exaSound e18 - e20 - e28 - Info and Experiences Post All Here

[LewinskiH01]

I tried upconverting everything to DSD on JRiver and feeding 2xDSD to the e22, but it didn't work well. It stuttered and made noises similar to when latency is set up too low. The CPU was running at 5-6%, so it didn't seem like the PC was running out of power.

 

What might be causing this?

[Stereolab42]

I tried upconverting everything to DSD on JRiver and feeding 2xDSD to the e22, but it didn't work well. It stuttered and made noises similar to when latency is set up too low. The CPU was running at 5-6%, so it didn't seem like the PC was running out of power.

 

What might be causing this?

 

How many logical processors do you have? (You can see this information in Task Manager.) If the up-conversion task is single-threaded, and you have, say, 16 logical processors, 6% utilization could mean it's pegged. Also, run the "Benchmark" tool in JRiver under the "Help" menu, and report the result.

[dallasjustice]

My current plan is to run 4-way digital cross-over with room-correction, also for DSD, with upsampled output to e28...

 

Did you build a 4 way speaker? Are you using Acourate to build the filters?

[Miska]

Did you build a 4 way speaker?

 

Not yet, but my plan is to either build a 3-way loudspeakers + dual-sub, or strip out cross-over from an existing passive design, such these:

Divine 100.49 Walnut

99.36 MKII Piano

 

Most likely I would end up stripping down an existing one...

 

Are you using Acourate to build the filters?

 

Yes, Acourate (or Audiolense XO). For room correction part at least, maybe initially for cross-overs too.

[Sam Lord]

Not yet, but my plan is to either build a 3-way loudspeakers + dual-sub, or strip out cross-over from an existing passive design... ...Most likely I would end up stripping down an existing one... ...Acourate (or Audiolense XO). For room correction part at least, maybe initially for cross-overs too.

 

In my typically parasitic fashion I might do something similar, copying what you do. But I'm on my way to an upgrade: 4 channels of Hypex nCores arrived today . I hope to do as nice a job as barrows, he laid out his cabling and his modules better than anybody IMO, see his immaculate dual-mono amp in post #402 here:

 

nCore builds

 

I would love to use B&W802s and my current Velodyne DD12+ subs, after getting the Ferrari of course.

 

I would also love to see a two channel DAC with 4 buffered outputs though for multiamping, or a monster Exa E38 with 8 E22-class channels. Some would call it overkill. But then, I want my gear to sing like this guy skis. I want it to wrestle transients to be as effortlessly as Eon manages rocks. BTW, this guy is still alive, and now a doctor!

 

[dallasjustice]

I'll be interested to see how that goes. I've set up a 3way system with a pair of mono subs using Acourate.

 

Are you planning on measuring speakers/room with exa e28?

 

Not yet, but my plan is to either build a 3-way loudspeakers + dual-sub, or strip out cross-over from an existing passive design, such these:

Divine 100.49 Walnut

99.36 MKII Piano

 

Most likely I would end up stripping down an existing one...

 

 

 

Yes, Acourate (or Audiolense XO). For room correction part at least, maybe initially for cross-overs too.

[Audio_ELF]

Not yet, but my plan is to either build a 3-way loudspeakers + dual-sub, or strip out cross-over from an existing passive design, such these:

Divine 100.49 Walnut

Do I read correctly that the Divine speakers have direct access to the drivers (bypassing the crossover) so there would be no need to remove the passive crossover?

[Miska]

Do I read correctly that the Divine speakers have direct access to the drivers (bypassing the crossover) so there would be no need to remove the passive crossover?

 

No, XTZ used to have some models with external crossover box, but not anymore. But I would just take out the passive crossover and connect input connectors straight to the drivers. Easy modification for many loudspeakers that have enough connectors or at least space on the back panel.

[Miska]

Are you planning on measuring speakers/room with exa e28?

 

Yes, output would then naturally go through it, but I'm using Focusrite Forte for measurement mics.

[dallasjustice]

Yes, output would then naturally go through it, but I'm using Focusrite Forte for measurement mics.

How do you plan to synchronize the focusrite with the exa?

[Miska]

How do you plan to synchronize the focusrite with the exa?

 

It should be possible to import recorded sweep. At least for the open source DRC-FIR it is not a problem because there you just record playback of pre-generated WAV file and use the recorded result.

 

Analyzer always has be to be able to find the sweep from longer recording, because there's always some amount of latency from output to input, including the latency caused by microphone distance. So it can never assume playback and recording to be accurately time-aligned. There's easily some 50 ms delay at least.

 

(ESS doesn't even document the codec latency, but for example AKM does for their chips)

[dallasjustice]

I am not referring to latency.

 

I think Uli has a manual method to synchronize the input with the output. But it takes a little work. I also know that Audiolense uses a method similar to DIRAC to sync the in and out. I don't think any method is perfect though.

 

It would be much simpler if the exa had an analog input.

 

It should be possible to import recorded sweep. At least for the open source DRC-FIR it is not a problem because there you just record playback of pre-generated WAV file and use the recorded result.

 

Analyzer always has be to be able to find the sweep from longer recording, because there's always some amount of latency from output to input, including the latency caused by microphone distance. So it can never assume playback and recording to be accurately time-aligned. There's easily some 50 ms delay at least.

 

(ESS doesn't even document the codec latency, but for example AKM does for their chips)

[Miska]

I am not referring to latency.

 

I think Uli has a manual method to synchronize the input with the output. But it takes a little work. I also know that Audiolense uses a method similar to DIRAC to sync the in and out. I don't think any method is perfect though.

 

It would be much simpler if the exa had an analog input.

 

Even if it had, there's no way to accurately sync the two because you have unknown latency factors on the way. DAC and ADC chips have latency and in the end, there's latency from the speakers to the microphone since sound in air is pretty slow (around 350 meters per second).

 

So the software must be able to find the sweep from longer period of time. That's complete non-issue compared to many other DSP problems, because at least you know what to look for. (compared to signal analysis of passive sonars for example)

[dallasjustice]

If there were only one clock in and out there would be no clock drift. There are methods to compensate for clock differences between different in and out devices though.

 

Latency is generally not a big problem.

 

 

Even if it had, there's no way to accurately sync the two because you have unknown latency factors on the way. DAC and ADC chips have latency and in the end, there's latency from the speakers to the microphone since sound in air is pretty slow (around 350 meters per second).

 

So the software must be able to find the sweep from longer period of time. That's complete non-issue compared to many other DSP problems, because at least you know what to look for. (compared to signal analysis of passive sonars for example)

[Miska]

If there were only one clock in and out there would be no clock drift. There are methods to compensate for clock differences between different in and out devices though.

 

Clock drift is not an issue in this context at all. And thus no need to compensate for it.

 

Latency needs to be compensated for, so at the capture side recording time has to be longer than the sweep playback and then the sweep needs to be located from the longer recording period.

 

 

P.S. And if it would become a problem for some strange reason, I can perform the measurements using Focusrite RedNet 1.

[LewinskiH01]

How many logical processors do you have? (You can see this information in Task Manager.) If the up-conversion task is single-threaded, and you have, say, 16 logical processors, 6% utilization could mean it's pegged. Also, run the "Benchmark" tool in JRiver under the "Help" menu, and report the result.

 

Thanks for answering.

 

I have 8 logical processors. FWIW, Virtualization is disabled and Hyper-V Support is enabled.

I tried running the same song in 16/44, then another version of same song in 24/192, then upconverting the 16/44 file to 2xDSD on JRiver, and had the Resource Monitor open while I was doing this.

While playing the 16/44 song all 8 CPUs were well below 5%. When playing the 24/192 song several CPUs showed "higher" activity, but they were all well below 5% (too low to really discern the actual value).

Then, while upconverting to 2xDSD one processor was all the time between 20 and 30%, while other two CPUs were consistently between 5 and 10%. This added up to 5-6% overall CPU usage. Seems like no CPU was clogged, right?

 

I ran the benchmark tool:

Math score: 1606

Image: 3936

Database: 2861

JRMark (version 19.0.146): 2801

 

Does any of this provide any clue as to why audio stutters when upconverting to DSD at the server?

 

Thanks!

[Stereolab42]

Then, while upconverting to 2xDSD one processor was all the time between 20 and 30%, while other two CPUs were consistently between 5 and 10%. This added up to 5-6% overall CPU usage. Seems like no CPU was clogged, right?

 

I ran the benchmark tool:

Math score: 1606

Image: 3936

Database: 2861

JRMark (version 19.0.146): 2801

 

Does any of this provide any clue as to why audio stutters when upconverting to DSD at the server?

 

Thanks!

 

My JRMark is 3890 (math score of 2385) and I have no troubles with upconverting to DSD256, let alone DSD128. Do you have issues upconverting to DSD64? If not, then it surely must be your PC bottlenecking.

[LewinskiH01]

My JRMark is 3890 (math score of 2385) and I have no troubles with upconverting to DSD256, let alone DSD128. Do you have issues upconverting to DSD64? If not, then it surely must be your PC bottlenecking.

 

JRiver only gives me the option of converting to 2xDSD. Maybe I'm doing this in the wrong place? I do it at Options>Audio>DSP and Output, and select the convert to 2xDSD to be fed natively to a capable DAC. Other options are DSD thry DoP, but I purposefully avoided those as I understand they are inferior to DSD fed natively.

 

My computer is pretty powerful (see signature below) so I must have a setting screwing things up. What could it be? Actually I was surprised to get such a low score considering what Chris got with the highest performance CAPS and those specs vs. mine.

[Stereolab42]

Try upgrading to JRiver 20.

[EuroDriver]

 

My computer is pretty powerful (see signature below) so I must have a setting screwing things up. What could it be? Actually I was surprised to get such a low score considering what Chris got with the highest performance CAPS and those specs vs. mine.

 

what clock speed is the CPU running at ?

[Miska]

what clock speed is the CPU running at ?

 

It's 2.5 GHz constant and 3.5 GHz turbo peak with four cores (8 threads). Certainly shouldn't have any problems upsampling to DSD128...

 

ARK | Intel® Xeon® Processor E3-1265L v2 (8M Cache, 2.50 GHz)

[LewinskiH01]

It's 2.5 GHz constant and 3.5 GHz turbo peak with four cores (8 threads). Certainly shouldn't have any problems upsampling to DSD128...

 

Exactly right. It shows running at 2.2 GHz.

[LewinskiH01]

Try upgrading to JRiver 20.

 

I might, but not as a way to solve the issue at hand. The problem is not caused by 2xDSD but rather being put in evidence by it.

[EuroDriver]

Exactly right. It shows running at 2.2 GHz.

 

The problem might be AudioPhil's optimizer under clocking your CPU to 2.2 GHz and not allowing it to speed up to 3.5 Ghz when beneficial.

 

Have you tried something like CPU-Z to monitor the clock speed during playback ?

 

you may also want to try Foobar with SACD plugin that sounds significantly better than JRMC doing Redbook to DSD 256 (make sure integer mode is selected).

 

The sound quality champion is definitely HQ Player with its Polysinc filter. HQ Player is not compatible with some, if not most of AudioPhil's optimizer's OS tweaks

[bmoura]

HQ Player is not compatible with some, if not most of AudioPhil's optimizer's OS tweaks

 

Sounds like the best bet will be to go back to the exaSound ASIO drivers vs. the third party ones.

 

I tried the third party route once with the exaSound e28 and found the performance and sound was better staying with exaSound's drivers. It may be the story here as well.