UAD still worth it?

[Burillo]

People think computers are just super-fast calculators, and math is math.

But surprisingly this is not so. Floating point vs integer math is totally different. DSPs may have specific routines that are particularly suited for audio processing which a floating point CPU can only approximate.

This is partially where these myths come from. Whether or not anyone can actually perceive those differences may be another story, though.

Can you give any specific examples of instructions available on a DSP chip that are not possible to reproduce on a general purpose CPU?

[Scotty]

A distinction without a difference and a post carefully written to say nothing.

[noiseboyuk]

My thread summary - there's bugger-all difference between most of the good companies.

Buy whatever the best deal is from a company you like.

[bitman]

I was an old UAD-1 user when they quit supporting it. I quit em, and now don't miss em.

[theviirus]

DSP is still useful. See: Apogee, Antelope Audio, Avid, Motu.

DSP is very useful because once you start piling on the hungry plugins it gets harder to keep that ASIO RTL at low levels. Its even worse when you're running softsynths live.

DSP can be avoided if you use more hardware instruments and effects. Just one option.

considering the bill for an RMX16, a 224, a 480L, a Voxbox, a Massive Passive, an MXR flanger, a Culture Vulture, a Studer, and the UA 175B/176 pair would easily push above $30k…i’m more than happy having these as software (as if I had the room in my life to house or maintain these anyway).

We can debate the accuracy of analog modeling and non linearity for ages and I just don’t care to, but as much as I love the RMX16/224/480L units, i’d be ok if I never had rEaL ones. UA was given access to the original source code by AMS and Lexicon so any concerns of “accuracy” fall into the realm of “id rather care about my music than care about AD/DA converter accuracy here.”

[vitocorleone123]

DSP is still useful. See: Apogee, Antelope Audio, Avid, Motu.

DSP is very useful because once you start piling on the hungry plugins it gets harder to keep that ASIO RTL at low levels. Its even worse when you're running softsynths live.

DSP can be avoided if you use more hardware instruments and effects. Just one option.

considering the bill for an RMX16, a 224, a 480L, a Voxbox, a Massive Passive, an MXR flanger, a Culture Vulture, a Studer, and the UA 175B/176 pair would easily push above $30k…i’m more than happy having these as software (as if I had the room in my life to house or maintain these anyway).

We can debate the accuracy of analog modeling and non linearity for ages and I just don’t care to, but as much as I love the RMX16/224/480L units, i’d be ok if I never had rEaL ones. UA was given access to the original source code by AMS and Lexicon so any concerns of “accuracy” fall into the realm of “id rather care about my music than care about AD/DA converter accuracy here.”

I just use hardware synths and some effects and then add a few plugins like that. I don’t need a massive passive hardware unit or plugin. That’s all I’m saying - DSP isn’t needed. It can be avoided. It’s very straightforward if someone’s computer is lagging: get a new computer or use more hardware, or both. No judgment, was just stating facts.

[lfm]

Floating point is more precise that integer.

32-bit float I believe is 7 digit accuracy.
- means it can represent 9 999 999 integer part and no decimals.
- or it can represent integer 0 and 7 decimals

64-bit double precision 17 digits.

A signed 32-bit integer can represent +/- 2 147 483 650 which is 10 digits. Todays 64-bit windows(amd style) still have 32-bit integers but believe there are true 64-bit registers as such in cpu. They can run true 64-bit as IA64 architecture.

if you divide an integer and end up with results or intermediate result close to zero, you loose all the rest of information. It is truncated.
- if you multiply this elsewhere in a formula you lost all information just about

With float you still have all the digits accuracy like 17 digits as double float.
- make intermediate calculations and have results that you later multiply with some other number you still have 17 digits, even if close to zero in between. It's all left in decimals.

If this can help see how it works maybe.

A double precision float number is 8 bytes long, 64-bit and is nicely fit into a processors registers. And math processor in todays cpu's do this just fine without any UAD dsp processors.

Are these SHARC processors in UAD better than 64-bit float, I don't know. But if so in theory it could be more precise and sound better for specific calculations. If audible I have no clue.

Native versions are likely just as good.

But SHARC may have real long shift registers to delay stuff, not sure how architecture is. 128 bit registerns, 256, what?

Normal 64-bit cpu most likely has instructions to handle two 64-bit registers as one 128-bit register. I know 32-bit cpu did have for 64-bit length so assume that is the case.

Maybe SHARC does have an upper hand here, or not. But if actually audible in audio that's another matter and up to creators of algos.

Just some thought on the matter....

EDIT: thanks gearwatcher for correction. Updated....

[gearwatcher]

it's not SHARP, it's SHARC and no, back in the day of UAD cards development they were still only 32 bit floating point.

Moderne ones are SoCs with ARM cores for the general Von Neumann stuff, and ARM cores can do 64bit IEEE754, but the SHARC dsp cores are, AFAIK, still 32bit

https://www.analog.com/en/product-categ ... -socs.html

[Burillo]

I don't see how anything you said makes reproducing this math on a general purpose CPU impossible - just that it might be more difficult than a simple instruction. You can do shifts like that without such wide registers.

[Dirtgrain]

Nerds!

[sprnva]

Nerds!

https://www.youtube.com/watch?v=dD3XbBA_DF8

[Uncle E]

We can debate the accuracy of analog modeling and non linearity for ages and I just don’t care to, but as much as I love the RMX16/224/480L units, i’d be ok if I never had rEaL ones. UA was given access to the original source code by AMS and Lexicon so any concerns of “accuracy” fall into the realm of “id rather care about my music than care about AD/DA converter accuracy here.”

Plus, the difference aren't necessarily worse. It's so nice that you can turn the UAD Lexicons' noise off. I still like the hardware because I'm of a certain age (please make every singer sound like Lauryn Hill) but I doubt anyone else cares.

If you haven't tried it yet, the 224 Atmosphere preset is stunning on synths.

[Scotty]

BX codes plugins for the UA platform and also released several of them natively. Apart from being able to run scads of them natively on a modern processor (far more efficient than dsp) they null when compared to their dsp versions. I asked these questions previously and their response was there are no sonic differences between the dsp and native versions confirming reports from others that these are identical. DSP has some advantages in terms of real time performance and UA algorithms get a boost from inclusion in their interfaces.Their unison technology is also a value add in that context.

[HREQ]

UAD versions of the 1176, LA-2A, LA-3A, Pultec, 224, Dimension D, Fairchild, and Space Echo are exceptional, and they're so good that I'm not searching for better, because UAD's versions are already there.

jamcat just seems to have something against UAD, and it's not clear to me why.

I have nothing against UAD. Earlier I said UA are one of the few developers who really care about accuracy.

I’m just realistic. UAD plugins have the same limitations as everyone else. They’re always good, but it’s just not true that there aren’t others that are as good or even better. And they really can no longer justify the disparity in price.

Audiority’s Chorus Ensembler is better than UAD’s CE-1 model. Arturia and IK’s 1176 models are as good as UAD’s. But UAD’s 1176 can do any combination of buttons, not just “all in.” That’s what I mean about their attention to accuracy.

IK’s LA-2A is as good as UAD’s, except that IK didn’t include the R37 pre-emphasis, because IK just doesn’t care about accuracy as much as UAD.

UAD’s Pultec is clearly better than IK’s, but not NoiseAsh’s. And the Lindell Audio plugins are on par with UAD.

IK’s Hammond/Leslie are better. IK’s Studer also edges out UAD’s. But it takes 3x as much CPU for a 3% improvement.

And of course Relab’s Lexicon 480L is head and shoulders above UAD’s.

The models I am most interested in from Universal Audio aren’t available for Spark. Things like the Helios channel strip.

Thanks for list, it

I'd be interested in an extensive list of "UAD Alternatives" that on on par, I currently have PA Everything bundle for 2022, as well as some waves and IK everything deal from anniversary sale.

I wonder if one exists? Cause I'd rather not go down the UAD route if alternatives get me just as good results. Besides it seems it'd be a waste since UAD is porting it all to spark sub soon lol

[Funkybot's Evil Twin]

I'd be interested in an extensive list of "UAD Alternatives" that on on par, I currently have PA Everything bundle for 2022, as well as some waves and IK everything deal from anniversary sale.

I wonder if one exists? Cause I'd rather not go down the UAD route if alternatives get me just as good results. Besides it seems it'd be a waste since UAD is porting it all to spark sub soon lol

Seek and ye shall find:

viewtopic.php?t=355566

...there's a Gearspace thread from that time. More recent KVR threads. YouTube videos. There's lots of stuff out there.