overhead.
reminder on importance of MSI mode
Examples of such cards: EMPTY AS OF Q4 2025

(Xonar SE supports MSI mode & it is functional, but driver schedules on asmtxhci.sys instead of HDaudbus.sys, meaning it's a USB-based card & has a larger CPU overhead than a PCIe-based soluton)

List of cards which don't support MSI mode (no or minimal reduction in DPC & ISR usage):
Delock 89640
Kalea OXYGEN CM8828 CM9882A
Startech PEXSOUND7CH source (WARNING: CAUSES INSTANT BSOD, REQUIRES MANUAL HIVE DELETION PROCESS IF YOU ENABLE MSI MODE)
Creative AE-9 source1 source2 source3
Creative AE-5
Creative Soundblaster X-FI Titanium source

Possible contenders: Audigy FX V2 (+ daughterboard), Soundblaster Z

It can even cause CPU errors (his system very likely has other, larger lingering issues, this crash is a symptom only)

2.) HDMI audio extractor

This solution allows for skipping ALC4080's CPU overhead entirely & rely on Nvidia's GPU audio implementation.
The disadvantage to this solution is that one cannot use 3.5mm-based microphones, as ALC4080 will be disabled.

Notable models which worked for 2 community members:
Startech HD202A
ROFAVEZCO (The HDMI 2.1 variant didn't work for one user)

3.) Getting a ALC1220 or ALC897-based motherboard
Self-explanatory.

General recommendations for audio gear

• 3.5mm microphone through onboard ALC897–ALC1220
  NOTE: You're sacrificing microphone quality for lower total latency & DPC/ISR overhead from USB DAC's)
  Examples (no particular order, no data on quality but I don't expect anything good here):
  Search query
  https://genesis-zone.com/product/radium-300-xlr
  https://antlionaudio.com/collections/microphones/products/modmic-uni-2
  https://www.delock.com/produkt/66640/merkmale.html
  https://www.delock.com/produkt/66582/merkmale.html
  https://geizhals.eu/trust-madell-desktop-microphone-21672-a2485760.html
  https://geizhals.eu/r-de-smartlav-a1211806.html
  https://geizhals.eu/r-de-lavalier-go-a2113082.html
• Optical input standalone DAC's or DAC+AMP's
  Examples:
  https://rme-audio.de/adi-2-dac.html
  https://www.audiosciencereview.com/forum/index.php?threads/measurement-and-review-of-emotiva-dc-1-dac.2306/
  https://chordelectronics.co.uk/mojo-2
  https://www.schiit.com/products/fulla-2
  https://www.schiit.com/products/modi-plus + https://www.schiit.com/products/magni-unity
  https://ifi-audio.com/products/zen-one-signature
  https://beta.prismsound.com/products/callia/
  https://www.tonepublications.com/review/arcam-rdac/
  https://www.lynxstudio.com/products/mesa/
  https://www.lynxstudio.com/products/hilo2/
  https://apogeedigital.com/product/apogee-symphony-desktop/
  https://www.mcintoshlabs.com/products/d-a-converters/MDA200
  (scout used market for some of these brands)

(dont bother with topping, smsl, fiio, fx audio, fosi audio, cambridge audio, teac — reports of shoddy build quality, electrical noise, toslink stability & grounding issues on chinese-based dac+amp's)