Building a DIY Music Server

[JJSim]

IIRC it has been mentioned the Noratel transformer has its own internal soft start circuitry, hence the suggestion to omit the Hypex soft start module when using that transformer. This is an important clarification, as one should continue to utilize the Hypex soft start module with different transformer brands.

[JJSim]

4 hours ago, Johnseye said:

Hi all, now that I have a Taiko case I'm starting a new DIY Taiko Sage build. Given Taiko is using custom RAM, what low speed, low refresh DIMMs are compatible with the Sage board? 

 

Any 288 pin DDR4 ECC RDIMM/LRDIMM should work. Max frequency supported by that motherboard is 2933Mhz. Frequency is further dictated by the CPU; in the case of the Xeon 4210 it's limited to 2400Mhz. Memory refresh rate can be adjusted in BIOS.

 

Apacer has part # 78.D1GMP.4010B that would work with the Xeon 4210. Note that it is not the wide temp version that is more desirable. Samsung also has a good selection of server memory.

[JJSim]

5 hours ago, baconbrain said:

 

Curious as to where one can find the details regarding the cpu / memory frequency limitations. Do you have a specific reference?

 

Update: Nevermind, I found the information in the Intel CPU Specifications. 🤭

As you discovered, the spec list will mention the max memory frequency supported by the CPU. Same for motherboards.

 

I also want to clarify my original post. You can run memory at a lower frequency than rated, meaning you can run memory rated at 3200Mhz at 2400/2666/2933Mhz. To that end, Apacer has part # D51.27267S.002 in 16GB capacity, and part # D51.27244S.002 in 8GB capacity that would work with the Intel scalable CPU's. These are the wide temp "industrial" grade memory.

[JJSim]

Yes, for maximum bandwidth it's recommended to fill each DIMM slot, so for the dual socket Asus Sage that's 12 total DIMM's. If you use a CPU that allows for max 2400Mhz memory, you can save cost with the lower frequency DIMM's. The SKU's rated at 3200Mhz are more expensive.

 

Generally Samsung memory is lower priced. Apacer uses Samsung IC's on their memory, so I would expect quality and performance to be on par, though there has been high praise for the Apacer wide temp DIMM's on this forum.

[JJSim]

Mundorf Angelique is a new option. I bought 14 meters of it for my server build. Much lower cost than the Mundorf silver/gold.

 

Best price is from Audiophonics (price is by the meter).

 

https://www.audiophonics.fr/en/hook-up-wires/mundorf-caw115bk-wiring-cable-copper-ptfe-18mm-p-16651.html

[JJSim]

I agree–15awg is probably too large and stiff for the ATX cable (if utilizing all 24 pins). I will be using 18awg for the ATX, and 15awg for the EPS.

 

https://www.audiophonics.fr/en/hook-up-wires/mundorf-caw110bk-wiring-cable-copper-ptfe-079mm-p-16650.html?search_query=mundorf caw&fast_search=fs

[JJSim]

Quite a few people around here have. Ghent Audio offers their cables with JSSG360 as an option. It appears Taiko also uses JSSG360 on their ATX and EPS cables in the Extreme. I don't have any experience with it, but plan to use it on my cables.

[JJSim]

It would be an interesting comparison between low power yet slower/higher latency drives vs high power yet faster/lower latency drives. For example, the Intel Optane P5800X is the lowest latency SSD available, but can draw over 21W when active (for the 1.6TB version, the 400GB version draws 15W).

 

Taiko utilizes the Optane 900P for the OS drive which can draw over 14W, so it's pretty obvious which camp they sit in.

[JJSim]

For their new SG10 Streacom partnered with Calyos, which has an interesting portfolio of products. Their "2-phase" cooling solutions are particularly intriguing for use in DIY servers, by using a refrigerant instead of water in a 100% passive system without the need for a pump to circulate the fluid.

 

https://www.calyos-tm.com/technologies/loop-heat-pipes

 

It doesn't appear they have any off the shelf products, so anything they offer would be custom made and probably very expensive.

 

 

[JJSim]

Conceptually, yes. Although the use of a refrigerant instead of water is unique. Utilizing a dedicated condenser in a closed-loop system is also different than the traditional use of a metal chassis wall acting as a heatsink.

 

Only first hand experience will tell if this solution is a viable alternative to present offerings. Turemetal released their passive chassis a while ago, and then Taiko developed an effective solution with their heatpipe coolers.

[JJSim]

18 minutes ago, littlej0e said:

  

 

 

Now THIS is interesting...

 

I did some testing this past weekend with a couple of higher core Gold and Platinum CPUs. Most ended up beating the 4109Ts, 4110s, 4210s, and 4214s quite handily in nearly every category, save for naturalness and texture (Taiko/Emile definitely new what they were doing using the 4210s). Differences between some of the CPUs were not small. Unfortunately, the higher core CPUs cannot operate exclusively using passive cooling on the Taiko DIY chassis (without tripling or quadrupling the heat sink surface area at a minimum). I ended up using active cooling with external fans/fan controller hooked up to a separate power supply just so I could test other CPUs without melting my rig. The results were so promising, however, that I am now considering purchasing a dedicated, externally controlled/monitored/powered liquid cooling chassis (which means my music server would then be comprised of three separate chassis because...stupid lol. Perhaps I should call this build "The Faiko Extreme - Extra Chromosome Edition").

 

In any case, there does appear to be a loose, but tangible, correlation between core count and audio performance specific to Roon. Though to what degree and how well it scales, I'm not really sure. Emile's design proved that high power servers can achieve equal, or better, performance compared to low power. So I can't help but wonder what would happen if we take Taiko's design methodology a step further and combine it with an even higher power ULPS and higher core count CPUs. Could we achieve performance beyond that of a legit Taiko Extreme? Seems worth investigating seeing as DIY'rs have the luxury of not being constrained by commercial viability.

 

Any idea what the "real world" heat dissipation/relative performance is of the Calyos solution(s)? They seem to support up to 2 x 350W CPUs, but I'm always skeptical. 

 

Thanks for reporting your findings. It's long been a topic of debate over low core/power CPU's vs high core/power CPU's. Looking at the market of music servers available today, you can see the division between the two ideologies. The question always remains, however–how much is too much? Even a 12-core CPU is left twiddling its thumbs under a normal music playback workload (save for upsampling in HQPlayer). You can argue that even with extreme core counts and TDP, with such a small workload placed on the CPU it will draw less power and generate less heat than it's rated for. And having more cores allows for the workload to be distributed more evenly across the cores. But at what point is there diminishing returns?

 

There are also opponents and proponents of active cooling. On one hand active cooling can create vibrations and noise (both audible and electrical) that are detrimental to sound quality. On the other hand high temperatures can put greater stress on components, and keeping components at a lower temperature can improve sound quality. The best approach is a well implemented passive cooling system that is capable of dissipating the total amount of heat produced by the server. This is also the most difficult to develop. Taiko put much effort into developing the heatpipe system in the Extreme. But sometimes additional cooling is still needed. Taiko had to redesign their network card to include a copper heatsink as the cards were shutting down due to high temperatures. This buildup of heat inside the chassis can have a parasitic affect on the entire server. Active cooling, even in the form of an externally powered fan blowing air over the components, can have a positive result on sound quality.

[JJSim]

Nothing more I can really suggest. I haven't even completed my server build, but have been studying the cooling aspect as I expect it will present one of the greatest challenges. The CPU's I am considering using all have TDP over 100W, but the only upsampling I intend to do is to high rate PCM through HQPlayer, which presents a very low load on the processor. But as I alluded to in my previous post, there are components other than the CPU that generate heat and the cumulative buildup of temperature can have a very negative impact on performance. I expect that I will at first try all passive cooling and then experiment with adding a slow spinning fan inside the chassis to circulate air. There are ways to address the negatives of active cooling (soft mounting to lessen vibration, externally powering fans, etc.) The problem with AIO coolers is that there are now two sources of vibration and noise; the fans on the heat exchanger and the pump that circulates the fluid. This is why the all passive design presented by Calyos is intriguing. Or the fully external approach that you described.

 

IMO Taiko has the most elegant solution with the heatpipe coolers offered with the DIY chassis. Last I inquired they were able to sell just the coolers, but I have decided to use a 3rd gen scalable processor which uses a different socket.

 

I was considering sending an inquiry to Calyos but I expect they will need detailed measurements and even CAD drawings, all of which is over my head. I am interested to hear what they tell you.

[JJSim]

13 hours ago, MarcelNL said:

you can design your own passive cooling system, but you'll have to bend your own heatpipes and route your heatsink....two large -ish 300W heatsinks should be perfectly capable of taking care of around 125W TDP CPU's

 

What TDP are you looking at? 

 

I looked into building my own passive cooling system, but the problem is finding a heatpipe base for the LGA 4189 socket. Dynatron has a selection of passive coolers, but these are intended to have airflow moving over them in a server chassis. The N8 model cooler from Dynatron could be modified, but it would require pretty involved surgery to remove the aluminum fin stack and then add longer heatpipes to reach the side wall of the chassis.

 

https://www.dynatron.co/product-page/n8

 

I did ask Taiko if they could make a heatsink for the LGA 4189 socket. They told me they would look into it but no word from them since. I doubt they are interested in developing one since the market for it would be very small.

 

I've also considered just adding a CPU cooler with a fan, but that would only cool the CPU and do nothing to remove the heat produced by other components. I am tempted to just add a large, slow spinning fan inside the chassis and not have to worry about temperatures at all. Like I said before, there are ways to minimize the negative impact of active cooling.

[JJSim]

On 6/6/2023 at 3:36 PM, baconbrain said:

 

 

Out of curiosity, I reached out to Calyos and inquired if they also do solutions for normal consumers but unfortunately they only do B2B. 

 

That's what I expected. I was reading an article that mentioned they have military contracts, so not surprising they don't cater to the consumer market.

 

And so, the search for passive cooling solutions continues...

[JJSim]

21 hours ago, MarcelNL said:

what unmet need do you have? As in what TDP are you thinking of using?

 

I am favoring the Intel 4314, so 135W TDP. I decided on a 3rd gen scalable processor for multiple reasons. The only upsampling I intend to do is PCM-PCM, which presents a light load on the processor so during use the CPU won't be pulling anywhere near 135W. The problem is finding a passive cooling solution for the LGA 4189 socket. The Taiko heatsinks only fit the LGA 3647 socket, likewise for the HDPLex heatsinks. Dynatron has passive heatsinks that will work, but they are still intended to have forced airflow blowing on them.

 

https://www.intel.com/content/www/us/en/products/sku/215269/intel-xeon-silver-4314-processor-24m-cache-2-40-ghz/specifications.html

[JJSim]

8 hours ago, AudioDoctor said:

 

Check out, The Beast.

https://www.monsterlabo.com/

 

I know about that one, but I don't like the appearance of PC cases for a music server. I also already have the chassis I will be using (Streacom FC10 with Audio Optimized rear panel).

[JJSim]

9 hours ago, Kuba7777 said:

Is there a Taiko DC-ATX converter in Taiko Extreme?

thank you 

 

This post from member JayM indicates that the Extreme does use the Taiko DC-ATX. Unless the Taiko DC-ATX was developed for use in the Extreme but was never put into production units. AFAIK, no one has been able to capture a photo of what is underneath the capacitor board (the second Lundahl choke is visible, however).

 

Based on the topology of the Extreme's single rail power supply (single transformer secondary, single active rectifier, single capacitor array) plus the confined space underneath the capacitor board, it seems unlikely that Taiko is utilizing linear regulators for the different ATX, EPS, and expansion card voltages.

 

On 8/31/2022 at 5:48 AM, JayM said:

As confirmation to Nenon's post regarding R&D this is definitely a big cost driving factor, this is the reason we have to design our products with different markets in mind since there is no way R&D can be supported by only the DIY branche. This is the reason the ATX is designed that way, the holes and cutout corner of the copper plate make the DC-DC ATX mountable in te SGM Extreme under all the capacitors otherwise this would not not have fit. Emile really wants to support the DIY market purely out of joy, because he started that way himself. I can vouch for the fact that we have little to no margin on the DIY products as is, so having to PCB's made one that would have fit in the SGM Extreme and one purely for the DIY market would have increased the price many times over since the market is just to small. We do not want to do that to the DIY community.

 

[JJSim]

1 hour ago, dctom said:

Does anyone have experience of a good current mATX motherboard for a small server build?

I am using this motherboard in my single processor build:

 

https://www.supermicro.com/en/products/motherboard/x12spm-ln4f

 

This motherboard has many advantages:

 

1) Server/enterprise grade quality

2) PCIe 4.0

3) All three PCIe slots are CPU direct

4) Accepts up to 3200Mhz memory

5) Has SFF-8654 SlimSAS connectors for CPU direct connection of U.2 SSD's (such as 900P/905P/P4800X/P5800X). This connection method has a side advantage of allowing these drives to be externally powered by a 12v source with the correct cables.

6) Has robust VRM design as it can power processors with up to 270W TDP

7) Has an on-board BMC controller for graphic output, eliminating the need for a dedicated graphics card

[JJSim]

15 hours ago, jean-michel6 said:

Which processor do you use ? 
How is done the cooling ? 

 

My server is not yet completed, but I am choosing between these three processors:

 

https://www.intel.com/content/www/us/en/products/compare.html?productIds=215283,215277,215269

 

TDP's are 135W, 120W, and 105W respectively, so not an insignificant amount of power. Remember that TDP is the measure of power dissipation when all cores are under a complex load. I do not intend to upsample to high rate DSD, so any of these processors will be practically idling when in use and power consumption will be far lower than the rated TDP.

[JJSim]

On 12/9/2023 at 5:22 PM, jean-michel6 said:

Thank for the answer . 
 

How you will handle the cooling since this mb has a very specific cooling tower intended for use in server farm where they blow very cold air on it . G

There are passive CPU coolers available for the LGA-4189 socket, but they are designed to have forced airflow moving over them. I did contact Taiko to see if they could make a heatpipe cooler for this socket and was told they would check into it, but I assume it's not something they are interested in.

 

I plan to try using this heatsink with no forced airflow to see if it is sufficient:

 

https://www.dynatron.co/product-page/n8

 

My hope is that setting the CPU to base frequency and possibly undervolting will lower power consumption to the point where this heatsink will be enough. As I mentioned prior, I do not intend to do heavy upsampling so the load placed on the processor will be very low.

[JJSim]

Most desktop motherboards and CPU's are intended to have forced airflow cooling, so those of us building silent music servers are in the fringes. If asked, I anticipate that almost every motherboard and CPU manufacturer will say they do not recommend passively cooling their products.

 

The Dynatron heatsink I posted above isn't much different in terms of function and mechanics than the heatpipe coolers that some have DIY'd and that HDPlex and Taiko developed. The Dynatron heatsink has a copper vapor chamber base and copper heatpipes mounted in an aluminum fin stack. The main difference is the substrate the heatpipes are mounted to (aluminum wall in a Streacom/HDPlex/Taiko chassis vs aluminum fin stack with the Dynatron heatsink). With numerous small holes placed in the lid of the chassis, a certain amount of convective cooling can be achieved with warm air rising out of the chassis and cooler air being drawn through the sides and bottom of the chassis. This small amount of airflow may be sufficient to keep CPU temps down.

 

Of course this is all theoretical until I actually complete my server and experiment with passive cooling. I am not that worried if active cooling is needed, as there are advantages to active cooling and there are ways to mitigate the negative impacts.