Topic: FPGA Rig Photos - page 2. (Read 43785 times)

Anyone else here with over 4 Ztex single boards finding that they leak a HELL of a lot of power back into the USB connection?

All my Ztex FPGAs connect to pukka powered USB hubs. Each USB hub has a separate 5V 1A power cable. The USB2.0 spec says that each hub supports 4 connections, so the reason why there are so many 7 port hubs (a weird number in computing, generally) is because they have two hub controllers ganged together inside. I've pulled enough cheap USB hubs apart (that wouldn't work with my FPGA setup) to find that this is the norm, and even the more expensive Belkin kit does the same thing, and uses the same controllers.

My 20 unit FPGA board has a very low power integrated AMD logic board underneath and an 800W ATX PSU. The separate power rails use PCIe power leads from the PSU. This should be *massively* overprovisioned for power, and I've soldered up the power rails and 2.1/5.5 barrel plugs using 18AWG cable, which should be adequate at 12V to supply only 5 of the Ztex 1.15x boards. Fans are powered from a separate circuit.

This should be over-engineered... but the amount of trouble I had getting USB hubs to work surprised the hell out of me.

Using a passive hub and 5 FPGAs (two 4-port hubs, one with three FPGAs connected, the other with two FPGAs and the input from the first hub) took hours of trial and error to get all the FPGAs recognised. Stefan mentions repeatedly in his online documentation that multiple ground cables are required, and the problems I saw appear to be due to current returning from the FPGA board via the USB cable ground. These are tiny wires and can't carry much current.

I think this may be a very simple question so I've put it here... if it's more complicated then it ought to have its own thread.


I like the quality, proven performance, warranty and support that Ztex gives... but even following his 'cluster power supply' advice re: ground cables vs. power cables, I ended up taking a few days and various USB hubs to get all the units working. The logic board had plenty of powered USB sockets and I had to use most of them. I really want to use my Raspberry Pi to control a huge FPGA assembly, but the strict current limits on the Raspberry Pi USB socket seems to make this impossible.


Anyone know what's going on, from an electronic engineering perspective? Are ALL the competing boards going to have similar issues - i.e. am I doing something obviously wrong? Is there any way to sling a resistor into the USB ground return wire (or some other solution - I'm not an EE) to discourage the FPGA current from returning via the USB cable, and instead following the low-resistance multiple-ground-wire route like it should?

My 20 unit assembly ended up using 6 USB hubs, two of them powered, thus a snakepit of cables below the tidy top level and lots of power points required.

I know the total power draw is low and the system is incredibly efficient. But if I wanted to build a 100 unit assembly? I'd need 30 USB hubs, which is bloody ridiculous.

Big backplanes are the way to go but only if the data, as well as the power, is aggregated on the backplane.

I'd really appreciate feedback from those with first-hand experience of the kit I'm most likely to buy (Ztex, Enterpoint, BTCFPGA.com, any others competitively priced in EU?)...

The USB2.0 spec says that each hub supports 4 connections, so the reason why there are so many 7 port hubs (a weird number in computing, generally) is because they have two hub controllers ganged together inside. I've pulled enough cheap USB hubs apart (that wouldn't work with my FPGA setup) to find that this is the norm, and even the more expensive Belkin kit does the same thing, and uses the same controllers.

Most support a four-port hub system, but hubs using 16-port hub controllers are also available in the industry. although some 16-port hub actually consist of four cascaded four-port controllers.

Holy shit, now we know why BFL hasn't be delivering to everyone else.

I count 42, all on stock bricks (?!?!?!) - protip: Almost any standard power supply will be more efficient and require less cable clutter.

I'd seriously consider moving to a bigger PSU that will support all of the units when you get them.  The stock power supplies are HOT.  Cablez on the forum makes custom PCIe to BFL cables that work great.  I am using a Corsair AX1200 that ran 10 BFLs just fine (I think it can handle up to 12).  It's more efficient as well, I saw my total power usage drop to about 75 watts per unit, as opposed to 85 each using the stock PSUs.

With one it does not seem so but perhaps when I put 10 more on that shelf

My first single is up I am waiting for 10 -12 more from various sources

I got the rack from Sam's Club for $49 and the Custom Cooling Fan on the left from Walmart for $12

I used the new auto tuner from BFL and it seems to be running well

That single looks pretty lonely... What a difference from the GPU rigs stretched out on coat-hangers and USPS boxes.

Sorry I couldn't see the vent from here. A vent definitely will make the air flow better, especially when A/C is running. 
I haven't used a FPGA -- does it really need a box fan at all?

It works quite well there is an AC vent right above that blows down the wall. The single has averaged 853mh/s without one instance if throttling

Being so close to the wall, the airflow of the fan is greatly reduced. IMO.

My first single is up I am waiting for 10 -12 more from various sources

I got the rack from Sam's Club for $49 and the Custom Cooling Fan on the left from Walmart for $12

I used the new auto tuner from BFL and it seems to be running well

My first single is up I am waiting for 10 -12 more from various sources

I got the rack from Sam's Club for $49 and the Custom Cooling Fan on the left from Walmart for $12

I used the new auto tuner from BFL and it seems to be running well