HQ Player

[zaumator]

I try to install NAA on Beaglebone Black. I installed Linux Wheezy.

Updated to latest packages and installed: sudo apt-get install libasound2 alsa-utils

 

When I try to install Networkaudod I get an error message:

 

Unpacking networkaudiod (from networkaudiod_3.1.1-27_armhf.deb) ...

dpkg: dependency problems prevent configuration of networkaudiod:

networkaudiod depends on libstdc++6 (>= 5.2); however:

Version of libstdc++6:armhf on system is 4.7.2-5.

dpkg: error processing networkaudiod (--install):

dependency problems - leaving unconfigured

Errors were encountered while processing:

 

Where can in find "libstdc" files? Or should I try an older Version of NAA?

[One and a half]

Not sure what is going wrong, can you send screenshot of the Settings-dialog?

 

With ASIO backend and non-DoP the "AltDSD" setting should be enabled for PBD DACs. With WASAPI and DoP Also make sure you don't have "2wire" setting accidentally enabled and set "Buffer time" to 100 ms.

 

For the record, screen shot that works, suggestions listened to.

 

[Miska]

I try to install NAA on Beaglebone Black. I installed Linux Wheezy.

 

I recommend going for Stretch. Wheezy is really old. Current stable branch is Jessie and testing branch is Stretch. If you go with Jessie you need at least some modifications to kernel and libasound2.

 

Here you can find necessary files to install Stretch on BeagleBone Black or CuBox-i:

Studenten Net Twente - Index of /debian/dists/testing/main/installer-armhf/current/images/hd-media/SD-card-images/

[zaumator]

I recommend going for Stretch. Wheezy is really old. Current stable branch is Jessie and testing branch is Stretch. If you go with Jessie you need at least some modifications to kernel and libasound2.

 

Here you can find necessary files to install Stretch on BeagleBone Black or CuBox-i:

Studenten Net Twente - Index of /debian/dists/testing/main/installer-armhf/current/images/hd-media/SD-card-images/

 

Thanks Miska! I couldn't find the latest image.

[bibo01]

From FB HQPlayer page:

 

HQPlayer Desktop 3.12.0beta1 now available for download.

HQPlayer now includes nVidia CUDA GPU offload support. To utilize this offload nVidia GPU at minimum Compute Capability 3.0, 1 GB of graphics memory and nVidia driver is required. Latest CUDA driver for Mac can be found here. If suitable GPU is not found, offload is not performed even if requested.

 

Signalyst - Installing HQPlayer beta

[4est]

Yeah! I have been waiting to see this after noticing the facebook notice a month ago.

[bibo01]

There is also a new network icon on the toolbar:

 

When not selected (button up), control connections (Muso, Roon, hqp-control) are allowed only from the same computer (localhost). When selected (button down), control connections are accepted from other devices, such as other computer or tablet on the local network. Setting is remembered over restarts. This for security and convenience of laptop users who frequently use their computers at home and while traveling.

[Miska]

There's also a new check box in Settings dialog to enable/disable CUDA offload. Even when enabled it is still subject to automatic detection that the compute capability level is high enough.

 

CUDA offload is likely to help with multichannel, especially with digital room correction. And also with upsampling to highest DSD rates (best when combined with pipeline SDM).

[lmitche]

Way cool, will a PCIE x1 card meet the minimum requirements?

[Chodi]

There's also a new check box in Settings dialog to enable/disable CUDA offload. Even when enabled it is still subject to automatic detection that the compute capability level is high enough.

 

CUDA offload is likely to help with multichannel, especially with digital room correction. And also with upsampling to highest DSD rates (best when combined with pipeline SDM).

 

So I have a Cuda Nvidia card that is listed as acceptable. I checked the box for Cuda offload. I still get very high cpu usage listed in windows 10 task manager. How can I check that it is really offloading to Cuda? Am I to assume that all is not offloaded but only some portion of the computations? I am going to assume if it was offloading all then my task manager would show some very low number when playing dsd. Now it is showing 59% usage which would certainly not have worked before. I am using an I7 cpu.

 

Just for the record I am assuming it is working because I can now play closed form smoothly in DSD 256 which I could not do before. How can I verify it is doing this offload?

[4est]

I too am curious about this, and what it will take to do 2 channel (closed form?) up to 512. For that matter, what are the CPU requirements along with offloading? I'll likely have to buy a dedicated GPU card, and I am wondering how to go about an upgrade.

So I have a Cuda Nvidia card that is listed as acceptable. I checked the box for Cuda offload. I still get very high cpu usage listed in windows 10 task manager. How can I check that it is really offloading to Cuda? Am I to assume that all is not offloaded but only some portion of the computations? I am going to assume if it was offloading all then my task manager would show some very low number when playing dsd. Now it is showing 59% usage which would certainly not have worked before. I am using an I7 cpu.

 

Just for the record I am assuming it is working because I can now play closed form smoothly in DSD 256 which I could not do before. How can I verify it is doing this offload?

[Chodi]

For whatever its worth I found this:

 

CUDA-Z

 

It is a little program that lets you see what Cuda capable card you have and what it is doing in realtime. In my case it seems to indicate that Cuda is running and constant activity indicates that HQP is in fact offloading something to Cuda.

[Chodi]

I must update my results using CUDA checked. I used GPUZ to verify activity in the gpu. HQP is not sending anything on my windows 10 system to the gpu. I downloaded the CUDA toolkit to re-install CUDA just in case and it solved nothing. So....checking CUDA in this beta version is not using my gpu to process anything. I would be very interested if anyone else had success with this. My Nvidia card is on the approved list. I wondered why I did not see a drop in cpu usage through task manager and now I know why. It isn't working.

[bibo01]

What Nvidia card do you have?

[Chodi]

What Nvidia card do you have?

 

GTX670 2GB approved for sure.

[bibo01]

GTX670 2GB approved for sure.

 

True. It is 3.0 and it has 1344 CUDA cores. Is its driver Win10 approved?

 

I am curious to see Miska troubleshooting this on Win platform.

[Chodi]

Yes the driver is the latest approved for Windows 10. I even download the CUDA toolkit and had it install CUDA just to double check. I have no idea why it's not working. This will require Miska to assist. GPUZ confirms that CUDA is working on my computer.

[Miska]

So I have a Cuda Nvidia card that is listed as acceptable. I checked the box for Cuda offload. I still get very high cpu usage listed in windows 10 task manager. How can I check that it is really offloading to Cuda? Am I to assume that all is not offloaded but only some portion of the computations? I am going to assume if it was offloading all then my task manager would show some very low number when playing dsd. Now it is showing 59% usage which would certainly not have worked before. I am using an I7 cpu.

 

Not everything is offloaded, only the part of the work that benefits from GPU's capabilities. The work that performs better on CPU is run on CPU in parallel with the GPU working on the things it is better at. So you can expect to see somewhat lower CPU load and possibly being able to reach higher output rates.

 

Just for the record I am assuming it is working because I can now play closed form smoothly in DSD 256 which I could not do before. How can I verify it is doing this offload?

 

There's no easy way at the moment to see it. I've been thinking about how to nicely implement indication that offload is being performed.

[Miska]

I must update my results using CUDA checked. I used GPUZ to verify activity in the gpu. HQP is not sending anything on my windows 10 system to the gpu. I downloaded the CUDA toolkit to re-install CUDA just in case and it solved nothing. So....checking CUDA in this beta version is not using my gpu to process anything. I would be very interested if anyone else had success with this. My Nvidia card is on the approved list. I wondered why I did not see a drop in cpu usage through task manager and now I know why. It isn't working.

 

There are reports that GPU-Z load indication stays at 0% even when performing operations on GPU. If the load indication works correctly, then you should be seeing activity there for example when loading or scrolling a relatively heavy web page on Firefox, Chrome or Opera browser (those utilize OpenGL for graphics acceleration).

 

Based on nVidia documentation, full GPU information is available on Quadro and Tesla cards intended for this kind of use. GeForce series is gaming oriented and they limit some functionality on those on purpose (to help sales of the more expensive pro cards).

[Chodi]

There are reports that GPU-Z load indication stays at 0% even when performing operations on GPU. If the load indication works correctly, then you should be seeing activity there for example when loading or scrolling a relatively heavy web page on Firefox, Chrome or Opera browser (those utilize OpenGL for graphics acceleration).

 

Based on nVidia documentation, full GPU information is available on Quadro and Tesla cards intended for this kind of use. GeForce series is gaming oriented and they limit some functionality on those on purpose (to help sales of the more expensive pro cards).

 

To summarize your comments: with CUDA checked using my gpu it is working but it is only offloading part of the load. Due to the exact card that I have you are saying that reporting of the activity on the gpu is not possible. The only way to tell if it is working would be to see a lighter cpu load in task manager.

 

I can say after some experimentation with various filter combinations using DSD 256 upsampling from 16/44 as the benchmark, the difference in cpu load is minimal. A couple of percentage points perhaps. The difference is small enough to not make it easy to quantify. I am only running 2 channel. Perhaps someone doing multi-channel would get more obvious benefit from this. In my case the benefit is not positive enough to justify whatever small offload is occurring. Might also benefit someone with a less powerful cpu.

[Miska]

I can say after some experimentation with various filter combinations using DSD 256 upsampling from 16/44 as the benchmark, the difference in cpu load is minimal. A couple of percentage points perhaps. The difference is small enough to not make it easy to quantify. I am only running 2 channel. Perhaps someone doing multi-channel would get more obvious benefit from this. In my case the benefit is not positive enough to justify whatever small offload is occurring. Might also benefit someone with a less powerful cpu.

 

Yes, it depends on how fast the CPU is compared GPU and how many CPU cores are available. Now when using offload it makes sense to use Pipeline SDM also with dual-core CPU because GPU adds more cores to the system.

 

Most notable difference is when doing closed-form filter to DSD256 or using single-pass poly-sinc (non -2s) to DSD256. Both of these will also require relatively fast GPU.

 

My target is to be able to do at least 5.1 channels or preferably 8 channels DSD256 upsampling to exaSound e28. Target hardware being Quadro M4000 and Skylake i7 CPU.

[Jud]

Based on nVidia documentation, full GPU information is available on Quadro and Tesla cards intended for this kind of use. GeForce series is gaming oriented and they limit some functionality on those on purpose (to help sales of the more expensive pro cards).

 

Years ago I bought an nVidia GeForce card because you could double the number of cores used just by installing the proper driver.

 

I assume they no longer make such mistakes these days.

[Miska]

Years ago I bought an nVidia GeForce card because you could double the number of cores used just by installing the proper driver.

 

I assume they no longer make such mistakes these days.

 

Probably they fuse those parts on the chips or something like that, or then it's done in the card's BIOS. With my GeForce card the nVidia API return "Not available" error in response to querying detailed information. While it works on a Quadro card...

 

Other difference typically between GeForce and Quadro cards is amount RAM.

[4est]

I hate to ask this, but can you elaborate a little on how this interaction takes place. For me the natural implementation for this is on my CAD/CAM station that also presently houses my music files on a Storage Space. It is an older z68 chipset with a i5 2500k. It would max out at using a 3.5 GHz i7 3770k because of it's 1155 socket. I happen to have one that I can transfer to it from another machine.

 

My question then is in a situation such as this, should I look into a GPU that is faster and more powerful to offset the the slower/less powerful CPU, or does that become superfluous and wasted? You spec'd a min of 1Gb of RAM, but does the amount matter or DDR3 vs DDR5 matter over the net Compute Capability score? I ask because there big differences in price on these video cards, and honestly I do not need one for video capabilities. My wait times on that station are minimal. Ultimately if I am wasting my efforts attempting to use this older machine or if I need to spend $800USD on a GPU instead of $400, I'd just as soon learn now as opposed to later. Thank you in advance.

 

Should we perhaps start another thread about hardware for upsampling DSD? It would seem as if there have always been questions about what hardware.

Yes, it depends on how fast the CPU is compared GPU and how many CPU cores are available. Now when using offload it makes sense to use Pipeline SDM also with dual-core CPU because GPU adds more cores to the system.

 

Most notable difference is when doing closed-form filter to DSD256 or using single-pass poly-sinc (non -2s) to DSD256. Both of these will also require relatively fast GPU.

 

My target is to be able to do at least 5.1 channels or preferably 8 channels DSD256 upsampling to exaSound e28. Target hardware being Quadro M4000 and Skylake i7 CPU.

[Miska]

There's no easy way at the moment to see it. I've been thinking about how to nicely implement indication that offload is being performed.

 

OK, I have now implemented a status bar message that is shown at every track switch to inform about offload status. Will be on the next release...