I have been messing around with a Sound Devices Mix Pre6 for a couple months.
Upcoming 32-bit float Audio Interfaces in 2024
- Thread starter Tranasaurus Reks
- Start date
Help Support GroupDIY Audio Forum:
Too early to decide anything yet, to be fair I have barely used it so far. We have the usual suspects here to compare them to, digi, antelope, mytek, UA, etc. Just have not had time to really play with it and compare to anything. Really will not have time until October or November. The 32 bit thing caught my attention, kind of the reason I wanted to check it out, but I also needed a super compact quality device with 4 inputs minimum for field work.
Wavelength
Well-known member
Gang,
There are a lot of new AD converts being made now from AKM and ESS. unlike early ladder type units these new AD converters use digital filters which extend the real conversion into 32 bit territory. I don't totally agree with people much smarter than me... but most determine the worth of the AD chip with signal to noise ratio (SNR). Most of these 32 bit units are still in the -120'sdB range and some ekking out low -13xdB range, were -144dB would signify 24bits.
To me the use and type of the digital filter along with the analog pre-circuitry makes all the difference in the world. That really can't be measured in SNR because it's a type of sound.
~~~
Any interface can output floating point over USB, Thunderbolt or Network. Is that conversion cost money. No not really! Does it take away from the conversion? Well it can since 32 bit float is 24 mantissa + 8 bits float. The real reason people don't include it is really PCM is easy and a full measure. The only time you need float32 is when you are combining a ton of calculations and filters it makes it much easier and provides a wider result. I remember some company did a complete DAW using 32 PCM so no float the idea was the calculation took much less time, much less latency and sounded better. Who knows???
~~~
I do a lot of work with computers at a low level. I have written device drivers since 1981 on a ton of platforms from assembler, C, C++ and now C# (actually prefer assembler and C the new stuff is too distant from what comes out). Therefore I do T state calculation on drivers i.e. how long does this procedure take.
Setting up a DAW system is interesting and I gave a talk to engineers at Vintage King ounce and got a lot of feedback. RAM is something you can never have enough of! If you don't have enough the system will use virtual memory which means memory is swapped to disk and back, which is very bad. Also watch where you put your projects and so forth not to put anything on the system disk except for the applications. Also never put another drive with projects on the same controller as the system disks. I like to say keep your lanes clear. If your using a USB interface don't put a drive on USB, instead use a thunderbolt drive. SSD over rotating is always a plus. Go to OCW (macsales.com) and grab a Thunderbolt enclosure and throw an SSD in it. There cheap now and will work better.
Enough for now, all have a good weekend!
Gordon
There are a lot of new AD converts being made now from AKM and ESS. unlike early ladder type units these new AD converters use digital filters which extend the real conversion into 32 bit territory. I don't totally agree with people much smarter than me... but most determine the worth of the AD chip with signal to noise ratio (SNR). Most of these 32 bit units are still in the -120'sdB range and some ekking out low -13xdB range, were -144dB would signify 24bits.
To me the use and type of the digital filter along with the analog pre-circuitry makes all the difference in the world. That really can't be measured in SNR because it's a type of sound.
~~~
Any interface can output floating point over USB, Thunderbolt or Network. Is that conversion cost money. No not really! Does it take away from the conversion? Well it can since 32 bit float is 24 mantissa + 8 bits float. The real reason people don't include it is really PCM is easy and a full measure. The only time you need float32 is when you are combining a ton of calculations and filters it makes it much easier and provides a wider result. I remember some company did a complete DAW using 32 PCM so no float the idea was the calculation took much less time, much less latency and sounded better. Who knows???
~~~
I do a lot of work with computers at a low level. I have written device drivers since 1981 on a ton of platforms from assembler, C, C++ and now C# (actually prefer assembler and C the new stuff is too distant from what comes out). Therefore I do T state calculation on drivers i.e. how long does this procedure take.
Setting up a DAW system is interesting and I gave a talk to engineers at Vintage King ounce and got a lot of feedback. RAM is something you can never have enough of! If you don't have enough the system will use virtual memory which means memory is swapped to disk and back, which is very bad. Also watch where you put your projects and so forth not to put anything on the system disk except for the applications. Also never put another drive with projects on the same controller as the system disks. I like to say keep your lanes clear. If your using a USB interface don't put a drive on USB, instead use a thunderbolt drive. SSD over rotating is always a plus. Go to OCW (macsales.com) and grab a Thunderbolt enclosure and throw an SSD in it. There cheap now and will work better.
Enough for now, all have a good weekend!
Gordon
Absolutely. This is so important. I NEVER put any important data on the same drive (or partition) as the operating system and applications. I can backup and restore my OS drive in about 3 minutes using a proper drive image program and external drives. My data is safely tucked away on a completely separate drive and that gets backed up to external drives. I disconnect the external drives to keep them safe from whatever happens on the main box, and to keep them from interfering with my transceivers and audio gear. But I NEVER keep my data on the OS/applications drive. I learned this as a beta test site for major software producers (IBM, Lernhout&Hauspie, Kurzeweil AI, Microsoft, DragonSystems, Nuance, and others.)
This turns on how well your motherboard is built, what CPU, chipset and etc. you are using. No problems here, but I can imagine it could be an issue on a case-by-case basis.
This may be overkill, but it may be good advice on some computers. It certainly could not hurt. I do, however, agree it is best to keep potentially conflicting USB devices on different USB channels on the motherboard. And I NEVER plug any significant USB device into a hub - I always connect it to a connector with a direct route to the mobo. I also turn off USB power down and other power saving features in the OS which can raise havoc with all sorts of applications as the system tries to save energy powering-down USB and other ports after x minutes. Using highest available baud rates can improve throughput and avoid stuttering. Lots more, but I am with Mr. Wavelength - how you set your system up matters.
Just MY take. James
Wavelength
Well-known member
James,
The USB thing is rather simple. If you are reading from the AD and writing to a drive both on USB then your synchronous operation is going to run into problems (one IN, the disk OUT). USB Audio Class has a higher priority over the disk and therefore it can queue up and cause a real problem. When we were working on the UAC and UAC2 protocols we saw this happen a number of times. Sure if you have an i9 Windows or a M3 Studio with an SSD, probably not going to be a problem. BTW on the Studio there are 2x PCIe USB controllers the Thunderbolt have one and then the USB-A and front USB-C run on a different one. If you have a Studio BTW the second one had a big software update recently.
Motherboard controllers sharing with the system disk is a problem. The system disk has the highest priority and anything else connected to that controller has the lowest. Now PCIe disks are all separate so that's not a problem but SATA and others which have dual controllers in one should be avoided.
Thanks,
Gordon
The USB thing is rather simple. If you are reading from the AD and writing to a drive both on USB then your synchronous operation is going to run into problems (one IN, the disk OUT). USB Audio Class has a higher priority over the disk and therefore it can queue up and cause a real problem. When we were working on the UAC and UAC2 protocols we saw this happen a number of times. Sure if you have an i9 Windows or a M3 Studio with an SSD, probably not going to be a problem. BTW on the Studio there are 2x PCIe USB controllers the Thunderbolt have one and then the USB-A and front USB-C run on a different one. If you have a Studio BTW the second one had a big software update recently.
Motherboard controllers sharing with the system disk is a problem. The system disk has the highest priority and anything else connected to that controller has the lowest. Now PCIe disks are all separate so that's not a problem but SATA and others which have dual controllers in one should be avoided.
Thanks,
Gordon
And yet, while I appreciate your take - really I do - I have never heard about or encountered these issues - perhaps because I use AMD based boards - I not only use latest PCIe 4 SSD drives - I also have multiple SATA SSD drives in all my computers and have never encountered any of the issues you mention.
Seriously - and I am NOT being snarky - just how big a problem, how often does this occur?
James
That sounds like you are remembering the old ATA controllers which had two disk connections with implicit prioritization. I don't think that has been true of SATA controllers for many years (although with the caveat that I typically deal with server and workstation class hardware, so it is possible that some desktop chipsets inherited that behavior for longer than necessary).
I teach at a Music college and run a multi-room commercial studio and I see issues weekly with users who record to the system disk (especially the desktop). I also see issues with bus-powered low-cost drives, especially on laptops with only one or two USB-C ports, especially when using docks or hubs. I do see dozens of users per week, so I get a pretty large sample of users and problems. Laptops are obviously much more prone to these issues than desktop systems.
For desktop systems, I recommend SSD drives for daily use and spinning drives for longer term storage- with the caveat that the storage drives be spun up and cataloged at least once every year. I recommend Neofinder for this (Mac OS).
capacitorless
Well-known member
I have a certain amount of experience with the (relatively ancient) Creamware/Sonic Core Scope DSP audio system. It uses fixed 32 bit processing and imo sounds pretty great still. Back in the day, the thinking was that using 32 bit integer doing internal calculations was part of that sound. Of course, the ADC/DAC was 24 bit.
Just wondering if 32int would sound better than 24+8 in the digital filtering part of conversion.
Just wondering if 32int would sound better than 24+8 in the digital filtering part of conversion.
Last edited:
Wavelength
Well-known member
James,
As I said this doesn't effect PCIe drives because in a sense they each have their own controller.
Say you had a SATA based motherboard (no PCIe) and the controller handled 2 drives. Drive0 would be the system, using Drive1 for projects would cause problems because Drive0 has priority and almost always in use with DAW software.
~~~~~~
In general integer math though is light speeds faster in the # of t-states it takes to do the math. Less computations, lower system usage, lower power required which always equals lower noise.
Thanks,
Gordon
As I said this doesn't effect PCIe drives because in a sense they each have their own controller.
Say you had a SATA based motherboard (no PCIe) and the controller handled 2 drives. Drive0 would be the system, using Drive1 for projects would cause problems because Drive0 has priority and almost always in use with DAW software.
~~~~~~
Audio playback systems like Audirvana and others do this and it does sound much better. The obvious problem with fixed vs float is summation and filtering. Any type of filtering or other digital processing requires tables of coefficients. These are in a table the number of rows called taps. A sample is multiplied by it's corresponding coefficient and summed and the summation over the number of taps is the result. All the coefficients are less than 1 or fractional. Early DSP's had integrated divide by so (sample * coefficients')/x = output. Where x was usually some multiple of 2 (2, 4, 8, 16....) which could easily be done by shifting the number x# of places (i.e. divide by 2 is shift right 1, 4 = 2 and so forth).
In general integer math though is light speeds faster in the # of t-states it takes to do the math. Less computations, lower system usage, lower power required which always equals lower noise.
Thanks,
Gordon
