Article: Computer Audiophile Pocket Server C.A.P.S. v4 Pipeline

[Lowlands]

That s5533 isn't the board I was referring to, the one I bouhgt is the s5535, there is two versions, the one ending in HE is the server board that takes ECC RAM.

 

Right. My bad.

 

Is it this one: TYAN - Motherboards_S5535-HE_S5535AG2NR-HE

Can you confirm that the picture is the HE version? If so, that board looks likeTHE solution for Streacom FC5 builds, since there is no obstruction for heatpipes....

 

[robbbby]

That's the one. It was ~$35 cheaper than the supermicro board, fits in a much cheaper hdplex case that looks just as good IMO and does not require the $40ish? heat pipe riser to clear the ram. It's the poor mans pipeline!

The only thing I am not sure of is if it supports IPMI, the manual does not say anything about it since it shares a manual with the non HE. A few articles and TYAN ads say it does support IPMI 2.0 but I cannot yet verify that as I am still waiting for my power supply.

[Lowlands]

That's the one. It was ~$35 cheaper than the supermicro board, fits in a much cheaper hdplex case that looks just as good IMO and does not require the $40ish? heat pipe riser to clear the ram. It's the poor mans pipeline!

The only thing I am not sure of is if it supports IPMI, the manual does not say anything about it since it shares a manual with the non HE. A few articles and TYAN ads say it does support IPMI 2.0 but I cannot yet verify that as I am still waiting for my power supply.

 

Good stuff..!

Now.... all we need is a retail point of sales in Europe for Tyan.... Anyone ??

[robbbby]

By the looks of things they are tough to find. That specific board is no longer in stock at the place I purchased it from. Searching on google I found only one other place in north america that had them in stock but they were selling for more than the supermicro board.

[capaill]

has any research been done to see if this or a similar system could be modified or redesigned to incorporate multi room audio, with the same song being capable of being synced across multiple rooms. Possibly have a master unit, such as this one for main listing area, and a lower spec unit for around the house. Everyone agrees sonos does this really well, but they don't, as most others also don't, do Hirez or allow us to use our playback software and app of choice, such as Jriver/Jremote.

[unsleepable]

Chris —

 

I've been checking this C.A.P.S. v4 Pipeline server as I am looking for components to build my own digital source. And I think there are a couple of issues about which I hope I can get your input.

 

On the one hand, why a 80W processor? The E3-1241 v3 processor is very powerful, but similar processing power can be found a lower power consumption and heat dissipation. For the Streacom FC10 case, the vendor specifies a maximum of 95W TDP CPU. But if you open the user guide at http://www.streacom.com/downloads/ug/fc10.pdf, on page 3 it more clearly recommends a 65W TCP one. The vendor warns that a high TDP will require the chassis to be "placed in a location with adequate air flow and moderate ambient room temperature."

 

I understand that nothing major will happen if the chassis is not able to keep the processor cool. The CPU will just throttle down to lower the temperature. But then, it begs the question of why this processor was chosen when, depending on the environment, it might not be possible to obtain its maximum performance.

 

The other issue is power. The HDPLEX Linear PSU provides a maximum of 100W regardless of the power rating of the DC power supply. But a quick calculation shows that the maximum power consumption for the whole setup is as follows:

 

- Processor: 80W

- Motherboard: since it doesn't provide full-fledged video, let's say a conservative 20W.

- Memory: again conservatively, ~ 3W per module

- SSD: 3.3W

- SOtM tX-PCIexp card: ~ 15W plus 2.5W per connected device—a maximum of 20W.

 

Everything adds up to 129.3W—without considering additional disks, a potential graphics card, or the optical drive that the chassis allows for. Even assuming a 100% efficiency on the linear power supply, we are ~ 30% over maximum—and again, the culprit is clearly the processor. I understand that this is under absolute maximum load, and I assume that normally the system will function fine—but still, it seems that we are not exactly on the safe side.

 

Am I maybe missing something here? Also, if you have your own calculations about these aspects of the system design, I would very much appreciate it if you could share them to have a better insight.

 

And one last question, what's the advantage of the FC10 over the FC5 chassis? Since the motherboard is µATX, it would fit in the FC5 chassis and still leave room for 3 x 2.5" disks, while removing the need for the pipe extensions, and I guess simplifying the setup. What other considerations have gone into the chassis selection?

 

Thank you in advance for any help you can shed on this.

[aps]

Chris - Thanks. Your photographs show the cables from the power supply threaded through the IEC socket cut-out on the FC10. Is this the final solution or is there a neater solution that allows one to simply connect / disconnect the FC10 from the power supply?

[Knuck]

Great questions unsleepable. Chris could you also tell us why you recommend the X10SLH-F over some of the cheaper X10 models like the X10SLM-F.

[Elberoth]

Here is a small connection tutorial (with pics) I posted in another thread:

 

CAPS v4 Pipeline build tutorial

[Elberoth]

(...)

On the one hand, why a 80W processor? The E3-1241 v3 processor is very powerful, but similar processing power can be found a lower power consumption and heat dissipation.

 

My thinking exactly.

 

I have went with the Xeon 1230Lv3 variant, which is only 25W max TDP.

 

The CPU runs unbeliveably cool - it reads only 5'C over ambient at isle and 8'C doing PCM to DSD upsampling. EIGHT degrees !

 

This is with open case and 24h run in repeat.

 

When I came the first time in the morning, and I touched the heatpipes, I though that the computer has switched off for some reason. It was so cool - both the heatpipes and the riser's aluminium plate directly over the CPU - that that it felt like touching any metal object in the room. I was not able to detect ANY heat. Only after I left the server doing PCM to DSD upsampling (sth my CAPS v3 was not able to do on the fly due to computing power constraints) I was barely able to detect some heat (8'C over ambient according to BIOS readings) but only when touching riser's aluminium plate. The heatpipes, and riser's top assy, were still as cool as everything else in the room.

 

Here are two screenshots I have taken.

 

24h idle:

 

24h upsampling:

 

With this particular CPU & MoBo combination, you can probably go with no heatpipes at all, just a big, solid cooler with no fan.

 

Sure, the CPU is not as fast as the one proposed by Chris, but it is no slouch either. It's Jriver benchmark score is 2300 - 4x the score of my old, Atom based Caps V3 Lagoon (did 603). And if it does DSD upsampling, it will probably also do DRC and other stuff.

 

Here is the Jriver benchmark score for this CPU:

 

=== Running Benchmarks (please do not interrupt) ===

 

Running 'Math' benchmark...

Single-threaded integer math... 5,897 seconds

Single-threaded floating point math... 3,233 seconds

Multi-threaded integer math... 1,683 seconds

Multi-threaded mixed math... 1,022 seconds

Score: 1605

 

Running 'Image' benchmark...

Image creation / destruction... 0,854 seconds

Flood filling... 0,989 seconds

Direct copying... 0,875 seconds

Small renders... 1,819 seconds

Bilinear rendering... 2,529 seconds

Bicubic rendering... 1,838 seconds

Score: 2471

 

Running 'Database' benchmark...

Create database... 0,635 seconds

Populate database... 1,391 seconds

Save database... 0,308 seconds

Reload database... 0,141 seconds

Search database... 1,631 seconds

Sort database... 1,539 seconds

Group database... 1,921 seconds

Score: 2842

 

JRMark (version 19.0.163): 2306

 

For the record - the CPU Chris has chosen did 4323. Twice as fast, but I'm not sure that extra processing power is really needed. I went for lower heat.

 

I'm gonna build a next server for a friend in the next 2-3 weeks. I think I'm gonna use the slightly faster Xeon 1240L v3, which is also a 25W TDP.

[robbbby]

From what I can see this CAPS is purpose built to be able to do everything well.

Everyone is asking why processor x was picked over processor y, motherboard x over motherboard y, etc. etc. It doesn't matter what he chose, people are going to question his choices not matter what he uses. I don't think there is anybody out there with enough time and money to try every combination of hardware to find the perfect balance between power/functionality/power usage/aesthetics. This looks like a CAPS that would be right at home simply streaming internet radio or doing on the fly DSP and transcoding simultaneously or any other high cpu usage audio/video jobs.

Instead of critiquing and questioning every part of the build use is as a guideline, nobody says you must build this CAPS with exactly these parts. Elberoth built one but used his own judgement to buy a lower power CPU, I built one and used a different motherboard, processor and case because it fit my needs and wants better.

 

And with regards to the 85W TDP CPU, that is not power being used by the CPU so your electrical consumption calculation does not make sense. Here is a description on processor TDP straight from an intel whitepaper

 

- Because TDP is a worst case value when running a “worst case” application, most processors, when running a more “typical” workload, will dissipate power that is less than the rated TDP value; how much less will depend on the application and the specific part being tested. This is true not only for the processor, but for every electrical component in a server.

- It is important to note that thermal design power is the maximum thermal power the processor will dissipate, but not the same as the maximum power the processor can consume. It is possible for the processor to consume more than the TDP power for a short period of time that isn’t “thermally significant”. For example, a processor might consume slightly more power than the rated TDP value for say one microsecond...but then consume less power than the rated TDP value for a long period of time.

[Elberoth]

Great questions unsleepable. Chris could you also tell us why you recommend the X10SLH-F over some of the cheaper X10 models like the X10SLM-F.

 

Because of it's guaranteed 7-years life ?

[unsleepable]

In case you were referring to my post, I'd like to point out that I wasn't criticising anything. I've simply asked about the thermal and power layouts, assuming that the designer's has considered these aspects and that there might be some background for the choices made.

 

These aspects are vital for any computer, but much more for one without active cooling and that uses a low-powered supply. This is not just a matter of tastes… The chassis and the power supply must allow for the system to work at full capacity in normal conditions. As I said, the numbers just don't add up, and seem to indicate that the design might be flawed.

 

I am afraid that the text you quoted only reinforces my point:

 

It is possible for the processor to consume more than the TDP power for a short period of time that isn’t “thermally significant”.

 

This means that the processor may consume more than 80W, even though the heat dissipation won't be proportional to the power consumption. Peak power consumption will of course be larger than the dissipated heat.

 

I know that the power calculation I laid out is very conservative, and that realistically higher numbers should be considered—for example, no efficiency is rated for the linear power supply, and I very much doubt that it is 100%. Linear power supplies usually have a much lower efficiency than switch-mode ones, a 50–60% figure would be quite normal. All in all, with this particular component selection and considering a linear power supply, a 250W figure would seem more appropriate if we want the system to be able to run at full capacity.

 

But again, rather than speculating maybe Chris has this information from the vendors of the selected components, and his own calculations.

 

As for heat dissipation, I simply quoted what appears in the user guide of the chassis selected for this particular design, a recommendation made by the company that makes these things.

[unsleepable]

I have went with the Xeon 1230Lv3 variant, which is only 25W max TDP.

 

Are you using the HDPLEX linear power supply?

[Elberoth]

Are you using the HDPLEX linear power supply?

 

No. I went with TeraDak linear ATX PSU. See a separate thread. I'm still waiting for one though.

[Knuck]

Does anyone have an opinion about replacing the e3-1241 v3/X10SLH-M combo suggested by Chris with an e5-2609 v3/X10SRi-F combo. The cost is pretty close but with the e5 you get 15MB of Smart Cache vs 8 on the e3, lower voltage DDR4 memory on the e5 vs DDR3 on the e3 and 4 channels of memory with 53 Gb/s bandwidth on the e5 vs 2 channels with 26 Gb/s bandwidth on the e3. Although the DD4 memory is more expensive, I don't think 16 Gb is necessary especially if you use the C.A.P.S. as an audio PC. The tradeoff is that the e5 has a 1.9 Ghz processor vs 3.5Ghz on the e3. I would think the memory benefits would outweigh the processor disadvantage if the PC is being used only for audio. There are other advantages to the e5 that I haven't listed. At similar cost, I think it is worth consideration.

 

 

 

Quick reply to this message Reply Reply With Quote Reply

[Knuck]

Does anyone have an opinion about replacing the e3-1241 v3/X10SLH-M combo suggested by Chris with an e5-2609 v3/X10SRi-F combo. The cost is pretty close but with the e5 you get 15MB of Smart Cache vs 8 on the e3, lower voltage DDR4 memory on the e5 vs DDR3 on the e3 and 4 channels of memory with 53 Gb/s bandwidth on the e5 vs 2 channels with 26 Gb/s bandwidth on the e3. Although the DD4 memory is more expensive, I don't think 16 Gb is necessary especially if you use the C.A.P.S. as an audio PC. The tradeoff is that the e5 has a 1.9 Ghz processor vs 3.5Ghz on the e3. I would think the memory benefits would outweigh the processor disadvantage if the PC is being used only for audio. There are other advantages to the e5 that I haven't listed. At similar cost, I think it is worth consideration.

 

 

 

Quick reply to this message Reply Reply With Quote Reply

[allen_bh]

Chris, thank you so much for putting this guide out for everyone. I have a ZUMA I built based on your past guide and I truly appreciate it.

 

Just a couple of observations and changes I made to build my version of a "Pipeline":

 

1. The Pipeline isn't as much a server as it is a workstation when you're using external storage on your network such as a NAS or a SAN. If you're storing all your data on the Pipeline, you'd want to run a Windows Server OS and a Xeon and a server based motherboard and ECC RAM as Chris recommends. FWIW, in my past experience, ECC RAM can be slower and has it's own issues despite additional costs, but if you're running a XEON CPU it's a must.

2. The newer Intel i7 based motherboards are optimized for multimedia so I "personally" chose to go with the i7-4790S, the Gigabyte GA-H97M-D3H, and 32GB of Crucial Ballistix (BLS2KIT8G3D1609DS1S00) with a single Samsung 256GB 850 Pro SSD. I also opted for the "Streacom ST-ZF240 ZeroFlex 240 Watt Passive PSU - Silent Power Supply" as I have a DCS Debussy and it is galvanically isolated from the USB connection. My ZUMA has a traditional (as opposed to a linear) power source and I've never had issues. I see the need for what Chris recommends if your DAC doesn't have great isolation or if you're concerned about the PC leaking noise back into your A/C line and negatively affecting the rest of your gear so YMMV if you change the design of course.

3. I also run "My Movies" off this workstation so the integrated 4600 Intel Graphics comes in very handy and it can support streaming HD video from my SAN without sweating.

4. These builds are great baselines for those who want to make a few changes and are technically savvy. They're excellent by-the-number guides to build what is being proposed.

 

Once again, awesome guide. I know I don't post much, but I certainly get a lot from reading the guides and builds here so thanks again.

 

Bryan

[Elberoth]

I have just finished my second Pipeline bild. This time I went with slightly more powerful Xeon 1240L v3, which has max TDP of 25W.

 

Here is Jriver benchmark for this build:

 

=== Running Benchmarks (please do not interrupt) ===

 

Running 'Math' benchmark...

Single-threaded integer math... 4,954 seconds

Single-threaded floating point math... 3,023 seconds

Multi-threaded integer math... 1,494 seconds

Multi-threaded mixed math... 0,876 seconds

Score: 1836

 

Running 'Image' benchmark...

Image creation / destruction... 0,385 seconds

Flood filling... 0,558 seconds

Direct copying... 0,840 seconds

Small renders... 1,376 seconds

Bilinear rendering... 1,093 seconds

Bicubic rendering... 0,640 seconds

Score: 4498

 

Running 'Database' benchmark...

Create database... 0,230 seconds

Populate database... 1,157 seconds

Save database... 0,190 seconds

Reload database... 0,055 seconds

Search database... 1,041 seconds

Sort database... 0,919 seconds

Group database... 0,686 seconds

Score: 5026

 

JRMark (version 19.0.163): 3787

 

This is way faster than my first build using Xeon 1230L v3 CPU (which did 2306) and only slightly less than the CPU Chris had chosen (which did 4323 and which has max TDP of 80W).

 

However, I just noted that those Jriver Benchmark readings are very inconsistent. I run the test several times and each time I got a different reading - from 3200 all the way to 3787 :-/

 

The computer was idling, with the only program running beeing the Jriver.

 

I also got an impression that the readings got higher after I installed the Audiophile Optimizer, which shoots down many unnecessary services.

[allen_bh]

Here's the benchmark I just got from my Pipeline:

 

=== Running Benchmarks (please do not interrupt) ===

 

 

Running 'Math' benchmark...

Single-threaded integer math... 3.738 seconds

Single-threaded floating point math... 2.273 seconds

Multi-threaded integer math... 1.145 seconds

Multi-threaded mixed math... 0.719 seconds

Score: 2413

 

 

Running 'Image' benchmark...

Image creation / destruction... 0.144 seconds

Flood filling... 0.308 seconds

Direct copying... 0.415 seconds

Small renders... 1.020 seconds

Bilinear rendering... 0.793 seconds

Bicubic rendering... 0.449 seconds

Score: 7030

 

 

Running 'Database' benchmark...

Create database... 0.180 seconds

Populate database... 0.976 seconds

Save database... 0.335 seconds

Reload database... 0.086 seconds

Search database... 0.790 seconds

Sort database... 0.897 seconds

Group database... 0.627 seconds

Score: 5526

 

 

JRMark (version 20.0.63): 4990

 

 

Not too shabby

[stevebythebay]

How does this design compare to the AudioAanZee Reference Flow Google Translate They seem to be using many of the same key parts and LPS.

[stevebythebay]

Chris: so how does this stack up sonically to the Aurender W20? Seems that has been your go-to server for sonic quality for some time. And you felt, at the time you reviewed the W20, that a CAPS approach could not achieve what Aurender has accomplished with that product. Still true???

[extracampine]

I'm still keen to hear Chris' reply to the points raised earlier about the power supply.

[firedog]

Regarding Windows 10: is microsoft not giving the free upgrade to these servers b/c of the Xeon board?

I can't get MS to let me "apply" for a free upgrade to Windows 10 from my Pipeline machine.

[firedog]

Found a workaround:

https://www.microsoft.com/en-us/software-download/windows10ISO

 

You can download and install,or make an iso disc or stick