Article: CAPS Twenty | Part One

[Le Concombre Masqué]

10 minutes ago, The Computer Audiophile said:

Perhaps I wasn't clear enough when I described this as a two PC system and this is the audio endpoint that's the NAA? 

what do the gorgeous gauges read then ? the CPU GPU etc load of the endpoint or of the horsepower PC ? 

[barrows]

13 minutes ago, The Computer Audiophile said:

Perhaps I wasn't clear enough when I described this as a two PC system and this is the audio endpoint that's the NAA? 

Or perhaps some did not actually read the introduction which I feel was very clear and spelled out your approach perfectly!

[The Computer Audiophile]

5 minutes ago, Le Concombre Masqué said:

what do the gorgeous gauges read then ? the CPU GPU etc load of the endpoint or of the horsepower PC ? 

The gauges in the endpoint display the resource utilization of the endpoint. The high power server just might have gauges that display its resources as well, but my lips are still sealed on this one. 

[Le Concombre Masqué]

7 minutes ago, The Computer Audiophile said:

The gauges in the endpoint display the resource utilization of the endpoint. The high power server just might have gauges that display its resources as well, but my lips are still sealed on this one. 

Pardon me then for I was far far away of awareness that a GPU was required for a NAA endpoint and furthermore that its load (and CPU's and RAM's) monitoring was critical...

[The Computer Audiophile]

Just now, Le Concombre Masqué said:

Pardon me then for I was far far away of awareness that a GPU was required for a NAA endpoint and furthermore that its load (and CPU's and RAM's) monitoring was critical...

In typical fashion you've found a way to be abrasive. Nobody said any of this is required. In fact I said the only requirement is the LattePanda and a power supply. 

[firedog]

10 minutes ago, Le Concombre Masqué said:

Pardon me then for I was far far away of awareness that a GPU was required for a NAA endpoint and furthermore that its load (and CPU's and RAM's) monitoring was critical...

Chris said that, " I was evaluating a couple DACs that could handle very high rates of native DSD content, but I couldn't get any of my existing audio endpoints to work perfectly with the DACs. I often heard loud noises, pops, and much of the time the DACs were just not recognized by the audio devices. I needed a solution to this issue, so I decided it was time for a new CAPS design." 

and

" if there was more Linux support for native DSD playback with HiFi DACs, I would've never gone down this path. I would use my Sonore Signature Rendu SE optical for one half of this CAPS design."

so no, I don't think anything like that was stated. 

[The Computer Audiophile]

I totally forgot to mention in the article that CAPS Twenty and CAPS Twenty.One were set to be used by EMM Labs at RMAF this year, but due to the show's cancelation the CAPS computers won't get a public debut this year. Perhaps they will make an appearance at a show in 2021 :~)

[agillis]

On 6/23/2020 at 2:37 AM, 20hertz said:

I hope Smallgreen computer can build some for us uneducated ones!

 

We have the sonicTransporer i9 now winch has an i9-9900 (8 core). The i9-10900 will be available and has 10 cores BUT the cores are slower! For upsampling most of the processing is done on one core per channel so you want really fast cores.

 

The sonicTransporter i9 is still the fastest upsamling machine available for Roon or HQplayer (or both together).

 

Chris as always has some amazing new tech here. That LattePanda Alpha 864s is super fast AND draws only 5w. Very good for a very small low power Roon box.

[The Computer Audiophile]

20 minutes ago, agillis said:

 

We have the sonicTransporer i9 now winch has an i9-9900 (8 core). The i9-10900 will be available and has 10 cores BUT the cores are slower! For upsampling most of the processing is done on one core per channel so you want really fast cores.

 

The sonicTransporter i9 is still the fastest upsamling machine available for Roon or HQplayer (or both together).

 

Chris as always has some amazing new tech here. That LattePanda Alpha 864s is super fast AND draws only 5w. Very good for a very small low power Roon box.

The i9-10900K looks to be faster. 
 

https://cpu.userbenchmark.com/Compare/Intel-Core-i9-9900K-vs-Intel-Core-i9-10900K/4028vs4071

 

 

 

 

 

 

 

https://www.cpu-monkey.com/en/compare_cpu-intel_core_i9_10900k-1139-vs-intel_core_i9_9900k-890

 

[barrows]

Jussi has mentioned quite a few times that the most important spec for running serious oversampling algos (the EC modulators) in HQP is the base clock speed.  The processor with the higher base clock speed is the one to use.  It appears to be well accepted that the I9-9900K processor runs the EC modulators perfectly well, but the SGC I9 machine does not use the "K" processor, so I believe the base clock speed is lower.

I do not think the I9 machine from @agillis can run HQPlayer 4, with the EC 7th order modulator at DSD 256 (poly sinc 2 EXT filter), but I would love to proven wrong on this.

 

I would prefer to run a Sonic Transporter here, if it could do HQP 4, with the EC modulators to DSD 256...

 

Folks have already confirmed that the I9-9900K does fine, and I am sure the slightly faster I9-10900K will do it fine as well.

 

 

[Account Closed]

53 minutes ago, barrows said:

Jussi has mentioned quite a few times that the most important spec for running serious oversampling algos (the EC modulators) in HQP is the base clock speed.  The processor with the higher base clock speed is the one to use.  It appears to be well accepted that the I9-9900K processor runs the EC modulators perfectly well, but the SGC I9 machine does not use the "K" processor, so I believe the base clock speed is lower.

I do not think the I9 machine from @agillis can run HQPlayer 4, with the EC 7th order modulator at DSD 256 (poly sinc 2 EXT filter), but I would love to proven wrong on this.

 

I would prefer to run a Sonic Transporter here, if it could do HQP 4, with the EC modulators to DSD 256...

 

Folks have already confirmed that the I9-9900K does fine, and I am sure the slightly faster I9-10900K will do it fine as well.

 

 

If memory serves the i9-9900KS is even a better choice as it will turbo two cores even higher than the 9900K. However, if there is going to be a serious GPU for off loading the filters then an i5 9400 or 10400 will work just fine. I run an i5 8400 without a GPU and I can easily do DSD 256 ASDM7EC with poly-sinc ext2. It is the high demand filters that drive the need for the i9-9900x chips not the modulators. For the modulators it is all about getting two cores to run at 4.0 Ghz or slightly higher constantly without thermal throttling. This is why Intel sent Chris an i5 as well as the i9 to test. The problem with the i9 when running without a GPU is that it becomes very difficult to passively cool. The TDP under load is way over 95 watts and most passive cool cases that I have seen top out at 95 watts. This leaves active liquid or active air cooling each with greater complexity and/or noise. The i5 on the other hand is a 65 watt chip so under load it might reach 95 watts and still be able to be passively cooled.  The real real problem comes when trying to cool a GPU passively. The ones that can be heat piped are under the needed capacity and the capable ones require some form of active cooling which is usually air cooling and that means a potential noise issue. Sadly there is no magic bullet right now that is silent, can run every filter and that can do DSD 256 via EC modulators. And, as I have said elsewhere here there is nothing on the horizon that will allow for doing DSD512 via EC modulators. That would require a chip that could run two cores at 8 Ghz continuously without melting. Maybe the i19-999999 S to be released in Q2 of 2030. It will likely come with a liquid nitrogen cooler and require a dedicated 30 amp circuit.

 

[Account Closed]

1 hour ago, agillis said:

 

We have the sonicTransporer i9 now winch has an i9-9900 (8 core). The i9-10900 will be available and has 10 cores BUT the cores are slower! For upsampling most of the processing is done on one core per channel so you want really fast cores.

 

The sonicTransporter i9 is still the fastest upsamling machine available for Roon or HQplayer (or both together).

 

Chris as always has some amazing new tech here. That LattePanda Alpha 864s is super fast AND draws only 5w. Very good for a very small low power Roon box.

Andrew, this first piece is strictly an endpoint like the Rendu series but for home builders and experimenters. Now the one to come may give the ST i9 a run.

[barrows]

49 minutes ago, bobflood said:

If memory serves the i9-9900KS is even a better choice as it will turbo two cores even higher than the 9900K. However, if there is going to be a serious GPU for off loading the filters then an i5 9400 or 10400 will work just fine. I run an i5 8400 without a GPU and I can easily do DSD 256 ASDM7EC with poly-sinc ext2. It is the high demand filters that drive the need for the i9-9900x chips not the modulators. For the modulators it is all about getting two cores to run at 4.0 Ghz or slightly higher constantly without thermal throttling. This is why Intel sent Chris an i5 as well as the i9 to test. The problem with the i9 when running without a GPU is that it becomes very difficult to passively cool. The TDP under load is way over 95 watts and most passive cool cases that I have seen top out at 95 watts. This leaves active liquid or active air cooling each with greater complexity and/or noise. The i5 on the other hand is a 65 watt chip so under load it might reach 95 watts and still be able to be passively cooled.  The real real problem comes when trying to cool a GPU passively. The ones that can be heat piped are under the needed capacity and the capable ones require some form of active cooling which is usually air cooling and that means a potential noise issue. Sadly there is no magic bullet right now that is silent, can run every filter and that can do DSD 256 via EC modulators. And, as I have said elsewhere here there is nothing on the horizon that will allow for doing DSD512 via EC modulators. That would require a chip that could run two cores at 8 Ghz continuously without melting. Maybe the i19-999999 S to be released in Q2 of 2030. It will likely come with a liquid nitrogen cooler and require a dedicated 30 amp circuit.

 

I have tried to run EC modulators with higher level processors than the I5 you says works fine, were you running overclocking, and specially configured bios?  Were you also running Roon on the same machine?  I have no interest in having to run Roon on a separate machine, for example.  Indeed, I did not attempt to run the more difficult filters (1x) and am quite happy with the performance of the poly-sink ext2 filter.

I must say I still do not understand anyone's resistance to having active cooling?  I have no problem with having a well configured fan based cooling system to keep the machine at a reasonable temperature, after all, there is no way i am going to have a computer doing this much processing, and all the electrical noise and airborne RF noise it creates, anywhere near the audio system: I keep powerful computers off in the work room, where they have no effect on the audio system in the listening area.  To me it seems much simpler to use an active, air cooled system, and this approach should also be able to contribute to proper performance and reliability.

[Account Closed]

54 minutes ago, barrows said:

I have tried to run EC modulators with higher level processors than the I5 you says works fine, were you running overclocking, and specially configured bios?  Were you also running Roon on the same machine?  I have no interest in having to run Roon on a separate machine, for example.  Indeed, I did not attempt to run the more difficult filters (1x) and am quite happy with the performance of the poly-sink ext2 filter.

I must say I still do not understand anyone's resistance to having active cooling?  I have no problem with having a well configured fan based cooling system to keep the machine at a reasonable temperature, after all, there is no way i am going to have a computer doing this much processing, and all the electrical noise and airborne RF noise it creates, anywhere near the audio system: I keep powerful computers off in the work room, where they have no effect on the audio system in the listening area.  To me it seems much simpler to use an active, air cooled system, and this approach should also be able to contribute to proper performance and reliability.

Hi Barrows,

 

Yes, I do run Roon on the same machine and yes it did take some BIOS tweaking. It also takes a capable MB (ASUS Z370 I in my case). I am using Win 10 as the OS and it took some tweaks there as well. I am able to keep the CPU at or below 70 C. I also use this as my  desktop (but only when music is not playing). I wanted a passively cooled machine so I used a Streamcom 65 Watt TDP case. My office is too close to my main system in the living room so I needed a quiet machine. If I were to invest in a separate server machine at this point in time I would go with active liquid cooling, an i9 variant and a kick ass GPU (2080ti probably). But, I won't as I am happy with DSD 256 EC and the 2s filters I can run. Keep in mind that many people cannot get a noisy server machine far enough away from their music system so hence the need for passively cooled silent server hardware.

[57gold]

OK, amongst serious CA folks with tech savvy, so excuse the continued silliness one my part, but, if the 20.1 is just handling already upsampled data from a heavy duty computer (which is doing the serious lifting), why the big heavy, mostly empty box, with heat sinking and oversized or super duper power supplies, not in the big box, to run relatively light weight stuff?

 

Just an exercise of the possible...belts, suspenders and gaffers tape, because one can?

[The Computer Audiophile]

5 minutes ago, 57gold said:

OK, amongst serious CA folks with tech savvy, so excuse the continued silliness one my part, but, if the 20.1 is just handling already upsampled data from a heavy duty computer (which is doing the serious lifting), why the big heavy, mostly empty box, with heat sinking and oversized or super duper power supplies, not in the big box, to run relatively light weight stuff?

 

Just an exercise of the possible...belts, suspenders and gaffers tape, because one can?

 

 

There are endless ways to explain what I did and why I did it, but it's likely already in the article. Perhaps CAPS just isn't your thing, just like D to D converters aren't your thing. 

[jrobbins50]

Chris, rightly or wrongly, I use VNC Viewer / Server to remote into my CAPSZuma server and other home PCs being used as Roon multichannel endpoints. Any reason that VNC would not work on the 20.1?

 

I will also ask if the 20.1 is intended to support multichannel playback.  Thanks. JCR 

[The Computer Audiophile]

Just now, jrobbins50 said:

Chris, rightly or wrongly, I use VNC Viewer / Server to remote into my CAPSZuma server and other home PCs being used as Roon multichannel endpoints. Any reason that VNC would not work on the 20.1?

 

I will also ask if the 20.1 is intended to support multichannel playback.  Thanks. JCR 

VNC will work just fine. 

 

Multi-channel will work just fine as well. 

[firedog]

2 hours ago, 57gold said:

OK, amongst serious CA folks with tech savvy, so excuse the continued silliness one my part, but, if the 20.1 is just handling already upsampled data from a heavy duty computer (which is doing the serious lifting), why the big heavy, mostly empty box, with heat sinking and oversized or super duper power supplies, not in the big box, to run relatively light weight stuff?

 

Just an exercise of the possible...belts, suspenders and gaffers tape, because one can?

As Chris said in the article, he doesn't claim this build improves SQ over other solutions, etc. He built it to solve a problem between Linux based streamers and certain specific DACs. He also said, without those DACs, no reason not to stick with his Rendu solution.

Besides that, it seems pretty clear he was having fun building something extreme, both in terms of the HW and the gauge displays.  Clearly not for every one, and clearly not for those who aren't going to DIY or find a very capable and probably expensive custom builder.

 

I'd assume the CAPS Twenty server build is going to have a much broader appeal to the audience here, which includes lots of people doing heavy uspampling and DSP.

[57gold]

9 hours ago, The Computer Audiophile said:

 

 

There are endless ways to explain what I did and why I did it, but it's likely already in the article. Perhaps CAPS just isn't your thing, just like D to D converters aren't your thing. 

 

Will the CAPS connect to a DAC with a D to D, or will all the efforts to handle the signal received from the heavy duty computer make one of those unnecessary?

[The Computer Audiophile]

6 minutes ago, 57gold said:

 

Will the CAPS connect to a DAC with a D to D, or will all the efforts to handle the signal received from the heavy duty computer make one of those unnecessary?

I believe you're joking but I'll answer for those who may have a serious interest in the question. CAPS can connect to anything and the use of a D to D is up to the listener or DAC manufacturer. 

[barrows]

12 minutes ago, 57gold said:

 

Will the CAPS connect to a DAC with a D to D, or will all the efforts to handle the signal received from the heavy duty computer make one of those unnecessary?

It would probably be a compromise to directly connect a very powerful computer (as this will be) directly to a DAC, either via direct USB, or through an intermediary D-D converter.  A computer with this level of processing power is going to be very noisy electrically and in terms of RF (airborne) noise.  Combining this powerful server/oversampling engine with a remote endpoint, like the Twenty.one, or another Ethernet Renderer of one's choice, and getting it well away from the audio system is a much better idea.

[Account Closed]

38 minutes ago, barrows said:

It would probably be a compromise to directly connect a very powerful computer (as this will be) directly to a DAC, either via direct USB, or through an intermediary D-D converter.  A computer with this level of processing power is going to be very noisy electrically and in terms of RF (airborne) noise.  Combining this powerful server/oversampling engine with a remote endpoint, like the Twenty.one, or another Ethernet Renderer of one's choice, and getting it well away from the audio system is a much better idea.

Direct connection might work fine IF one can isolate the USB output with something like the JCAT USB card as the output device as it can be independently powered or one of the many USB isolation devices available. Still, I personally would not want a high powered server machine in close proximity to my audio system but that said many here are doing just that and reporting good results.

[barrows]

11 minutes ago, bobflood said:

Direct connection might work fine IF one can isolate the USB output with something like the JCAT USB card as the output device as it can be independently powered or one of the many USB isolation devices available. Still, I personally would not want a high powered server machine in close proximity to my audio system but that said many here are doing just that and reporting good results.

My testing and measurements led me to the conclusion that servers like this with any proximity to the audio system are very problematic.  The airborne RF emitted by a server which includes storage drive(s) and and an OTS motherboard is not to be underestimated.  Additionally, the amount of noise coming out of the power supply side, and being put back on the AC power system is also very significant.  Even with very powerful line filters, (even AC regenerators) this noise went right onto the AC line, and was seen on the AC power cabling going into the audio system components.  Back when i was trying to optimize direct connected (USB) servers I ended up with a fully battery powered server to stop the AC line noise problem, but the airborne RF still existed.  Linear power supplies made little or no difference, the noise was being produced by the MoBo and the hard drives (SSD or spinning discs did not matter either, in fact it usually appears that SSDs were worse, which was odd).

The only real solution which ended up working, was getting the serving computer well away, electrically and physically, from the audio system.  Best to have it in another part of the home, powered from the opposite phase of the AC home power, and isolated from the audio system by optical fiber Ethernet.

The electrical connection via USB is not the only problem.

 

[Account Closed]

54 minutes ago, barrows said:

My testing and measurements led me to the conclusion that servers like this with any proximity to the audio system are very problematic.  The airborne RF emitted by a server which includes storage drive(s) and and an OTS motherboard is not to be underestimated.  Additionally, the amount of noise coming out of the power supply side, and being put back on the AC power system is also very significant.  Even with very powerful line filters, (even AC regenerators) this noise went right onto the AC line, and was seen on the AC power cabling going into the audio system components.  Back when i was trying to optimize direct connected (USB) servers I ended up with a fully battery powered server to stop the AC line noise problem, but the airborne RF still existed.  Linear power supplies made little or no difference, the noise was being produced by the MoBo and the hard drives (SSD or spinning discs did not matter either, in fact it usually appears that SSDs were worse, which was odd).

The only real solution which ended up working, was getting the serving computer well away, electrically and physically, from the audio system.  Best to have it in another part of the home, powered from the opposite phase of the AC home power, and isolated from the audio system by optical fiber Ethernet.

The electrical connection via USB is not the only problem.

 

Agreed!