You don’t need to install CUDA toolkit. You only need nVidia graphics driver…
You only need CUDA toolkit if you want to develop CUDA applications yourself.
What GPU do you have? Can you post results of Performance tab of CUDA-Z?
It’s just a Quadro K2000.Its only a 3 on compatibility and only has 384 cuda cores.
How does it perform with CUDA-Z test (Performance tab)?
My GTX980 is 154 GFLOPS double precision float.
You can compare against the CUDA-Z performance figures here:
http://www.computeraudiophile.com/blogs/miska/upgrade-gpu-more-cuda-processing-power-683/
Double precision float is only 28.7.I’m guessing this card isn’t powerful enough ?
If the GPU is slower than the CPU, then it doesn’t really help much… Number of CUDA cores is pretty bogus figure. The true performance needs some testing.
May 27th was sales start for nVidia GTX 1080 which is first normal graphics card with their Pascal generation GPU. (GTX980 is Maxwell)
CUDA 8 has support for Pascal.
Figuring I will get either the GTX 1080 or GTX 1070.
Hi @jussi_laako
Running HQP on Win 10, I have CUDA offloading enabled (ticked) but I’ve been using CAM to monitor GPU usage (thanks @andybob) and GPU usage remains 0%, even running the intensive poly-sinc filter at DSD512 (non short).
Music plays without stuttering (surprising with my i7-7700HQ) but I expected the GPU to still contribute. I tried the non-2s xtr filter at DSD512 just to see if GPU gets engaged (not expecting it to work without stuttering of course) but still no GPU activity…
See attached for my GPU model. It supports CUDA and has the latest NVidia drivers.
Is there anything else to try and get the GPU to be used? What else could be preventing CUDA offloading from being engaged?
I seem to tick all the boxes for HQP CUDA offloading requirements.
Thanks
By your screen shot those are not the latest drivers.
Good spot.
I was going by the latest supported drivers on the Dell site (Inspiron 7567), v390 Mar-2018.
Looking at the GeForce site there is a v397 so I’ll see if that makes a difference with HQP CUDA offloading.
I did a clean install of the latest driver and now HQP has the GPU working at ~30%… with the intensive poly-sinc (non short) filter at DSD512…
Great? Nope, playback stutters.
Turned off CUDA support in HQP and it works fine again.
Jussi has said in the past (on CA Forum somewhere) that sometimes the GPU and CUDA offloading can make things worse if it’s the weakest link.
Might be the case for me with my i7-7700HQ and 1050 Ti.
Be careful what one wishes for in life.
@jussi_laako an interesting observation. This same laptop running your full bootable HQPe can’t play poly-sinc (non short) at DSD512 but the same laptop running HQP on Windows can without any issues and CUDA is disabled on both so that’s not a factor. I would have expected HQPe to work fine and Windows to struggle (less efficient?) but it’s the other way around.
One difference is on Windows it’s using ASIO driver (ProJect S2 DAC) and on HQPe it’s using ALSA.
I’ve tried ALSA buffer of 0 (default), 100ms and 250ms - no difference. With HQPe it’s not stuttering, it’s actually dropouts, every 10 seconds or so. But it’s the same chain that Windows has and Windows plays without any dropouts.
Dell Inspiron 7567, i7-7700HQ , 1050 Ti
OS or drivers don’t matter that much when we talk about computing intensive work. But compiler used to build the binary for the two platforms is different… This will have varying impact one way or the other depending on the CPU model and precise type of load… So one cannot even generalize the result across different hardware.
1 Like
@jussi_laako: you ever thought about a Cuda support for Radeon cards by compile against the Coriander library?