HQPlayer settings to upsample everything to DSD512

[sbenyo]

I may be starting to understand something. I  probably cannot use CUDA offload with poly-sinc-xtr because my GPU is not strong enough. Is this correct?

 

If I try to force CUDA and use 44.1k I see GPU is being used and chocking (gets too high). With 48k, GPU not used, CPU is 99% and system gets completely stuck.

 

I wonder if I change my GPU to something strong like RTX2080, will it help? Is it going to enable me to use full CUDA offload  and will it work well even for 48K?

 

[Miska]

6 hours ago, ddetaey said:

I am still struggling with the difference -  non-difference between the single stage filters and their two-stage variant.

 

Is the end-result of the filter calculation of .e.g. poly-sinc-xtr and poly-sinc-xtr-2s filters the same before sent to the DAC, or is there a difference?

If no difference, why would one use a single stage filter?

 

There is difference, but it is extremely small. It is not that one would be better than the other, but they are just very slightly different. So if one looks at the output samples at enough resolution, there is detectable difference.

 

For ext2 I took different approach and it is always two stage if the conversion factor is  32x or higher. Intermediate rate being at minimum 16x the source rate. This is double the rate compared to 2s filters, but otherwise same logic.

 

For example if you convert something like DXD to DSD64 with -2s or ext2, then the filter operates the same as single stage filter and there's no difference (factor being 8x in that case).

 

Higher the conversion factor, bigger the CPU load difference between single and two stage filters are from there on they approach each other and after reaching the base factor they become the same.

 

 

[sbenyo]

Another question, I can upgrade my GPU from gtx 1060 to rtx 2080 or upgrade my CPU from 1920x to 2970wx (64MB L3).

Upgrade price is more or less the same. Which one is better for hqplayer? Will any of these enable xtr-lp without issues?

 

Sorry for all the questions. I really like to find the right solution and have xtr-lp working with 48k DSD512.

 

[Miska]

2 hours ago, sbenyo said:

Another question, I can upgrade my GPU from gtx 1060 to rtx 2080 or upgrade my CPU from 1920x to 2970wx (64MB L3).

Upgrade price is more or less the same. Which one is better for hqplayer? Will any of these enable xtr-lp without issues?

 

Sorry for all the questions. I really like to find the right solution and have xtr-lp working with 48k DSD512.

 

Are you trying to convert 44.1k source to 48k x512 DSD with xtr? You won't find hardware capable of doing it. RTX 2080 can do the more typical xtr from 48k source to 44.1k x512 DSD.

 

[sbenyo]

1 hour ago, Miska said:

 

Are you trying to convert 44.1k source to 48k x512 DSD with xtr? You won't find hardware capable of doing it. RTX 2080 can do the more typical xtr from 48k source to 44.1k x512 DSD.

 

 

I am not sure I don't think that's what I do. I set bit rate limit to 48k x 512. I also use adaptive rate which I assume prevents me from converting 44.1k to 48k (If I don't use it CPU is 99%) . I still cannot run xtr-lp with 48k family source to DSD512 smoothly. I get "freezes" (music stops for a few ms). It can happen a few times during song play. When song is playing between freezes it's smooth.

CPU is ~40%. I cannot use CUDA offload so it's set to partial. If I try to use it, GPU get too high (even 80-99%) and play is not smooth. 

 

If I use better GPU like RTX, I may be able to use full CUDA offload and maybe avoid the freezes.

 

If I also tried to limit to 44.1k x 512. This way 48k songs play at DSD 256 automatically instead of 512. It's currently the best option to be able to play all rates smoothly  but not always at 512. Other option is to use other filters like ext or ext2 which work well and can also use GPU.

 

My goal is to play smooth with xtr-lp at DSD512 for all rates, at the same rate family.

If RTX can help, at the same price I can upgrade CPU from 1920x to 2790wx. I wonder what would give me the right or better solution (if both may work). 

[Miska]

2 minutes ago, sbenyo said:

My goal is to play smooth at DSD512 for all rates, the same rate family.

If RTX can help, at the same price I can upgrade CPU from 1920x to 2790wx. I wonder what would give me the right or better solution (if both may work). 

 

OK, I see. At least that case works for me with RTX 2080. I don't know about 2790wx but likely that works too. But it would be good to ask someone with similar CPU (2790 or 2990) if modulators run fine on it. With lot of cores, clocks generally go down.

 

[sbenyo]

Thanks. I will try to find out.

These CPUs have larger L3 cache, 64MB vs 32MB I have.  Can it make a difference? 

 

[Miska]

8 minutes ago, sbenyo said:

Thanks. I will try to find out.

These CPUs have larger L3 cache, 64MB vs 32MB I have.  Can it make a difference? 

 

For filters yes, for modulators not much.

 

Note the clock speed difference also between 1920x and 2790wx!

 

[sbenyo]

I managed to loan RTX 2070 super card.

 

Now I am running with full CUDA offload. The results is very disappointing. It's worse than running CPU only.

The CPU usage is low (~15%) and GPU is between 60-75%. Both 48k and 44k rates have freezes. It plays as expected for a few seconds and then it stops for a few milliseconds and continues.

 

The freezing is much worse this way than with CPU only!

 

Is it possible RTX 2070 is not good enough? Is there a way to increase CPU usage and also do some CUDA offload?