LowOrbit Posted February 9, 2016 Share Posted February 9, 2016 Hi I don't want to derail any ongoing discussions. I am currently enjoying the trial version of HQP on Windows 10/AObeta and am very encouraged. I will be pressing the purchase button soon. The PC I am running HQP on is a i7 3770k cpu machine and using some of the more intensive interpolation/filter combinations I get occasional signal dropouts at DSD128 and DSD256 (outputting to a Auralic Vega via a second PC running NAA). Does anyone yet have experience of CUDA offload? Would this likely stop the stuttering/drops? I am open to adding a suitable GPU but my concern is fan noise. Both my audio PCs have passive heatsinks and are effectively silent and I am worried that a useful CUDA GPU will add lots of cooler noise. Does anyone have any experience (I guess Miska does!)? Mark Link to comment
Popular Post LowOrbit Posted July 29, 2019 Popular Post Share Posted July 29, 2019 Hi I've been following this discussion around CPUs for EC modulators. I'm a longtime user of HQP, now on v4 Desktop, running into a Beaglebone Black/botic NAA and Twisted Pear 9038Pro Buffalo dac. I don't post much on fora these days but I just built a Ryzen 3700X pc for audio purposes. Using a MSI X570 mobo, 16gb 3200 DDR4 ram and a beefy 700w ATX PSU, no CUDA support, basic gpu. My experience over the last two weeks is this: I can run upto ASDM7EC with any filter (poly-Sinc-XTR or Sinc-M are preferred) upto 48k x 128 without issue and it sounds fantastic. The EC variant modulators are a real step forward for SDM playback. Going up to 44.1k x 256 rates I get varying degrees of stutter introduced even though resource monitoring in Windows 10 shows some headroom on all 8 cores. I have the multi-core CPU box greyed out in HQP, Auto Rate Family enabled settings, and multithreading disabled in bios. I have played around the overclocking profiles but nothing will get me to the point where x 256 rates will play glitch-free. Sometimes I get maybe 10 bars into a track before the hiccups begin, and they are sometimes very brief but enough to kill the music of course. It seems that HQP is heavily loading 4 cores with modulator activity (changing to low intensity modulator variants shows big reductions in core loading) and that's the bottleneck. I am not sure if adding a GPU (RTX2070 Super are looking like a good bet pricewise and for general performance) will help as Jussi has said that only the filters (and convolution processes, iirc, which I don't use) are suitable for the high level of parallel processing the GPU adds. It could be that offloading the filter processing will reduce the thermal load on the CPU enough to deliver a few more CPU cycles, and kill the stutter, but I am sceptical right now. Maybe going to a 12 core 3900X will be the way to go. Maybe it just needs a 5ghz 8-core Intel Coffee Lake Refresh cpu. Just thought someone might find my recent experiences useful. k6davis, Rune and rickca 3 Link to comment
LowOrbit Posted July 31, 2019 Share Posted July 31, 2019 4 hours ago, rickca said: @k6davis thanks those are very useful results. I had been considering i7-9700K and Ryzen 3700X. The 3700X is particularly attractive because it's 65W, but I'm not sure whether 3700X will outperform 2700X enough to change the Intel/AMD decision in your use case. I've looked at the published benchmarks, but they may not be representative of HQPlayer performance. Ryzen 3700x has twice the L3 cache vs 2700X, but I'm not sure that will do the trick. Hi As per my post from a couple days ago - I have experimented extensively with Ryzen 3700X on X570 mobo. It seems it doesn't improve on your 2700X as far as running EC modulators. I can almost get 44.1 x 256 SDM running, but never totally smoothly. Always quite brief glitches. I can run any filter with ASDM7EC at DSD128, and that seems the sweet spot for reproduction. Sounds terrific. Looks like the AMD chips just don't boost to high enough core speeds for a stable period of time to deliver the DSD256 experience we crave with the current HQP code. Going back to Intel may be the only way right now. Good to hear that the 9700K can hit the target, even if it needs some GPU assistance for the heavier filter loads. rickca 1 Link to comment
Popular Post LowOrbit Posted July 31, 2019 Popular Post Share Posted July 31, 2019 49 minutes ago, k6davis said: EC is beautiful Ted. It's hard to describe without going into the usual audiophile cliches, but there's a significant increase in clarity on transients with absolutely zero harshness. There's better imaging too. It takes you deeper into the music. I just switched back to non-EC DSD512 when I read your question and while I used to think it was the pinnacle, it seemed a bit flat by comparison. I wanted to switch back immediately. I'll take it even further. With the AMD chip, I could do non-EC DSD512 all day but was limited to EC DSD128... and I still listened in EC! Jussi has been really modest about it. This sums up my experience very nicely. I am not bothered about DSD512 - my primary DAC only runs upto DSD256, and I get so close to EC DSD256 with my setup that it is tantalisingly close. I prefer EC DSD128 over non-EC at any rate. It's both more correct and more musical, bass is cleaner, deeper, transients come across more potently and voices are clearer. Those songs with the odd mumbled or indistinct lyric - now much more intelligible. And the tonal variation and truthfulness (tenor sax, strings, percussion sound richer, more alive). I downloaded HQP-Pro, did the 1 minute sample thing, learned little, went back to listening to music. Le Concombre Masqué and k6davis 2 Link to comment
Popular Post LowOrbit Posted August 15, 2019 Popular Post Share Posted August 15, 2019 11 hours ago, Luca72c said: Did you try disabling SMT on 3700X? Were your test conducted using Miska's multicore optimised EC modulators version or first, non-optimised version? More, nobody tried 3900X with hqplayer yet? Yes, most of my testing was with SMT off. Tried Creator Mode, Game Mode and set up manual OC. Nothing got me to DSD256 with ASDM7EC without stuttering. All other Modulators - fine. All Filter variants fine (at least at redbook source rates). I am using Desktop 4.1.0.1. I sold on the AMD setup and sunk my pennies into an i7 9700K on an Asus Tuf Z390 mobo. I can run DSD256 ASDM7EC, with all filter options except Closed Form-16M (I don't generally use the non-apodizing filters, just an experiment). CPU runs cooler than the AMD chip, no need to faff about with bios setting, just assemble and play. In reality I don't hear a real difference stepping up from DSD128 rates with the ESS Sabre9038Pro dac, but mine is not a typical setup and already performed at a very high SQ anyway. k6davis and asdf1000 2 Link to comment
LowOrbit Posted August 15, 2019 Share Posted August 15, 2019 1 hour ago, Rune said: Did you get over the critical 4Ghz with your manual overclock of the 3700X? My 3700X was stable upto 4.5ghz - made no difference. I suspect the early bios is not giving the chip full rein, but I wasn't going to wait around for AGESA to catch up. Intel delivered out of the box. Link to comment
Popular Post LowOrbit Posted August 15, 2019 Popular Post Share Posted August 15, 2019 48 minutes ago, asdf1000 said: No GPU (CUDA offloading)? Not bought a GPU yet. Probably no need to for HQP. asdf1000 and k6davis 2 Link to comment
Popular Post LowOrbit Posted August 15, 2019 Popular Post Share Posted August 15, 2019 1 hour ago, asdf1000 said: When using HQP, is the machine using any other programs at the same time? Or only HQP? Just HQP. It's set to autostart and I use HQPD Control on a Sony android tablet to control playpack. Music library is local on a multi-SSD Storage Space under W10 (1903). CPU Utilisation is never over 75% (averaged over all cores). Temps peak around 65c. 9700K is watercooled with a CoolerMaster 240. 16GB DDR3200 ram. NAA on Beaglebone (Botic kernel). k6davis, luisma and asdf1000 3 Link to comment
LowOrbit Posted August 16, 2019 Share Posted August 16, 2019 31 minutes ago, luisma said: Thank you @LowOrbit for posting your results. There are some problems with Ryzen in regards to full performance on every core https://www.tomshardware.com/reviews/amd-ryzen-3000-turbo-boost-frequency-analysis,6253-4.html IMO that should not prevent the CPU to deliver some good processing, Turbo for the intel is 4.9 vs 4.4 for the 3700X, Miska have said before and we know the speed per core have more influence that the actual number of cores although for the new modulators it seems the most cores the merrier. I haven't been able to test, I'm in the process of getting a new Ryzen and the practical results of LowOrbit got me a little discouraged. You are using Windows 10 correct @LowOrbit? WIndows 10 - yes. I had an MSI X570 motherboard for the Ryzen cpu - the bios available from MSI did not have the latest AGESA microcode updates from AMD, so it may have been a factor. The 3700X was stable at 4.5ghz, could run any filter and the EC modulators but I could not get past fairly frequent short interruptions at DSD256. No issues at DSD128 with ASDM7EC and it did sound damn fine. With non-EC modulators I could run at DSD512 without issue (using a DSC1/Amanero lashup). The problem may have been motherboard specific, but I sold the Ryzen and mobo on and the 9700K is performing well without any overclocking or messing. Link to comment
LowOrbit Posted August 16, 2019 Share Posted August 16, 2019 2 hours ago, Rune said: You probably only had on core at 4.5Ghz. If you had done an all core overclock to 4GHz it would most likely have worked. That may be the case but was not what Ryzen Master was showing. Manual set all cores. And believe me, I tried many configurations and combinations to try to isolate and offer best performance for the cores which deemed to load heaviest which I assumed to be the ones carrying the Modulator processes. Anyway, academic now. Not an experiment I will readily repeat. Link to comment
LowOrbit Posted March 8, 2020 Share Posted March 8, 2020 I've been leaning on the convolution feature of HQPlayer Desktop lately, and was wondering if there is an optimum impulse format? I have mostly redbook derived files, but increasingly downloads are in "higher res" formats. I have been creating impulses at 44.1khz and 24bit depth, and am happy with the results. I use rePhase to generate impulses and have the option to vary number of taps and other aspects of the generation process. Is there an advantage to going all Rob Watts on the tap length? Link to comment
LowOrbit Posted March 8, 2020 Share Posted March 8, 2020 2 hours ago, Miska said: Use 352.8 or 384 kHz sample rate and 64-bit floating point if possible. The actual filter defines number of taps needed. Since usually corrections should be mostly in bass range, filters tend to be somewhere from 16000 to 256000 taps long at 44.1 kHz rate. Every time you double the rate also number of taps gets doubled. Thanks Jussi - I shall draft some new Impulses and have a play. Link to comment
LowOrbit Posted May 29, 2020 Share Posted May 29, 2020 I upgraded to HQP Desktop 4.6.0 this morning and have been listening to some familiar music using the new Sinc L filter. Early impressions indicate a nice step forward in retrieval of low level detail and a nice musicality. I like it. It does seem to present a very calm, clear portrait of the music. I am running ASDM7EC at DSD256 from my i7-9700K/RTX2080ti pc via NUC/Audiolinux. Thanks Jussi. Solstice380 1 Link to comment
LowOrbit Posted September 2, 2020 Share Posted September 2, 2020 I've installed CUDA 11 and HQP Desktop 4.7 this morning on my i9700K based Windows machine with BIOS defaulted. 16gb RAM, using either SINC L or Sinc S @ ASDM7EC/256fs. I have a 2080ti GPU, Multicore DSP Greyed, CUDA Offload ticked, Adaptive Rate ticked. I use NAA via IPv6 backend, just to complete the picture. I get a brief drop out every 5 - 10 seconds, only tried with one redbook rip. Reverting to Desktop 4.6 and the dropouts disappear. Link to comment
LowOrbit Posted September 2, 2020 Share Posted September 2, 2020 5 hours ago, Miska said: Note that you don't need to install anything CUDA related. Only latest regular Nvidia display drivers. Thanks for the pointer. Uninstalled all the Cuda guff and just update the drivers. Make no differential, still getting regularly brief interruptions at 256Fs. Dropping to 128Fs, all is good. Link to comment
LowOrbit Posted December 6, 2020 Share Posted December 6, 2020 20 hours ago, Traktorist3d said: Hello everyone, I have a PC based on a Core i5 2500k, even when overclocked to 4.2GHz, and when playing an ASDM7EC 256, the processor is only loaded by 50% and gags occur, respectively. Is the processor too old for HQP to make full use of it? Windows 7 64 Hi You may find that the cpu load may show as 50% but the core(s) running the ASDM7EC modulator is not able to process fast enough to keep the output realtime which causes the stuttering you are hearing. 4.2ghz is possibly not fast enough for this process. It may help if you are running a lightweight filter, but you may simply need to try a less heavy duty modulator. Link to comment
LowOrbit Posted May 18, 2023 Share Posted May 18, 2023 @Miska Recently I spent some time after parting with my Chord Dave using a variation on the DSC1 (one of the iterations developed over on DIYAudio a few years ago). This I fed from HQPlayer, and the files were derived from offline upsampled material, converting it to DSD to feed the converter. I know neither you or the author of PGGB (ZB) condone this sort of behaviour and I understand the rationale for doing one or the other. But sounded really good. Really good. This got me to wondering: When HQP invokes its filter prior to DSD modulation, what intermediate sample rate does it upsample to? 8FS. 16FS, 32FS? This value, might let me use PGGB-256 to upsample to this intermediate sample rate at 64-bit float output... Then when we fed this to HQP, could Oversampling 1x and Nx be set to None? What I want is to feed the 64bit files to the modulator without going through an upsampling filter in HQP. Yes, I can use HQPlayer to feed the DSD decoder in my new T+A Dac200, or I could use 32bit PGGB files to feed the pcm path. I'm curious to see if I can achieve something that doesn't compromise the accuracy of the PGGB 64bit files, avoids extra upsampling stages, maybe allows me to use HQP convolution (I have use for a filter with my Raal headphones), and delivers the resultant output to the DSD decoder (which you of course have maintained is the purist route to reproduction, for as long as HQPlayer has been around.) Can you confirm if my thought process makes any sense and what the data rate is that gets sent to the modulator process in HQP? Link to comment
LowOrbit Posted May 18, 2023 Share Posted May 18, 2023 Miska, this is nothing but Obfuscation and deliberately missing the point - pretty much as expected. Disappointing. Bogi 22 minutes ago, bogi said: You asked at which sample rate HQPlayer performs convolution with PCM input. Answer is: at source PCM content rate. Then HQPlayer upsampling follows to reach modulator rate at which output DSD content is generated. No I didn't. I know about that. I asked what is the rate that HQPlayer upsamples pcm to prior to sending it to the modulator. And the answer is not the same as the DSD output rate. That makes no sense. kennyb123 1 Link to comment
LowOrbit Posted May 19, 2023 Share Posted May 19, 2023 People - I did not ask my question intending to pick a fight or get people rattled. I asked a question, Miska deflected (presumably because he doesn't want to engage in this, which is fine, but his answer was not straightforward and only informative by omission, also fine but I would have preferred an honest "not getting dragged into that" response). I move on. HQPlayer is good, other software also good. kennyb123 1 Link to comment
Recommended Posts
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 accountSign in
Already have an account? Sign in here.
Sign In Now