ReClock 1.8.6.0

[VPhile jr.]

Btw. I just got insane sound experience with those Bach recordings linked page before on yet another PC/old Thinkpad & Senn PX100: 1411 KBps output, madFLAC, MPC, 16bits PCM DSound (bit exact), 44.1kHz and rabbit plugin @ "Best Sinc" - is it applied in PCM output at all (not seeing any cpu hit ~1%) in this setup or is it just nightly placebo? Perhaps just bare ReClock renderer in bit exact mode showing its supreme tricks? Or is it? Because higher mp3 radio streams >200Kbps are not playing fluently with "Best Sinc" at bit exact. Also, just on subjective audio level comparison to stock VLC, WinAMP or MPC the difference is noticeble.

 

[GuyG]

Just provide you with some run of the gun performance numbers for Atom based (dual core) system while reclocking mostly some jazzy tunes, for a comparison/future development purposes..

--
Atom ION 330 (two physical cores, two virtual), onboard audio Realtek ALC662/Dsound render
WinXPSP3, MPC-HC (using its native FLAC and mp3 decoder in IEEE float), ReClock 1.8.6.0. + resampler 1.0.0.2 (MT edition)
---
mp3 stream:
128Kbps - 192KBps radio
Reclock @ 96kHz (Good Sinc Interpolation) ~60% CPU
Reclock @ 192kHz (Good Sinc Interpolation) ~95% CPU
both silky smooth but anything above setting "Good Inter." SQ stutters as hell..
---
mp3 local:
192Kbps - 256KBps
Reclock @ 96kHz (Good Sinc Interpolation) ~45% CPU
Reclock @ 192kHz (Good Sinc Interpolation) ~85% CPU
both silky smooth but anything above setting "Good Inter." SQ stutters as hell..
---
FLAC local:
1411 KBps output was "playable" at Good Sinc settings as the max, CPU was <90% @ 192kHz and <50% 96kHz, BUT with occasional brief stutter or crack, Audio clock shows 104ms samp. Sorry don't have extensive FLAC library at hand (perhaps 1411 KBps is too much) with this PC now, new Czech Radio recordings of complete six Bach's Brandenburg Concertos, downloaded free FLACs here (scroll down): http://www.rozhlas.cz/d-dur/download_eng
---

=> I´m kind of surprised that those two top SQ settings are in practice unavailable for this system (for mp3), but this is really no big number cruncher platform to star with in terms of its CPU ~1.6GHz.

However, Atoms feature powefull video chip, Atoms are quite popular home cinema platforms, perhaps future resampler version using CUDA could give us that "Best Sinc" at least in 96kHz. Shall we contact the CUDA guys on their forum, they are quite responsive, to look at it with James as to whether is it feasible project at all? Thanks

This reminds me. I've been doing more testing with reclock and using the latest resampler, I cannot get completely smooth playback with the best sinc option. (192/24 padded) I have an E5200 running at 3.5ghz but regardless of cpu load playback is not smooth.

Reclock reported about 25% cpu, task manager reported about 80% peak cpu usage (high update speed) and madVR reported no dropped/delayed frames but it was still obviously stuttering onscreen.

If I used 'as source' (48/16) with best sinc cpu use dropped much lower but playback was still not smooth. I reduced the cpu/gpu load of everything else (like madVR/ffdshow) by setting to the lowest quality settings and still playback was not smooth with best sinc.

As soon as I took it off best sinc I was able to go back to 192/24 padded and turn everything else up and have smooth playback even though cpu load was now the same/higher.



Also, when watching a film last night after about 60-90 mins the framerate seemed like it had dropped in half. Not sure exactly what the framerate was but madVR was not reporting dropped/delayed frames and quitting/reloading fixed it.

I don't know if this was a reclock or other issue though and don't have the time to test it atm.



Was also wondering if someone could explain the sound pre-buffer and latency options as I was seeing a lot of lip-sync problems until I started changing it, but I'm not sure what it should be set to or how to work out the best settings.



Also, with DACs does anyone do a DAC with HDMI yet? If something had HDMI and USB I could use it for all my sources.

 

[yesgrey3]

Btw. I just got insane sound experience with those Bach recordings linked page before on yet another PC/old Thinkpad & Senn PX100: 1411 KBps output, madFLAC, MPC, 16bits PCM DSound (bit exact), 44.1kHz and rabbit plugin @ "Best Sinc"

The resampler only is used with media files. With pure audio files the resampler is bypassed, hence the "bit exact" words.

By the way, if you just had such an insane sound experience using PCM DSound (and Microsoft internal resampler), you should try with PCM Kernel Streaming (if XP) or WASAPI (if Vista/W7).

 

[leeperry]

The resampler only is used with media files. With pure audio files the resampler is bypassed, hence the "bit exact" words.

By the way, if you just had such an insane sound experience using PCM DSound (and Microsoft internal resampler), you should try with PCM Kernel Streaming (if XP) or WASAPI (if Vista/W7).

DS can be bit-perfect if all the sliders are set to the max apparently...it's been thoroughly discussed on some forums.

just to be clear, the resampler will only be bypassed on audio files if you chose "same as input" for the output sample rate, or manually set the same one as the input file.

 

[James]

Btw. I just got insane sound experience with those Bach recordings linked page before on yet another PC/old Thinkpad & Senn PX100: 1411 KBps output, madFLAC, MPC, 16bits PCM DSound (bit exact), 44.1kHz and rabbit plugin @ "Best Sinc" - is it applied in PCM output at all (not seeing any cpu hit ~1%) in this setup or is it just nightly placebo? Perhaps just bare ReClock renderer in bit exact mode showing its supreme tricks? Or is it? Because higher mp3 radio streams >200Kbps are not playing fluently with "Best Sinc" at bit exact. Also, just on subjective audio level comparison to stock VLC, WinAMP or MPC the difference is noticeble.

bit exact = no resampling.

 

[namaiki]

edit: one webshop as reference price point, here kitty:
http://www.decacom.co.kr/
http://myworld.ebay.com.sg/gyrocom/
(they've listed 96kHz DAC called AUDIOTRAK Prodigy CUBE, similar to UHF-1 (but diff. components), for ~$130)

I advised this one to a friend, and it made him sell his STX: http://www.audiophonics.fr/info-ng27-amplificateur-casque-usb-ipod-dac-pcm2702-p-5006.html

Would I be better off going for the above AUDIOTRAK, at least, over the NG27?

 

[leeperry]

Would I be better off going for the above AUDIOTRAK, at least, over the NG27?

maybe you could create a seperate thread so we don't pollute the main Reclock thread w/ hardware talk

it all depends on your budget, but the NG27 w/ a LT1364 sure would make it worth your while...especially over a Realtek 8)

 

[Mark_A_W]

This reminds me. I've been doing more testing with reclock and using the latest resampler, I cannot get completely smooth playback with the best sinc option. (192/24 padded) I have an E5200 running at 3.5ghz but regardless of cpu load playback is not smooth.

Reclock reported about 25% cpu, task manager reported about 80% peak cpu usage (high update speed) and madVR reported no dropped/delayed frames but it was still obviously stuttering onscreen.

If I used 'as source' (48/16) with best sinc cpu use dropped much lower but playback was still not smooth. I reduced the cpu/gpu load of everything else (like madVR/ffdshow) by setting to the lowest quality settings and still playback was not smooth with best sinc.

As soon as I took it off best sinc I was able to go back to 192/24 padded and turn everything else up and have smooth playback even though cpu load was now the same/higher.



Also, when watching a film last night after about 60-90 mins the framerate seemed like it had dropped in half. Not sure exactly what the framerate was but madVR was not reporting dropped/delayed frames and quitting/reloading fixed it.

I don't know if this was a reclock or other issue though and don't have the time to test it atm.



Was also wondering if someone could explain the sound pre-buffer and latency options as I was seeing a lot of lip-sync problems until I started changing it, but I'm not sure what it should be set to or how to work out the best settings.



Also, with DACs does anyone do a DAC with HDMI yet? If something had HDMI and USB I could use it for all my sources.

I have the same issue.

With Best Sinc at 192khz it stutters, despite reclock reporting ~56% CPU, and task manager showing available CPU on my Quad Q6600 at 3.0ghz. But there may be CPU spikes that really are maxing it out, that don't get shown due to the averaging.

Dropping the resampling back to second best (44% CPU in reclock) works.

I also use madVR.


Mind you I could only resample to 92khz/excellent or 192khz/high quality with earlier versions.

So there may be an stuttering issue, or we may just be out of grunt, hard to say.

Mark

 

[James]

I have the same issue.

With Best Sinc at 192khz it stutters, despite reclock reporting ~56% CPU, and task manager showing available CPU on my Quad Q6600 at 3.0ghz. But there may be CPU spikes that really are maxing it out, that don't get shown due to the averaging.

Dropping the resampling back to second best (44% CPU in reclock) works.

I also use madVR.


Mind you I could only resample to 92khz/excellent or 192khz/high quality with earlier versions.

So there may be an stuttering issue, or we may just be out of grunt, hard to say.

Mark

Simple: Don't use it. The "good" setting is really, really good.

 

[Mark_A_W]

That's what I have done


But it's possible there is an issue, rather than a CPU limit, that's why I posted.

 

[James]

That's what I have done


But it's possible there is an issue, rather than a CPU limit, that's why I posted.

The problem is, that not only the CPU load is important, but also the absolute time spent. I wouldn't go higher than 20-30% CPU load shown by ReClock. Higher values might cause samples not to be rendered in time.

 

[el Filou]

Now that's interesting! Would there be any way to check if it skips some samples in logs (or in ReClock CP but I doubt that) ?

 

[James]

Now that's interesting! Would there be any way to check if it skips some samples in logs (or in ReClock CP but I doubt that) ?

It doesn't "skip" them. You'll hear drop outs if they aren't delivered in time.

 

[VPhile jr.]

DS can be bit-perfect if all the sliders are set to the max apparently...it's been thoroughly discussed on some forums.

just to be clear, the resampler will only be bypassed on audio files if you chose "same as input" for the output sample rate, or manually set the same one as the input file.

Yep, that's what I meant thanks for steping in, color me crazy, but subjectively I prefer DS bit perfect w. Reclock render over WinAMP-KernelS, which isn't bad just different, in headphone application. It could be a different matter with megaspeakers, about that test later on..

-

HDMI - DAC availability, yes the previously discussed&linked Korean JAVS, which apparently manufs/develops for ESI/Audiotrak brands has got models with HDMI connectors as well. Specifically DAC-2 is HDMI input capable (330,800/$290)and UDT-1 supports digital audio out via HDMI.

 

[VPhile jr.]

The ultimate ReClock test bed would be real time upsampling of FLAC streams, there are radios @ ~700Kbs, which is the full original CD data bandwidth in this lossless format. I guess the current gen of computers will choke on that one in 96kHz "Best Sinc", even with MT plugin enabled and multi GHz cores at work, but should be piece of cake on real CUDA hardware as per big multi/GPUs..

LeePerry and others you can give it a spin here
(real freeair/sat classic music station and their experimental stream):
http://radio.cesnet.cz:8000/cro-d-dur.flac

If I understand it correctly they stream in 48kHz because of the overhead needed during transcoding/streaming:
http://www.cesnet.cz/doc/techzpravy/2008/using-flac-encoding/

Unfortunately FLAC streams work only in Foo[L]bar and VLC, not in MPC, or at least I didn't find a way how to run it..

 

[el Filou]

So you mean the new 1.8.6 resampler is actually more CPU-intensive than SRC, even with the multithreaded optimization?
Because my E7600 can resample 2 channel FLAC (even 96/24) to 24/192 Excellent just fine with 1.8.5.8, and it's only using one core...

successfully playing live streams with ReClock is just a question of sufficient buffering in the player I guess.

 

[Mark_A_W]

The ultimate ReClock test bed would be real time upsampling of FLAC streams, there are radios @ ~700Kbs, which is the full original CD data bandwidth in this lossless format. I guess the current gen of computers will choke on that one in 96kHz "Best Sinc", even with MT plugin enabled and multi GHz cores at work, but should be piece of cake on real CUDA hardware as per big multi/GPUs..

LeePerry and others you can give it a spin here
(real freeair/sat classic music station and their experimental stream):
http://radio.cesnet.cz:8000/cro-d-dur.flac

If I understand it correctly they stream in 48kHz because of the overhead needed during transcoding/streaming:
http://www.cesnet.cz/doc/techzpravy/2008/using-flac-encoding/

Unfortunately FLAC streams work only in Foo[L]bar and VLC, not in MPC, or at least I didn't find a way how to run it..

Well, I'm using reclock with movie soundtracks converted to FLAC. So that's up to 8 channels at 24 bit 48khz (Hellboy2, Prince Caspian), or 6 channels 24bit 96khz (Baraka, Chronos).

It's with this audio that 8 channels resampling to 192khz at Best Sinc is not quite possible.

2 channel 16bit 44.1khz audio is a doddle.

Zoom Player will play .FLAC audio files, no probs. You need madFLAC decoder installed.

 

[James]

So you mean the new 1.8.6 resampler is actually more CPU-intensive than SRC, even with the multithreaded optimization?
Because my E7600 can resample 2 channel FLAC (even 96/24) to 24/192 Excellent just fine with 1.8.5.8, and it's only using one core...

successfully playing live streams with ReClock is just a question of sufficient buffering in the player I guess.

SRC is much more efficient with multi channel using only one core:
http://www.mega-nerd.com/erikd/Blog/CodeHacking/SecretRabbitCode/rel_0_1_5.html

If you compare throughput of one channel and six channels, the sample throughput is only a little less than 1/2, not 1/6.

With the 1.8.6.1 resampler the drop in throughput is (almost) linear, 6 channels will take 6 times longer than one. However, processing 6 channels will be done in 6 independent threads, so a 6 core CPU would be fully utilized.

On a dual core CPU throughput of 2 channels will be the same as 1, 6 channels will have 1/3 throughput compared to 1 or 2 channels.

 

[VPhile jr.]

If I recall it correctly, James, actually increased the quality settings in that plugin, so crazy people can test new toys, therefore I doubt it can't be directly compared with older versions of ReClock..

 

[leeperry]

SRC is much more efficient with multi channel using only one core:
http://www.mega-nerd.com/erikd/Blog/CodeHacking/SecretRabbitCode/rel_0_1_5.html

meaning SRC might not be completely dropped due to ppl still running a single core CPU?