What's new

PC specs for lowest possible latency?

If I were building a networked audio studio today I'd go with Dante, it won the battle with AVB - largely a marketing victory, but what matters is that they won. I don't know that they won the war - AVB is huge in the auto industry, and like almost every other industry, it dwarfs us, even when you lump live sound, theme parks, theatre, etc in.

Are you aware that Samsung bought Harman (who bought Lexicon, BSS, AKG, Crown, JBL and handful of other audio companies)? Why would Samsung buy an audio company? Not for the audio, but because Crown and BSS had some real expertise in AVB. And Dante, and CobraNet, and their own proprietary network as well. When the folks from Harman used to introduce themselves at industry panels they'd call themselves the ABCD company because they had gear that worked on all four networks. But I digress...

AVB is making a small comeback in the professional audio market. I think it will be slow, and I do not think they will unseat Dante, but they could remain a force. So don't discount AVB.

IP over Ethernet, or just plain Ethernet (pick your poison) is the most promising way to get high channel counts at low latency at a reasonable cost. MADI is expensive and complex, PCIe cards are old news, Thunderbolt and USBc are promising, but even smaller niche players than Dante.

In a few years the idea of using multiple Lightpipe based solutions will seem as silly as magnetic tape seems today.

Back to the topic at hand (finally) - if I were Chillbot I'd build a faster machine, but I don't think I'd abandon the current workflow. It works, it is really flexible, and it works. Reducing latency may be more a tweak or two to workflow than expensive hardware. May be. Every studio is different, so these questions often become spirals into uncertainty.
 
If I were building a networked audio studio today I'd go with Dante, it won the battle with AVB - largely a marketing victory, but what matters is that they won. I don't know that they won the war - AVB is huge in the auto industry, and like almost every other industry, it dwarfs us, even when you lump live sound, theme parks, theatre, etc in.

Are you aware that Samsung bought Harman (who bought Lexicon, BSS, AKG, Crown, JBL and handful of other audio companies)? Why would Samsung buy an audio company? Not for the audio, but because Crown and BSS had some real expertise in AVB. And Dante, and CobraNet, and their own proprietary network as well. When the folks from Harman used to introduce themselves at industry panels they'd call themselves the ABCD company because they had gear that worked on all four networks. But I digress...

AVB is making a small comeback in the professional audio market. I think it will be slow, and I do not think they will unseat Dante, but they could remain a force. So don't discount AVB.

IP over Ethernet, or just plain Ethernet (pick your poison) is the most promising way to get high channel counts at low latency at a reasonable cost. MADI is expensive and complex, PCIe cards are old news, Thunderbolt and USBc are promising, but even smaller niche players than Dante.

In a few years the idea of using multiple Lightpipe based solutions will seem as silly as magnetic tape seems today.

Back to the topic at hand (finally) - if I were Chillbot I'd build a faster machine, but I don't think I'd abandon the current workflow. It works, it is really flexible, and it works. Reducing latency may be more a tweak or two to workflow than expensive hardware. May be. Every studio is different, so these questions often become spirals into uncertainty.


A most astute analysis and forecasting of technology.
Samsung ALWAYS gets the edge because they’re not nearly as regulated or restricted as the rest of the world and they buy patents and companies like candy.
 
John, how much latency are you experiencing?

It's like playing a church organ -- not a huge cathedral, but a real pipe organ so that the sound is not instantaneous. Very hard to play percussion accurately, and I write tons of that. Sort of rotten.

Is it because you're monitoring through Pro Tools?

Only indirectly. Pro Tools itself is not the problem; it's the fact that I am not using VE Pro's ethernet for audio. Still lightpipe, which still works with my Digidesign interfaces, which works with old RME cards....etc.

I would have to have a month off.
 
Back to the topic at hand (finally) - if I were Chillbot I'd build a faster machine, but I don't think I'd abandon the current workflow. It works, it is really flexible, and it works. Reducing latency may be more a tweak or two to workflow than expensive hardware. May be. Every studio is different, so these questions often become spirals into uncertainty.
You read it correctly, it's been a struggle. Won't tag him here but so much thanks to Evil Dragon for the help and input.

Latency is one factor that is a big deal to me, but it's not every factor. And I feel that I have a very unique workflow... in that I haven't actually ever met anyone who has the same setup as me. I'm somewhat of a hybrid between old school and current configs. So after much debate we are going with the 9920X @ 3.5GHz. Also going to give the Radeon a chance over the NVIDIA just because I've struggled a bit with NVIDIA so why not give something else a try.

For those that are into that sort of thing, here's the specs we went with, and adding about 6TB of SSD to this as well:

specs.png
 
It's like playing a church organ -- not a huge cathedral, but a real pipe organ so that the sound is not instantaneous. Very hard to play percussion accurately, and I write tons of that. Sort of rotten.

That sucks.

I don't have anything like that here, but I rarely even use two machines.
 
I wanted to ask this in the "buffers" thread but it's already at 4 pages and didn't want to hijack it.

Cost is not an issue. Assume a decent audio card and just focus on the PC. What would you assemble to give yourself the best chance of having super-low latency without pops or clicks.

Just hypothetically. (And because I'm getting a new PC built.)

For an upcoming gig, I'm using one the the HP Z8 workstation. Depending on your needs and budget, the sky's the limit. One single Z8 replaced my 12 cores nMP and 5 x i7/i9 PC slaves.

I'm running PT and CB. DAW buffer is at 64 or 128 when Sonokinetics libraries are involved as they are a hog. When used with VEP pro, DAW is always at buffer 64 and VEP pro is at buffer 1. I use Dante where I can send/receive 128 channels of audio to/from any DAW and VEP.
 
For an upcoming gig, I'm using one the the HP Z8 workstation. Depending on your needs and budget, the sky's the limit. One single Z8 replaced my 12 cores nMP and 5 x i7/i9 PC slaves.

I'm running PT and CB. DAW buffer is at 64 or 128 when Sonokinetics libraries are involved as they are a hog. When used with VEP pro, DAW is always at buffer 64 and VEP pro is at buffer 1. I use Dante where I can send/receive 128 channels of audio to/from any DAW and VEP.

Sounds like a great solution! How much memory and what type of drives and CPU?
 
I'm still testing different workflows. The machine is definitely a beast.
  • Dual 12 cores, 3.2 GHz base frequency | Intel Xeon Gold 6146
  • 512GB (8x 64GB) of 2666MHz DDR4 ECC
  • 512GB NVMe M.2 Solid-State Drive (OS)
  • 5 x 4TB SATA3 2.5" Solid-State Drive (Sounds)
 
I think you'll miss those 5 GHz over all cores with 9900K which can certainly bring an edge with lower latencies, even in spite of having less cores (do note that main audio thread is always calculated by a single core, when that one falls over you get glitches, this is why faster per core speed is more important than number of cores!), but it's your call :) I thought you'd listen to tack if not me at least :P
 
I know. But this way with every pop or glitch I can think *WHY DIDN'T I LISTEN TO EVIL DRAGON* you'll be forever on my mind.
 
So the current best scenario would be 1151 with 9900k at least going to 4.7ghz?
(I don´t know if 5ghz on all cores would make it very noisy?).
 
(I don´t know if 5ghz on all cores would make it very noisy?).

Depends on your cooling solution. I don't think it'd be too bad with some beefy Noctuas.

We were trying to find a solid compromise between the two...

Which is not really compromising the price, since the CPU itself is over $1k... :P

As I said in PM, it's a host machine, you don't necessarily need 128 GB of RAM in it (actually, as it was mentioned in another thread, 9900K will actually support 128 GB of RAM via a BIOS update on certain Z390 motherboards - if RAM was your concern)... I would still go with the highest possible all core frequency.


Question that remains - on your host machine, how many tracks do you tend to have in a project? I assume VEPro takes up a big chunk of mixing unto itself, relieving the DAW from having a too huge track count, yes? If that's the case, there's even less reason to go with a huge core count over larger all core frequency...
 
Question that remains - on your host machine, how many tracks do you tend to have in a project? I assume VEPro takes up a big chunk of mixing unto itself, relieving the DAW from having a too huge track count, yes? If that's the case, there's even less reason to go with a huge core count over larger all core frequency...
My DAW host machine barely wakes up from sleep. I might have 1 or 2 kontakt instances, omnisphere, stylus, zebra, engine, and 15-20 audio tracks. I don't use VEPro in server mode, just send MIDI out to slave computers. RAM was not a consideration for me, I'd probably make do with 16GB. But still I'd like to load taiko creator with less than a 1,024 buffer.
 
Is that all...

At over $20K that's all I can afford for now :)

Anyways, I wanted one machine to replace my main DAW and 4 slaves. Tried different set up including the latest/greatest/fastest i7/i9, unfortunately the performance was nowhere near what I expected.

In my case, multi cores Xeon still outperform i7/i9 by a long shot.
 
Top Bottom