• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

RME dac filter for competitive gaming

miero

Active Member
Joined
Aug 1, 2018
Messages
241
Likes
289
Try SPDIF/Toslink directly from PC motherboard or a sound card instead of USB. There is additional data buffering while USB packet is fill up with data and sent over a wire.
 

MC_RME

Addicted to Fun and Learning
Technical Expert
Audio Company
Joined
May 15, 2019
Messages
854
Likes
3,564
Should I be using the minimum phase filter for gaming for faster response time?
All my other gear is built for speed but the linear sharp filter seems like it is better for the sound, but bad for delay in response time of the sound.

It doesn't matter - I will explain. Let's start with a table of the delays caused by the various filters of the ADI-2 DAC, as stated in the AK4490 data sheet:
Code:
         44.1 kHz    96 kHz    192 kHz
SD Sharp  6.25       5.63       5.63
SD Slow   5.3        4.68       4.68
Sharp    29.4       28.8       28.8
Slow     6.63        6.0        6.0
NOS       ---         ---       ---

Remember that NOS is Super-Slow, and there is no value stated in the data sheet. Well, doesn't matter as these values are in samples (!). The worst case is the Sharp filter with 29.4 samples. At 44.1 kHz that equals 0.83 ms. Yes, less than a millisecond. If you use the (also phase linear) Slow you are at 0.3 ms. The statement 'significant delay with a linear filter compared to minimum' is true for typical EQs in the low frequency range, but not always for higher frequency aliasing filters.

But the main point is - these are just the DA filters. You have to add:

- buffers within the game for audio output
- buffers in the audio subsystem (WDM/ASIO etc)
- buffers in the audio subsystem processing (if not bypassing the kernel mixer)
- hardware buffers in PCIe, USB etc.
- hidden buffers in the driver
- buffers in the audio hardware
- processing the data within the audio hardware

That is everything from 1 sample up to a thousand samples, per item! The total amount for pure playback easily reaches 100 samples even in best case.

Measuring the latency from USB event to analog output is quite complicated - and somehow meaningless, because whatever tool you write will not represent how another program handles this. Using ASIO it gets more easy - but Games don't use that. So let's look at something that is easy to measure: the time between the arrival of the digital data at the SPDIF input until an analog output signal shows up. Doing this with the ADI-2 DAC we get (again in samples):

Code:
       44.1 kHz   96 kHz   192 kHz
SD Sharp  14       13       19
SD Slow   12.4     12.3     18.2
Sharp     36.5     36.6     42.2
Slow      13.8     13.8     19.7
NOS        8.4      8.7     14.7

Compare these values to the pure digital filter values and you will see they are naturally higher. The SPDIF receiver (1 sample) plus internal data processing (6 samples at single speed) are added. And as we measure we now also get values for the Super-Slow (NOS) mode. Here is the above in milliseconds:

Code:
        44.1 kHz    96 kHz    192 kHz
SD Sharp  0.3       0.14       0.099
SD Slow   0.28      0.13       0.094
Sharp     0.38      0.38       0.22
Slow      0.3       0.3        0.1
NOS       0.19      0.14       0.077

But all these numbers have to be taken in context. As mentioned by others the overall latency will be around 5 to 6 ms - if you are lucky. So whatever filter you choose you won't notice any difference. In the ADI-2 DAC's Settings dialog you can choose a buffer size down to 32 samples. That setting is also valid for WDM and your Games, which typically work best at 128 samples.
 
Last edited:

MC_RME

Addicted to Fun and Learning
Technical Expert
Audio Company
Joined
May 15, 2019
Messages
854
Likes
3,564
Sorry for the bad formatting. Tab doesn't work and multiple spaces are automatically reduced to one. How to get the above look like a readable table?

Edit: Thanks for pointing out the Quote workaround. But much too cumbersome. Next time I will just insert a pic of the table.

Edit 2: Value of Slow 96 kHz in ms corrected.
 
Last edited:

bravomail

Addicted to Fun and Learning
Joined
Oct 19, 2018
Messages
817
Likes
461
I have tested .15 to visually react and press a button. I've gotten faster, but that's pretty much my average.

https://www.humanbenchmark.com

try it out.

My monitor is 1ms for real though. Real world I think i do get 3ms in overwatch though. So to get your monitor to run 1ms real world isn't possible.

You gotta take a monitor refresh rate and your game-videocard FPS rate into consideration. Let's say you have 240Hz monitor and your hardware can pump 240 or better frames per second. That still gives you 4ms delay. So how can display manufacturers claim 1ms response? Where are 1000Hz refresh rate monitors?
 
OP
S

ShiZo

Addicted to Fun and Learning
Forum Donor
Joined
Sep 7, 2018
Messages
835
Likes
555
I have a 240hz monitor and have hardware good enough to push 2x that in fps. so between 480-600fps.

In games, real world, I get between 3-4ms input lag.
 

mkawa

Addicted to Fun and Learning
Joined
Sep 17, 2019
Messages
788
Likes
695
re: motherboard sp/dif out. you have to pass through the realtek codec to get there. almost certainly going to be slower than the direct child of a usb device on the chipset (i know intel's skylake+ doesn't have usb on-chip yet, but does zen 2 have it on the bridge controller/traffic cop that's on the MCM?)

that said, you're probably more limited by visual and your own game sense and sitch awareness than audio cues.

also, this level of support is why i'm sad that i can't afford and RME ADI2... :(
 
Last edited:

Berwhale

Major Contributor
Forum Donor
Joined
Aug 29, 2019
Messages
3,933
Likes
4,922
Location
UK
So I play PUBG which can simulate positional information by applying HRTF to a 2 channel output to headphones. I assume that the application of HRTF is going to add latency to the audio pipeline, but is that additional delay outweighed by the benefit of the extra positional information? i.e. having a better indication of where the footsteps are coming from.
 

RayDunzl

Grand Contributor
Central Scrutinizer
Joined
Mar 9, 2016
Messages
13,198
Likes
16,981
Location
Riverview FL
Assumiing multiplayer - add network delays...

Pinging audiosciencereview.com [104.27.182.13] with 32 bytes of data:
Reply from 104.27.182.13: bytes=32 time=31ms TTL=51
Reply from 104.27.182.13: bytes=32 time=31ms TTL=51
Reply from 104.27.182.13: bytes=32 time=30ms TTL=51
Reply from 104.27.182.13: bytes=32 time=48ms TTL=51

It would take me at least 31 and as much as 48 ms per above to post a goofy reply to someone's post in this game.
 

Berwhale

Major Contributor
Forum Donor
Joined
Aug 29, 2019
Messages
3,933
Likes
4,922
Location
UK
Assumiing multiplayer - add network delays...

Pinging audiosciencereview.com [104.27.182.13] with 32 bytes of data:
Reply from 104.27.182.13: bytes=32 time=31ms TTL=51
Reply from 104.27.182.13: bytes=32 time=31ms TTL=51
Reply from 104.27.182.13: bytes=32 time=30ms TTL=51
Reply from 104.27.182.13: bytes=32 time=48ms TTL=51

It would take me at least 31 and as much as 48 ms per above to post a goofy reply to someone's post in this game.

Many games have latency balancing to correct for the measured network latency differences between players.
 

mkawa

Addicted to Fun and Learning
Joined
Sep 17, 2019
Messages
788
Likes
695
good netcode has really good speculative/predictive wizardry to cut down effective latency in the common case. back in the day it was just dead reckoning, but now it's quite a bit better. the recent announcement video of riot games' "project a" has a portion of the video dedicated entirely to the new netcode.
 

Veri

Master Contributor
Joined
Feb 6, 2018
Messages
9,596
Likes
12,036
44.1 kHz 96 kHz 192 kHz
SD Sharp 0.3 0.14 0.099
SD Slow 0.28 0.13 0.094
Sharp 0.83 0.38 0.22
Slow 0.3 13.8 0.1
NOS 0.19 0.14 0.077
Awesome that you actually tested this all, was just wondering if the lone 13.8 there is correct :)?

Very interesting stuff, thanks a lot for sharing RME!
 

scott wurcer

Major Contributor
Audio Luminary
Technical Expert
Joined
Apr 24, 2019
Messages
1,501
Likes
2,821
5ms is around the time people can start to feel/ not feel a delay. This number is coming from the guitarists that were using digital processing effects, musicians using monitoring with effects plugins, and many people who work in pro audio scene. It generally holds true for gaming as well.

You said it first I know two guitarists that can't stand overdubbing with more that 5ms latency.
 

Eirikur

Senior Member
Joined
Jul 9, 2019
Messages
318
Likes
509
Sorry for the bad formatting. Tab doesn't work and multiple spaces are automatically reduced to one. How to get the above look like a readable table?
You can use "code" from in the "..." pull-down menu in the editor.

Code:
           44.1 kHz  96 kHz 192 kHz
SD Sharp       6.25    5.63    5.63
SD Slow        5.3     4.68    4.68
Sharp         29.4    28.8    28.8
Slow           6.63    6.0     6.0
NOS             ---     ---     ---

Bit of a pain if you need to edit (copy-paste into new tags works best).
 

zermak

Senior Member
Joined
Jun 2, 2019
Messages
372
Likes
251
Location
Italy
I would care more for the video latency in general rather than the audio latency. As far as I know and read (Rtings, TFT Central, hardware.nl, monitor.info) the lowest input lag is about 2.6ms and the panel activation is around the same value on a 240Hz TN monitor.

About netcode in games the youtube channel Battle(non)sense would disagree for the most part. It sill sucks most of the times and it is always a work in progress.

@MC_RME thanks for the testing and by the way you can use the tag CODE to properly shows the 'tabs'.
 

mkawa

Addicted to Fun and Learning
Joined
Sep 17, 2019
Messages
788
Likes
695
re: netcode. it used to be so so so much worse. bungie's first serious multiplayer game: myth TFL actually went so far as to use a global fixed clock and complete determinism everywhere in the engine (like, no floating point levels of determinism) to avoid speculative misfires. worse, if any client went desynch from the global state, the entire game was killed. the first iteration of quake also avoided prediction by designating a single host the master and slaving everyone to purely sequential consistency as seen by the master. basically, your latency was your refresh rate.
 

KSTR

Major Contributor
Joined
Sep 6, 2018
Messages
2,690
Likes
6,012
Location
Berlin, Germany
You said it first I know two guitarists that can't stand overdubbing with more that 5ms latency.
I once had a worst-case live event: I (bass player) had no monitor of the drums and had to rely on what I heard from the drum-set which was some 5 meters away (15ms) but the drummer had my direct signal on his monitors. He constantly thought I was trying to slow him down because I played way behind the beat, given the late arrival. We managed through the gig... making lot's of funny faces to communicate...
 

VintageFlanker

Major Contributor
Forum Donor
Joined
Sep 20, 2018
Messages
4,940
Likes
19,691
Location
Paris
It doesn't matter - I will explain. Let's start with a table of the delays caused by the various filters of the ADI-2 DAC, as stated in the AK4490 data sheet:
Code:
         44.1 kHz    96 kHz    192 kHz
SD Sharp  6.25       5.63       5.63
SD Slow   5.3        4.68       4.68
Sharp    29.4       28.8       28.8
Slow     6.63        6.0        6.0
NOS       ---         ---       ---

Remember that NOS is Super-Slow, and there is no value stated in the data sheet. Well, doesn't matter as these values are in samples (!). The worst case is the Sharp filter with 29.4 samples. At 44.1 kHz that equals 0.83 ms. Yes, less than a millisecond. If you use the (also phase linear) Slow you are at 0.3 ms. The statement 'significant delay with a linear filter compared to minimum' is true for typical EQs in the low frequency range, but not always for higher frequency aliasing filters.

But the main point is - these are just the DA filters. You have to add:

- buffers within the game for audio output
- buffers in the audio subsystem (WDM/ASIO etc)
- buffers in the audio subsystem processing (if not bypassing the kernel mixer)
- hardware buffers in PCIe, USB etc.
- hidden buffers in the driver
- buffers in the audio hardware
- processing the data within the audio hardware

That is everything from 1 sample up to a thousand samples, per item! The total amount for pure playback easily reaches 100 samples even in best case.

Measuring the latency from USB event to analog output is quite complicated - and somehow meaningless, because whatever tool you write will not represent how another program handles this. Using ASIO it gets more easy - but Games don't use that. So let's look at something that is easy to measure: the time between the arrival of the digital data at the SPDIF input until an analog output signal shows up. Doing this with the ADI-2 DAC we get (again in samples):

Code:
       44.1 kHz   96 kHz   192 kHz
SD Sharp  14       13       19
SD Slow   12.4     12.3     18.2
Sharp     36.5     36.6     42.2
Slow      13.8     13.8     19.7
NOS        8.4      8.7     14.7

Compare these values to the pure digital filter values and you will see they are naturally higher. The SPDIF receiver (1 sample) plus internal data processing (6 samples at single speed) are added. And as we measure we now also get values for the Super-Slow (NOS) mode. Here is the above in milliseconds:

Code:
        44.1 kHz    96 kHz    192 kHz
SD Sharp  0.3       0.14       0.099
SD Slow   0.28      0.13       0.094
Sharp     0.38      0.38       0.22
Slow      0.3       0.3        0.1
NOS       0.19      0.14       0.077

But all these numbers have to be taken in context. As mentioned by others the overall latency will be around 5 to 6 ms - if you are lucky. So whatever filter you choose you won't notice any difference. In the ADI-2 DAC's Settings dialog you can choose a buffer size down to 32 samples. That setting is also valid for WDM and your Games, which typically work best at 128 samples.
Thanks @MC_RME. As always, your posts are extremely valuable here.

Completely off topic: Why wouldn't you chose an Avatar proudly showing your brand's logo? Few suggestions:

Screenshot_20190926-173716_YouTube.jpgScreenshot_20190926-205753_Gallery.jpg20190925_174742.jpgScreenshot_20191026-192107_Chrome.jpg
 
Last edited:

q3cpma

Major Contributor
Joined
May 22, 2019
Messages
3,060
Likes
4,416
Location
France
Last edited:
Top Bottom