Jump to content
IGNORED

A novel way to massively improve the SQ of computer audio streaming


Message added by The Computer Audiophile

Important and useful information about this thread

Posting guidelines

History and index of useful posts

Most important: please realize this thread is about bleeding edge experimentation and discovery. No one has The Answer™. If you are not into tweaking, just know that you can have a musically satisfying system without doing any of the nutty things we do here.

Recommended Posts

18 minutes ago, romaz said:

HQP upsampling also benefits from CUDA cores with the Nvidia Pascal GPU as a minimum and the Volta/Turing GPUs being better still and so anyone doing serious DSD upsampling should consider incorporating such a GPU but note that these are power hungry and noisy devices.

 

Do experiment, I found my 1080 GTX on the Ryzen 3900x to leave a negative footprint, and the same on the 6800K but not as noticeable admittedly.

 

20 minutes ago, romaz said:

Force it to upsample to DSD512 and now it's consuming 300+ watts and despite it's very effective passive cooling measures, the Extreme is now running fairly hot.

 

It will very much depend on the setup though.  My Ryzen 3900x running poly-sinc-xtr-mp (non-2s, so full fat) is 125w (SMPS till new HDPlex 500w arrives later this month), with quiet cooling.  If I chose a filter that only uses a few cores sustained whilst playing that could drop to nearer 100w total wall draw. 

 

In Linux there's a useful tool called "turbostat" which helps to show per core utilization, including turbo multiplier usage.

 

HQPlayer Embedded version supports DLNA/UPNP and when NOT using NAA is very very reliable.  In my dual box setups, NAA is not only unreliable (e.g. DAC turns off, it doesnt recover well all the time) but your SQ is held victim to your network's quality too (opticalModules etc help a lot).  

Link to comment
3 hours ago, romaz said:

Force it to upsample to DSD512 and now it's consuming 300+ watts and despite it's very effective passive cooling measures, the Extreme is now running fairly hot.  At these "extremes", the Extreme is probably no longer sounding at its optimum best. 

 

Roy raises an extremely important point, one which I must re-emphasize. All of my — and I believe, his — SQ findings on both custom server/Euphony/Stylus and Extreme/Windows/HQPlayer have been either with bit-perfect playback (no upsampling), or PCM upsampling to 705.6/768, which has a negligibly light CPU load. As far as I'm concerned, all bets are off once you start using these servers for DSD512 with filters like poly-sinc-xtr-mp. I'm not saying there isn't an optimization path for this use case, but it is likely different from what we have been discussing. I certainly do not claim, and I make this very clear when asked, that my optimization journey applies to those on the DSD512 and DSD1024 upsampling journey.

 

On 8/15/2020 at 10:43 AM, hols said:

and running HQplayer embedded + NAA

 

Hols, could you clarify which mode (i.e. PCM upsampling, DSD512 upsampling, bit-perfect) you were running in HQPlayer? And since you mention NAA, were you running HQPlayer on another machine, and only NAA on the Extreme, when comparing with your custom server?

Link to comment
8 minutes ago, austinpop said:

Hols, could you clarify which mode (i.e. PCM upsampling, DSD512 upsampling, bit-perfect) you were running in HQPlayer? And since you mention NAA, were you running HQPlayer on another machine, and only NAA on the Extreme, when comparing with your custom server?

My HQplayer server and extreme are entirely separate. I run Jplay on extreme not using HQplayer in extreme. In my HQplayer server I run DSD256 ASDM7EC which is the EC modulators. It is dsd but it has a more transparent and dynamic sound compared with traditional DSD 512. I use a separate NAA computer to couple with my HQplayer server. Extreme not involved.

Link to comment
8 minutes ago, hols said:

My HQplayer server and extreme are entirely separate. I run Jplay on extreme not using HQplayer in extreme. In my HQplayer server I run DSD256 ASDM7EC which is the EC modulators. It is dsd but it has a more transparent and dynamic sound compared with traditional DSD 512. I use a separate NAA computer to couple with my HQplayer server. Extreme not involved.

 

Just for clarification how did you compare SQ of Extreme vs SQ of your server?

Thanks

 

Matt

"I want to know why the musicians are on stage, not where". (John Farlowe)

 

Link to comment

I guess I am confused over Roy and Rajiv's decision to use HQPlayer after spending HUGE bucks getting an Extreme.  But first:

1)  I am a huge HQPlayer fanboy;

2) I understand the zero-stress (massively over spec'd, under-utilized OS and hardware sq path) philosophy of the Extreme.

 

When one commits to a dual cpu approach, then isn't HQPlayer a non-starter?  Additionally, isn't the fact that the upsampling requirements of HQplayer are diametrically opposed to the zero-stress philosophy of the Extreme (like buying a Tesla Roadster then realizing you need to tow a 60 ft boat/trailer).  I mean, even in the use case of HQPlayer Pro (and using some computer, Extreme or otherwise, to do the offline processing) isn't the Extreme a poor choice as an HQPlayer playback engine, since it doesn't recognize half the processing in the box? 

 

Wouldn't it make more sense to either sell and reinvest in heavy lifting, or use another software player that can see both cpus?  Jussi tries to get around the idea of noisy, grunting heavy lifting by using a separate NAA, but the Extreme is so unbelievably overkill (and still has one hand tied behind its back) for that purpose. 

 

I may be missing something here, but Extreme and HQplayer do not coexist IMHO, unless one were to offline process (HQP Pro) then use another software player optimized for Extreme.  In that case, save some budget for more storage, as now your files are all DSD512.  :)

Link to comment
4 hours ago, guiltyboxswapper said:

HQPlayer Embedded version supports DLNA/UPNP and when NOT using NAA is very very reliable.  In my dual box setups, NAA is not only unreliable (e.g. DAC turns off, it doesnt recover well all the time) but your SQ is held victim to your network's quality too (opticalModules etc help a lot).  

 

I've tried enabling and streaming to HQPlayer Embedded after all of this HQPlayer talk... and so far I feel MPD or BubbleUPNP sounds more transparent, dynamic, and desirable in my setup. I am in no way equipped to run DSD modulators and my DAC would convert DSD back to PCM (Yggdrasil). I have not adjusted any of the HQPlayer configurations though, as all I did within GentooPlayer was enable the HQPlayer service. 

 

Research will commence now...For those not upsampling or converting to DSD with modulators have you tried/compared HQPlayer to a simple uPNP renderer such as BubbleUPNP or MPD?

 

Link to comment
13 minutes ago, ray-dude said:

Ted, remember that Chord-heads like Roy, Rajiv, and I are doing PCM upsampling, not DSD.  PCM upsampling is extremely low footprint from a system resource perspective (I had the actual numbers in my Extreme review where I talked about HQP, but IIRC, it is in the single digits CPU hit)

 

Extreme would not be appropriate for HQP if you're looking for the best DSD filters.  Aside from the difference in philosophy stuff, the CPU's just aren't fast enough to hand the hardcore modulators.  Us PCM guys are definitely outside of Jussi's core use case, but it does have a side benefit of running extremely well on the Extreme.

 


@ted_b
 

What he said. +1

Link to comment
5 hours ago, guiltyboxswapper said:

 

Do experiment, I found my 1080 GTX on the Ryzen 3900x to leave a negative footprint, and the same on the 6800K but not as noticeable admittedly.

 

Yes, agreed.  As I stated in my post, these GPUs are power hungry and noisy devices and so their pros have to be outweighed against their cons.  If your CPU doesn't have enough horsepower, offloading to a GPU can get the job done.  As for my statement that "Nvidia Pascal GPU as a minimum and the Volta/Turing GPUs being better still," I should have given credit where it's due as this comes from Jussi.  While I experimented with GPU offloading briefly, ultimately, it's not something I'm that interested in since I have no interest in DSD upsampling.

 

5 hours ago, guiltyboxswapper said:

It will very much depend on the setup though.  My Ryzen 3900x running poly-sinc-xtr-mp (non-2s, so full fat) is 125w (SMPS till new HDPlex 500w arrives later this month), with quiet cooling.  If I chose a filter that only uses a few cores sustained whilst playing that could drop to nearer 100w total wall draw. 

 

Yes, agreed once again.  SMPSs are much more efficient and run far cooler than linear power supplies.  My post was specifically with regards to the Extreme.  Once again, the point of it was to suggest that the Extreme is not an ideal platform for high rate DSD upsampling.

Link to comment
3 minutes ago, romaz said:

As for my statement that "Nvidia Pascal GPU as a minimum and the Volta/Turing GPUs being better still," I should have given credit where it's due as this comes from Jussi.

 

They're also good for GPUDirect RDMA / GPUDirect Storage

 

https://github.com/NVIDIA/gdrcopy

https://developer.nvidia.com/blog/gpudirect-storage/

https://docs.nvidia.com/cuda/gpudirect-rdma/index.html

 


 

Now that even Comet Lake (10th Gen Intel Core) could support Optane Persistent Memory, what's your view on that when compared to the "regular" Optane storage connected via PCIe / NVMe?

 

I found that ES (engineering sample) version of the 1st Gen Optane DCPMM would actually cost about $1 / GB so pricing ain't such a biggie anymore

 

https://item.taobao.com/item.htm?id=591150702088

Link to comment
2 hours ago, ted_b said:

When one commits to a dual cpu approach, then isn't HQPlayer a non-starter? 

 

Good points, Ted.  According to Emile, the older HQP v3.x can see both of the Extreme's CPUs.  Somehow, v4.x cannot and so something has changed and neither Jussi nor Emile have been able to rectify this.

 

2 hours ago, ted_b said:

Additionally, isn't the fact that the upsampling requirements of HQplayer are diametrically opposed to the zero-stress philosophy of the Extreme (like buying a Tesla Roadster then realizing you need to tow a 60 ft boat/trailer). 

 

Yes, this is the point of my post.  The Extreme is not the ideal platform for DSD upsampling.  As @ray-dude said, there are those of us who use it specifically for PCM upsampling which runs splendidly on the Extreme.  

 

2 hours ago, ted_b said:

II mean, even in the use case of HQPlayer Pro (and using some computer, Extreme or otherwise, to do the offline processing) isn't the Extreme a poor choice as an HQPlayer playback engine, since it doesn't recognize half the processing in the box? 

 

Yes, I would love to rectify this and hopefully, one day, I can get some version of HQP with the sinc-M function to see both CPUs because according to Emile, it should elevate SQ further.  With that said, since PCM upsampling is not CPU resource hungry, even with HQP just seeing one of the Extreme's CPUs, the outcome is excellent.

 

Quote

...but the Extreme is so unbelievably overkill (and still has one hand tied behind its back) for that purpose. 

 

It's easy to think this and I used to wonder about this myself until my Extreme arrived and I heard what it could do with bit-perfect playback.

Link to comment
21 minutes ago, romaz said:

Yes, agreed once again.  SMPSs are much more efficient and run far cooler than linear power supplies.  My post was specifically with regards to the Extreme.  Once again, the point of it was to suggest that the Extreme is not an ideal platform for high rate DSD upsampling.

Oh I realise, I just posted that as a perspective tbh (i.e. other machines don't draw 300w !!)

Link to comment

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now



×
×
  • Create New...