The GPU processing is simply an aid to get more performance out of the computer if required. Like for instance multichannel upsampling to dsd256 or higher with room correction etc running in the convolution engine.
For simple 2 channel resampling of all PCM formats to up to DSD 512, a cheap $500 desktop with Intel I7 4790 processor works perfectly fine.
Haven't decided on this yet. Mine is still at 50k for safety reasons, maybe later, but I cannot see why this needs to be stuck at SACD level for DSD256 or higher.
If I understood this correctly, based on discussion with Signalyst, the more powerful the processor the better the SQ as the algorithms get more juice to spit out things closer to real time. One of the reasons why he (Jussi) is not in favor of offline mastering as it will not take advantage of Moore's Law and other technological advances. Also, judging from the excitement on the other side looks more for 2CH systems
Yes more power may be better. But I've finding using the NAA system, as long as the CPU has enough power not to stutter, the sound is identical from any server. The reason for the better sound only applies when directly connecting the DAC to the server. As the CPU generates more noise that passes along to the DAC. the NAA system, especially with the extra layer of isolation from fiber, kills this detrimental effect.
His argument for online conversion is because that's what his product does. He offline converted 3 files for me and they sounded great. And they played back perfect on my $135 atom processor NUC. Moore's law doesn't matter when you offline convert. It simply takes longer to convert with a lower power CPU. That and the file size, as well as not being able to take advantage of conversion with online streaming services.
Last edited:
Hmmm... It does not make sense to me
regardless of server or direct connect the number crunching is still there! So the underlying concept holds. Stuttering is complete breakdown but even when running nicely something that does processing faster gets closer to real-time performance. Even my i7 MAC stutters at times if pushed too far so looks like not good enough for HQPlayerBtw, if one of your post is anything to go by, when you changed your MACMini/Win7PC to AudioLinux there was significant improvement in the SQ even though the former setup was not stuttering?
NAA is an option to pipe preprocessed data over TCP/IP. So likewise over USB layer. You have seen the benefits of NAA because your DAC has no isolation to start with. Secondly, your USB side is stuck with Linux only. So, NAA and the type of infrastructure around it has become the core requirement of your setup. e,g.your NUC could be replaced by this new Intel Compute Stick or similar mirco gadgets that are beginning to pop up. nobody will know you are using a PC! But you cannot consider this yet because your downstream device runs on Linux drivers for nativeDSD 🙁
.
Last edited:
Hmmm... It does not make sense to me
Btw, if one of your post is anything to go by, when you changed your MACMini/Win7PC to AudioLinux there was significant improvement in the SQ even though the former setup was not stuttering?
You have seen the benefits of NAA because your DAC has no isolation to start with.
Secondly, your USB side is stuck with Linux only. So, NAA and the type of infrastructure around it has become the core requirement of your setup. e,g.your NUC could be replaced by this new Intel Compute Stick or similar mirco gadgets that are beginning to pop up. nobody will know you are using a PC! But you cannot consider this yet because your downstream device runs on Linux 🙁
.
I'm talking about the server computer running the processing. If it's CPU is running at 60%, or 20% there's no audible difference when using the NAA system with optical isolation. I've experimented with 5 different PC's both MAC and windows, all different CPU power. I3 dual core to I7 quad core. I used DSD 128 for the comparison because they could all run it without stuttering. When direct connected to the DAC, they all sound different. Can't tell them apart when used as server in optically isolated NAA setup.
The renderer (NAA) computer is what I changed that caused the sound improvement. Changing that with different hardware will make a difference in sound. this is because that's the hardware connected to the DAC via USB. Everything upstream of that is isolated by the NAA FIFO buffering, and optical isolation. As long as the data is still intact once it arrives at the Ethernet port of the NAA computer, that's all that matters.
Mike, I just updated my post while ago, you have exceeded my real- time capabilities 😀 please check again and hope you can answer my queries.
I'm trying to spec a server pc at the moment. Do you have a feel for what is the minimum spec of processor and memory for this dsd realtime conversion?
Alright then, all noted - thanks! If that is the case it also does not matter if the server side is running on the vanilla Windows or this hot-rod realtime AudioLinux?
James, did you see my response to your post on the 'Best DAC is no DAC' thread? I suggest a processor that benchmarks at around 8000 or better and 6-8Gb of memory should be fine - that's what works for me.
Ray
Because piping DSD256- NAA over network consumes lots of bandwith and you may have to roll out something like a dual network arrangement like Ray has done
Not sure yet. If anything it still is only 29 Euros, and preloaded with HQplayer, and HQplayer embedded. Windows can't compete with that 🙂
The actual reason was my I3 NUC computer struggles with anything over DSD 128 resampling. DSD 256 over 2 channel Ethernet is nothing. 1gb Ethernet can handle 48 channels of DSD 256 streaming simultaneously without a hiccup, with the right hardware. IE Merging Ravenna.
😕 thought it was just streaming the piped DSD256 via NAA. Up sampling is done by your powerful server so the client side will not break a sweat rendering DSD256 so long as the network feed is healthy?
Don't know what optical setup this would be. Is this an ethernet to optical converter at one end and optical to ethernet at the other end?
Would the Twisted Pair Transporter be optimal? I believe it isolates the data lines and can transmit over 100 ft.
I seem to recall ethernet to optical converters can be very pricey.
No that's the Atom NUC that's the NAA. I also have a Broadwell I3 NUC I was using as a server. Works fine up to DSD128. Stutters with 256.
Don't know what optical setup this would be. Is this an ethernet to optical converter at one end and optical to ethernet at the other end?
Would the Twisted Pair Transporter be optimal? I believe it isolates the data lines and can transmit over 100 ft.
I seem to recall ethernet to optical converters can be very pricey.
Yes that's what they are. The twisted pear teleporter is for LVDS. Totally different application.
http://www.amazon.com/MC220L-Converter-1000Mbps-supporting-mountable/dp/B003CFATL0
2 of them and 2 SFP transceivers:
http://www.amazon.com/gp/aw/d/B003C...C_UL100_SR100,58_&refRID=18XCXJPPVF4SCACDACQJ
5 meters fiber optic cable:
http://www.amazon.com/gp/aw/d/B008L...ingle+mode&dpPl=1&dpID=41gutPgxaEL&ref=plSrch
Up to 10km cable length with no loss of quality!!
$130 for the whole setup with the fibre optic cable
Last edited:
if that is the case. I know Ray has got an Atom-based NAA doing DSD256, of course being served by a Xeon hardware with dual network cards
Last edited:
My Atom based NAA could handle DSD 1028 without breaking a sweat. This is because the server does all the work. CPU load doesn't change on the NAA with any resolution.
- Status
- Not open for further replies.