Сводная таблица HQPlayer - Little experience in Windows performance testing:
Test Configuration
Hardware
Music Server - Mac mini (Mid 2011), i7 2GHz, 4G DDR3-1333
Control Point for iPad mini, MacBook Pro
Processor PC - Asus X99-M WS, Intel i7 6900K, Corsair 32GB (4x8GB) DDR3-3200, Samsung 960Pro 256GB M.2, Asus GeForce GTX 1080 Ti
NAA - NUC Celeron N2820, 4GB RAM
DAC - T + A DAC 8 DSD
Software
Music Server OS - macOS the Sierra
Processor PC OS - Windows 10 64bit
NAA OS - Windows 10 PE 32bit
Music Library - Roon
Upsampler - HQPlayer
Control Software - Roon Remote
hardware connection
Music Server -> Network Switch -> Processor PC -> [Network Switch - > NAA] -> USB DAC
Music Path
16 / 44.1 CD-ripped file -> Roon -> Network Stream -> HQPlayer (Upsample to DSD512) -> [NAA] -> ASIO -> DAC
Test Content
The main test different Filters and different Hardware configuration performance respectively, set by the Redbook format music files (16 / 44.1) up to DSD512, using AMSDM7 512 + fs as a Modulator, CPU and GPU usage as a percentage. When the music can play normally, it is marked with green, while the music is stagnant or the broken sound appears marked with red. This test is not a rigorous test, and does not represent you to obtain the same result, only as a reference.
The Windows system has been optimized to some extent with Processor PC controlled within 50 processes, NAA PC controlled within 32 processes, Processor PC using only 2GB memory and CPU overclocked 4GHz. As NAA OS use Windows PE, Windows system can boot through the network, the entire system running in memory, it can work in the absence of Hard Disk.
Results Analysis
From the above comparison table, it can be concluded that poly-sinc-xtr (mp) uses the highest CPU / GPU resources, and in some cases, it is harder to play and closed-form / closed-form- - * - 2s (mp) uses the least amount of resources and poly-sinc (mp) / poly-sine-short (mp) is initialized for a longer period of time (more than one minute) .
Using CUDA offload (ie, GPU) can effectively reduce CPU usage, and can effectively help improve the quality of sound when encountering filters that use a lot of processing resources. However, there are some cases where the GPU fails to make significant improvements. This is because Filter uses only a lot more CPU resources and has fewer opportunities for effective assistance from the GPU. Therefore, special cases arise.
On the other hand, NAA can really improve some Filter, but also affect the operation of the CPU, although to some extent can reduce the CPU usage, but also increase the CPU on the network. As for the use and not depending on the circumstances, if the Processor PC fan noise or the use of switching power supply, it is recommended that the computer should be far away from the sound system, then use NAA more suitable.
If you only use poly-sine - * - 2s filters, I believe most i7 and low-power computer can cope with, if you need to use more complex Filters, GPU is a good way. But in fact, this test there is a very serious problem, that is, Monitor software itself will affect the CPU performance, that is, the contents of this test there is an error, this error may affect the normal playback, but in order to get the relevant figures, but can not but use Related software, I hope you understand that the test content can not fully reflect the actual situation.