Jump to content

AngeloVRA

  • Posts

    7
  • Joined

  • Last visited

  • Country

    Thailand

Personal Information

  • Location
    Bangkok

Recent Profile Visitors

270 profile views
  1. Just FYI, Those are specified for 18-120 AWG. The molex pins specified for 16 AWG are 538-45750-3211. You can see from the pic below that the crimp tabs for 16 AWG are longer.
  2. This is a follow-up to my initial post on my server build as I continue to work on it (seems never ending!) : https://audiophilestyle.com/forums/topic/58164-building-a-diy-music-server/?do=findComment&comment=1147932 The Taiko ATX has a 5VSB / 5A rail that has much lower ripple specs (0.01mVpp) than its other rails. How does that compare to an LPS when powering my JCAT USB XE card? I made a molex cable to power the JCAT USB XE card from the Taiko ATX 5vSB rail using Mundorf SG115 wires, same material as the DC cable from the SR7T 5v rail. Result: I was surprised that it was a big SQ difference in all areas and particularly in terms of dynamics, detail and resolution in favor of the external LPS. The Taiko ATX 5V SB rail was disappointing in this particular use case. Hence, the JCAT NET XE and USB XE cards will remain powered by two 5v rails from the SR7T The JCAT USB XE card has a power consumption of about 700 mA @ 5V. It does seem that the SR7T LPS gives better SQ than the Taiko ATX at low currents. Next, I wanted to compare powering the CPU from the LPS vs Taiko ATX. Since the Taiko ATX was consuming about 2.0 Amps @ 19v, I was estimating that the CPU alone would be consuming about 1-1.5 Amps @ 12V, making it a very viable candidate to be directly powered by an LPS. First step would be to fabricate an 8P+4P cable using the same Mundorf SG wires as I used for the EPS cable from Taiko ATX. I decided to use 1 15.5AWG + 5 18AWG wires per side. Crimping all the wires in a WBT copper cable sleeve and then soldering, with the bigger 15.5AWG wire extending past the sleeve and then soldered to the chassis connector. For safety and “best practice”, I decided to change the female chassis connectors to male. This is so that the connector at the destination side of the DC cable would be female sockets. Male pin plugs are prone to short circuits when the LPS is powered on and the destination connector wasn’t plugged in. Be mindful that the relative position of pin 2, which in my case is used as the positive terminal, is reversed with pin 1 when changing from a male to female chassis connector. You will have to swap the positions of your + and – wire. And above is the new EPS cable as installed. Some current measurements: - 185 mA with server off but LPS on - A max of 2.168 Amps during boot. It is possible that the actual max is much higher than this but my meter wasn’t fast enough to capture it. - Settles down to around 1.12 Amps when playing music This is for a Xeon W-1290 CPU running at its base frequency of 3.2 Ghz, 10 cores, HT off, Turbo off, C-states off. The SQ with the SR7T supplying 12V to the CPU was hands down much better than the 12V from the Taiko ATX. Both used the same wires and both have been burned in for at least 2-3 months. Some more current measurements, this time on the JCAT NET XE card with a) Euphony’s normal playback and b) Euphony’s “playback with network disconnected”. Current consumption goes down by about 265 mA as Euphony disables some functionalities of the card but doesn’t totally shut it off. And my last experiment was to test whether there were audible differences in the length / type of PCIE riser cables used to connect the JCAT NET XE and USB XE to the motherboard in my build. Above is HDPlex’s 20 cm x16-wide riser cable. The cable is very wide and is probably hampering good ventilation to the cards. The x16 connector is also too long for the JCAT (x1) cards and are sort of unstable. Below pic shows shorter 5cm (to USB card) and 15cm (to NET XE card), narrower x4 pcie riser cables from ADT Link. Motherboard connector is x16 so it gives a stable connection. X4 connectors to the USB/NET XE cards are also a better fit and more secure. Ventilation around the cards should be much better. As a bonus, the shorter, narrower (and different brand) of PCIE riser cables made for a positive and audible SQ difference. Relatively, the HDplex riser cable had more lower treble energy ans had less body. When it finally becomes available, it might be a good step up for me to change to the HDPlex H5 case so I can plug in the cards directly to the motherboard. Next on the horizon would probably be upgrading the standard footers. Once again, I thank all AS guys who have been generously posting and sharing their experience and knowledge!
  3. If you mean the 2nd 8 pin EPS cable for the CPU, I would say Yes, Its worth it. My build wasn't an AMD though. It was a Xeon W-1290 on Gigabyte Vision D which had 8+4 pin CPU EPS connectors. I started wiring with just 2 pins and added 1 pair at a time and heard an uptick in SQ, mainly in fullness and dynamics. I ended up fully populating the 8+4 connectors.
  4. @rico25 Found a pic of my experimental silk tubing covered DC cable.
  5. @rico25 The teflon tubes were a substitute for the standard teflon insulation of the insulated Mundorf SG wherein there is 100% contact between the wire and the insulation. I bought the cheaper bare, uninsulated Mundorf SG and placed them inside oversized teflon tubes so that there was minimal contact between wire and insulation. I didn't try cotton/silk tubing on its own without any other insulation as that didn't feel "safe" enough for me. My experiments with cotton and silk tubings were placing them on top of the oversized teflon tubes. My trials with JSSG360 was the same, with copper braid over the teflon tubes, teflon tape wrapped, then another copper braid layer, and finally a thin expandable PET braid. In both cases, in my system and to my ears (YMMV), transients, presence and immediacy sounded diminished, as if it was "over damped". I tried both for a couple of days but ultimately preferred leaving the teflon tube insulated wires bare. The above is for the external DC cables connecting the server to my PH SR7T. I did not do the same experiment on the server internal wiring as it was too stiff and putting on/taking off silk tubings and JSSG360 would mean disconnecting the pins from connectors, straightening the wires, installing the tubings, shaping the wire to fit and assembling the pins into the connectors again. For the internal wiring, I just used very thin PET expandable braid to organize the wires for a relatively neater install. Hope that helps. These things are very system dependent so I would encourage you to experiment on external DC cables in your own system.
  6. @ASRMichaelmany thanks for the appreciation 🙏🏻 ! Yes, I tried allocating 1/2/3 more )and less) cores to gstp . Sonic character changed but “0-1 stylus 2-5 gstp” still sounded the best. @Exocer Thanks for all the tips you generously shared! Most worked for me and some didn’t but all taught me something. Compared to the NUC NimitraS (also powered by PH SR7T) that it replaced, the new server is several steps up in all areas. Most notably it has more weight and gravitas, big room filling sound, macro and micro dynamics, more nuance, and much less digital edge/harshness. @mrkovenThanks!! Yeah, the wiring is the main DIY issue as the rest is just plugging things in. It also had a notable impact in my system. @flkin also Looking forward to your critique and improvement suggestions next week!! 😀😀 .
  7. Various posts of AS people building music servers have inspired me to build my own. I figured its my turn to share my humble build as well I started by ordering a 4 rail SR7T from PHD back in October last year. Although there were (extended) delays, I was one of the lucky ones to get mine just a few weeks before PHD closed down its business. I was able to get 4 rails on it: - 19V/10A adjustable to 15V/6A, 12V/5A, 9V/4A - 12V/10A adjustable to 9V/7A, 7V/5A, 5V/4A - 12V/6A adjustable to 9V/4A, 7V/3A, 5V/2.4A - 12V/6A adjustable to 9V/4A, 7V/3A, 5V/2.4A If I could do it all over again, I would opt for 5V/6A instead of 5V/2.4A from the 12V/6A rail. I wanted to use an HDPlex H5 chassis (so I could install USB/Network card direct to motherboard without use of riser cables) but was informed they were out of stock, probably till August so I decided to go with a Streacom FC10a instead. Having decided to mount the Taiko ATX DC-DC converter behind the front panel, I disassembled the front panel to drill and tap some blind holes so I can securely fix the Taiko ATX onto it. 2 Jaeger 3-pin terminals were also installed on the rear panel. 1 to receive the19V to supply the Taiko ATX and another to receive 12V to supply the CPU EPS if I didn’t like the SQ of 12V EPS from Taiko. I confirmed most appropriate orientation of the Taiko ATX (connectors up or down). Connectors on the bottom yielded a position that would be much easier to wire. This position also gave the shortest possible 24 pin ATX cable from Taiko ATX to motherboard For the CPU, I opted for a single Xeon W-1290 which has 10 cores, base speed of 3.2Ghz, Turbo speed of 5.2Ghz. This went onto a Gigabyte W480 Vision D full size ATX motherboard. This setup required the Streacom Thermal Riser HT4 so that the heatpipes could vertically clear the VRMs, and the extended length heatpipe set LH4 as the standard ones were about 2 cm too short. Initially, I only used 2 pins (+12V and Gnd) for the CPU EPS cable. For the ATX cable, also the minimum, 1 pair each for 12v/5v/3.3v and their respective grounds. I listened to it for over a week and then gradually added 1 pair of wire at a time. Each time, the additional wires gave me an uptick in fullness and dynamics. The wire itself could probably carry the current but the limiting factor was the current capacity of the pins. Utilizing more pins allowed more transient peak current to be delivered. I ended up fully populating all power connectors. The molex pins have two tabs on the top that slide into channels inside the Molex connector (e.g. both should be at 12 o’clock for the molex pins to slide in and lock properly. The Mundorf SG 1.5mm (AWG 15.5) is very stiff so I had to make the pin for one end, insert it into the connector at motherboard side, shape the wire towards the connector at Taiko ATX side, and only then crimp the molex pin at that end. This way the pin would be orientated correctly to slide in and lock properly. I opted to use the gold plated Molex pins because they have a bigger current capacity. There are also 2 sizes: 1 for 18AWG wire, and another for 16AWG wire. I opted to use bare Mundorf wire and use oversized Teflon tubes (2-2.5mm for 1.5mm wire, 1.5-2mm for 1.0mm wire) so that the resulting insulation would be mostly air. As a bonus, this was also much cheaper. Crimped solid core wires have a tendency to shake loose after some handling so I decided to also thinly solder the already crimped connection. Next was the power supply cable from the Jaeger chassis connector at the back of the server to the 6p input connector at the Taiko ATX. While a single pair of 1.5mm Mundorf SG should be able to carry the required current, the limiting factor would be the current carrying capacity of the molex pin contacts. As mentioned earlier, I had decided to use all 6 pins. For each polarity (positive and Gnd), I opted to use 1 length of 1.5mm AWG15.5 wire plus 2 lengths of 1.0mm AWG18. On the Jaeger connector side, simply twisting the 3 wires together seemed messy as the resulting wire won’t fit neatly into the solder cup of the Jaeger connector. I decided to use a WBT copper cable sleeve to crimp the 3 wires together, with only the 1.5mm wire coming out the other end which can then be neatly soldered into the Jaeger connector terminals. RAM is Apacer ECC industrial, wide temp (-40C to 95C), 4 x 4GB for a total 16GB. Music data storage is 2 x 2TB Intel 660p M.2 NVME which also has very low power consumption (40 mW at idle, 100 mW when active) OS drive is an 118 GB Intel Optane 800p which is relatively cheap, has very low read latency at 7 microseconds, and low power consumption (8mW at idle, 3.75w when active). This was mounted on an M.2 to PCIE adapter card which allowed me to install it on the 1st PCIE slot, direct to CPU. The 2nd PCIE slot, also direct to CPU has a JCAT USB XE, externally powered by a PH SR7T. The 3rd PCIE slot, also direct to CPU (PCIE bifurcation of x16 into x8/x4/x4), has a JCAT NET XE, also powered externally by a PH SR7T. Subsequently, I found out that the M2 Optane 800P needed a substantial heatsink as it really gets hot. All DC cables are Mundorf SG 1.5mm (AWG15.5) wires in oversized 2.5mm Teflon tubes with Jaeger connectors at the source and either another Jaeger or an Oyaide 5.5/2.1 mm DC plug on the other end. Surprisingly, cotton/silk tubing and JSSG360 shielding didn’t really work out for me in this scenario. In operation, I measured the following current consumption: - 19V to Taiko: 2.0 amps - 12V to EtherRegen: 800 ma - 12V to JCAT NET XE: 1.25 amps - 12V to JCAT USB XE: 680 ma BIOS Settings: After several weeks of trying various combinations, Ive settled on the following settings: - Turbo off - Hyperthreading off - C-States off - Onboard Lan Disabled - Onboard Audio Disabled - SATA disabled Euphony is currently on Version 421 and will try 716 soon - Ramroot disabled - CPU isolation is 0-1 stylus 2-5 gstp 6-9. Surprisingly 0-1 gstp 2-5 stylus 6-9 sounded different and worse. Probably because some cores run better than others. - Allocating IRQs to specific cores didn’t sound better for me. Many thanks to all AS DiYers who have posted their builds and have unselfishly shared their knowledge and experience!
×
×
  • Create New...