Gauging interest around some custom hardware and cabling for GPU rigs

[sidehack]

Initial testing on the Molex adapter board looks good. It's currently drawing about 18A off the 5V rail which is good since I intended to give it a 15A rating. A little airflow goes a long way too, just saying. Tomorrow I'll test how well it handles the heat of that much 5VDC conversion while drawing about 20A of 12V through the PCB.

I'm glad it's working well right off the bat. Maybe I'll put up pictures tomorrow.

[nerdralph]

Prototype PCBs for the Molex adapter have finally arrived. I'll be assembling and testing in the next few days and hopefully they work well and reliably.

I have already tested the new Dell 750W PSU board and it's in need of slight revision (including adding PWM fan control as suggested in this thread; it's been tested but not yet integrated), so I'll probably have another prototype PCB order going out soon.

If you're going to do PWM, I'd suggest adding temperature-based speed control.  There's a i2c temperature sensor (LM75 I think it was) on the i2c bus (pins B5/C5).  I forget the address, but you can find it with an i2c scanner; it's the lowest address of 3 you'll find on the bus.  Never figured out what the other 2 devices, but the LM75 gives 2 temperature readings in Celcius, one of which is off the heatsink for the mosfets.  I made a circuit with an attiny13a for PWM fan control (pin B2 I think), varied by ADC reading from a POT.  My plan was to do a bit-bang i2c implementation to read the temperature and make the fan speed temperature dependent, but still haven't got around to it.

[Nikolaj]

I have done a lot of deals with sidehack in the btc threads.

I have been active with btc eth and etc.

It is good to see you post here.

I will try to figure out what I could use.

R9 390 gpus need 8 pin

So a six pin to six + two pin  cable would work.

i have a lot of servers in house I would love to,adapt,one to run a pc

be careful with the loads, the wires needs to be thick, and the terminals needs to be soldered well (and the psu, capable in the rails distribution). The 390s and 290s are beasts in power consumption, potentially.

[MarkAz]

be careful with the loads, the wires needs to be thick, and the terminals needs to be soldered well (and the psu, capable in the rails distribution). The 390s and 290s are beasts in power consumption, potentially.

The quality of the cables from sidehack are night and day from the junk your typical PSU comes with... That's the advantage of getting them from someone who knows mining - they know that these cables are really going to be used.

[Nikolaj]

I know them perfectly, that's why I suggested to have caution. Surely they are

It's also dependeant by the load, for example 2x8 pin PCIE, linked, on a silverstone SST 1500W Gold aren't appropriate if you power an HD 7990.

You can easily melt the terminal in the female connector of the PSU, a thing that doesen't happen if you use them for a 290x, due the Amperage and thickness of the Copper in the cable (AWG).

I am curious to see the results sidehack, you seem to be doing a good job

[sidehack]

Testing on the Molex board looks good, so I have no qualms about manufacturing that guy.

I'm working now on designing ATX adapter boards, but figured I'd ask a question before finalizing anything.

A standard ATX PSU outputs 5V standby power until the motherboard sends it an "on" signal, at which time it fires up the other rails (3.3V, +/-5V, +/-12V). I intend to design a full-feature board with all these rails, including a switched (and on a fancier version, probably regulated) 12V rail for motherboard power and EPS, so that a computer can be run properly from a 12V source without requiring disconnection to safely and completely power down.

A simpler board would meet most people's needs, with an unswitched 12V rail and the like. Should it be designed with 5V standby and motherboard-signaled rails, or is a simple "rails come up when 12V is applied" board good enough for most folks?

EDIT - regarding cables, I have to use 18AWG on anything that gets two wires to a pin (so all leads on 6-pin splitters and one GND on 6+2 which jumps to the +2) but everything that's a one-to-one gets 16AWG wire. Been that way since I started making cables 3 years ago; in fact, I didn't even start stocking 18AWG until last month.

[MarkAz]

Sidehack just sent me the first trail shots of doing sheathed cables, here you can see them next to his normal cables.  In this case it's all of my PCI-e to PCI-e ones, so one splitter and one 6-pin to 8-pin:




I'm not overly surprised, but I like them quite a bit more than non-sheathed - well worth the extra money.  In terms of a practical build, it really helps to clean up the internals and make any wiring quite a bit more convenient.  The #1 is still getting cable lengths that are correct for your build, but this is a nice #2 values.  Here's the before and after - I shouldn't need to tell you which is which. 

[nerdralph]

where is the molex for powered riser getting its power? from the breakout board?

Right now the power comes from the 12v supply of the breakout from the PSU, and from the 5v step-down converter to the right of the PSU.  It's kind of a clunky setup, and Sidehack is working on a board which will take the 12v from the breakout and break it out to the 12v and 5v that the molex needs - it should be great and really clean things up.

You could probably skip the 5V step-down and provide just 12V to the molex.  PCIe has no 5V supply, just 3.3 & 12V, so the riser board has a 3.3V regulator powered by the 5V line from the molex connector.  Although I have not measured it, I suspect the GPU cards use so little power from the 3.3V line that the extra power dissipation on the 3.3V regulator wouldn't be a problem.

[sidehack]

Assuming the 3.3V regulator is okay with taking in 12V, and nothing else pulls from the 5V (depends on the riser I'd guess), that only increases the power dissipation on the 3.3V regulator by about 5 times.
A proper 5/12 Molex adapter is also handy if you want power for other necessities like a hard drive.

[Newwsr]

We can directly connect the psu without board
https://www.youtube.com/watch?v=COOnxZBQ_-c

[sidehack]

Which is great if you don't want to be able to change anything with your cabling, and it's also great if your PSU doesn't ever break. A modular solution is, in general, a better solution.

[Newwsr]

I'll Call 4 r9 280x it

[MarkAz]

I did some testing on my test platform with R9 390's, and they were pulling about 2a per card peak on the 5v side, with steady state being closer to 1a.  It's annoying that they step down 5v instead of 12v so we could just use PCIE connectors.

[sidehack]

MarkAz has had his order of test boards and cables for about a week now, and I've asked if he'll get some pictures of an installed setup. I just now found my camera so here are some standalone pictures:





This board is 1.2x3.4 inches with three screw holes for mounting; it takes in power from 2x PCIe 6-pin jacks (only one may be necessary depending on your load) and outputs 5V and 12V on three 4-pin connectors. With moderate airflow for cooling, the board will comfortably handle 20A on both 12V and 5V lines simultaneously. I can provide cables of any desired length, with or without sheathing, from the 4-pin jack to standard 4-Molex for risers and other computer peripheral hardware. I can also provide 6-pin or 6+2-pin cables for powering GPUs. Cables are custom-made to order in the USA from 16AWG and 18AWG wires, no cheap Chinese stuff.



I have PCBs and parts on order, so if anyone's interested in the Molex adapter boards they'll be for sale in about three weeks. I figure $15 apiece on this batch; if a second batch is made in larger quantity I might take the price down a bit to account for bulk discounts on parts.

If anyone is interested, let me know.

I'm working on an ATX24 adapter design which should complement this board nicely, but I've been kinda busy lately so that's a bit stalled.

[xleejohnx]

I would be interested in psu if you had one laying around?

[sidehack]

https://bitcointalk.org/index.php?topic=940317.0

A complete (and updated just this morning) listing of everything I have available right now.

[kilo17]

I did some testing on my test platform with R9 390's, and they were pulling about 2a per card peak on the 5v side, with steady state being closer to 1a.  It's annoying that they step down 5v instead of 12v so we could just use PCIE connectors.

I have switched to exclusively using the following risers:  They are 9.5" and I do not have any issues with them at all unlike the USB risers which have a short half life.  They are sourced from Padrino. I definitely need to order some stuff from Sidehack.  Love your new stuff.

[philipma1957]

MarkAz has had his order of test boards and cables for about a week now, and I've asked if he'll get some pictures of an installed setup. I just now found my camera so here are some standalone pictures:





If anyone is interested, let me know.

I'm working on an ATX24 adapter design which should complement this board nicely, but I've been kinda busy lately so that's a bit stalled.

so how much is 1  board and 3 wires?

Also I was thinking my psu has 2 cpu mobo cables  each is 4pin + 4pin

so if I want to is it possible for a cpu cable to six pin pci adapter cable  to be built.

I would connect the spare cpu to the adapter cable to your small board.
see photo of psu

[nerdralph]

I did some testing on my test platform with R9 390's, and they were pulling about 2a per card peak on the 5v side, with steady state being closer to 1a.  It's annoying that they step down 5v instead of 12v so we could just use PCIE connectors.

I have switched to exclusively using the following risers:  They are 9.5" and I do not have any issues with them at all unlike the USB risers which have a short half life.  They are sourced from Padrino. I definitely need to order some stuff from Sidehack.  Love your new stuff.

I'd say you have been lucky.  PCI-e signals are in the hundreds of Mhz, and are spec'd to use differential pairs or impedance-controlled PCB traces.  USB 3.0 cables are designed for high-speed signalling, while flat ribbon cables often will have problems once you get much over 100Mhz, even for low-voltage signals like PCIe.  I still use ribbon risers because they are cheap and I can get away with powering them from the mobo, but I have a couple of USB3.0 risers for testing when I have a bad ribbon riser.

You're also generally better off with the shorter ribbon risers (15 or 20cm), since the longer cables attenuate the signal more.

[kilo17]

I did some testing on my test platform with R9 390's, and they were pulling about 2a per card peak on the 5v side, with steady state being closer to 1a.  It's annoying that they step down 5v instead of 12v so we could just use PCIE connectors.

I have switched to exclusively using the following risers:  They are 9.5" and I do not have any issues with them at all unlike the USB risers which have a short half life.  They are sourced from Padrino. I definitely need to order some stuff from Sidehack.  Love your new stuff.

I'd say you have been lucky.  PCI-e signals are in the hundreds of Mhz, and are spec'd to use differential pairs or impedance-controlled PCB traces.  USB 3.0 cables are designed for high-speed signalling, while flat ribbon cables often will have problems once you get much over 100Mhz, even for low-voltage signals like PCIe.  I still use ribbon risers because they are cheap and I can get away with powering them from the mobo, but I have a couple of USB3.0 risers for testing when I have a bad ribbon riser.

You're also generally better off with the shorter ribbon risers (15 or 20cm), since the longer cables attenuate the signal more.

Not sure how much the signal translate to mining issues but I know that fried and non working USB risers cause a lot of downtime and headaches.  I point out on another thread that I have a BIG box of fried and nonworking USB risers.  

One of the biggest problems came when using Claymore's Dual Miner (which I abandoned a while ago due to the extra stress on the hardware). I fried 20+ USB risers which causes a lot of downtime.  The weakness is in the molex/sata wires which frequently burn but also they just stop working.  It occurs more after a couple of months but it happens and I have yet to have a problem with any of the above ribbons.

(I agree on the shorter ribbon cable but they are a little to short for 6+ GPU rigs.)