Bitcoin Forum

Hi there,

I'm in the process to create a dedicated FPGA Rig. So my question is, how many units do you control succesfull over USB on a single system?

I know, there is be a hard limit of 127 per system. But I'm sure I dont will hit that mark. I'm thinking more about 20, 40 or maybee 60 Units per Windows 7 system.

I know, many USB devices can slow down your system. It is better to choose an i3/i5 CPU with 8 GB memory or is an cheap K6 succesfull as well ?

Furthermore, i prefer CGMiner as mining software. Can it handle 20, 40 or 60 devices or must I open several instances ? Does several instances affect the overall system performance ? I think, the little windows of CGMiner can not show more as about 10 hashing units at once ?

Does anybody have already expirience with this ? Should be, as I'm pretty late with it....but better late as never :D

Thank you for your advise,
Luke

If this is such a technical feat, just buy 3x£200 laptops.

https://bitcointalk.org/index.php?topic=182892.0

People have already been using >40 USB miners on one computer.

AFAIK, The USB spec can only handle 127 devices per controller. You can always add a new controller (http://www.newegg.com/Product/Product.aspx?Item=N82E16815124118).

just use a USB hub with its own power supply and u wont have a problem

That is 100% correct. It's also important to note that a lot of motherboards ship with more than 1 USB controller. The one I'm using to type this has 4 distinct controllers.

Split it out as much as possible on separate controllers.

Your throughput will be better.

USB *may* be able to handle 127 devices... but it can only talk to one at a time

By splitting it out to 4 controllers, it can speak to 4 AT A TIME.

If one goes defective then you only loose 25% of your operation... not 100%

Note when I say CONTROLLER I mean HOST controller with a direct distinct & individual data-path back to the CPU.

Anyone had any luck on running lots of units on USB 3.0 vs 2.0? I wonder if a native implementation of USB 3.0 in a miner would reduce latency (http://stackoverflow.com/questions/13831008/what-is-the-minimum-latency-of-usb-3-0)? We don't really need the bandwidth, IIRC, but lower latencies might help in a system with 100+ 50GH/s USB miners.

Thank you for your Responses.

I changed now the design to additional 2 USB Controllers, so with the Onboard 3 in total. And of course I'm using only powered USB Hubs.

Will Report again, when I'm up and running, but that should take about an month from now, until I recieve the last FPGA's.

Cheers,
Luke

But the controller in the USB-Hubs also count as a "device" some of the bigger ones have two controllers in it.

That is NOT a controller......
It is not capable of initiating a transfer on its own, as such at the most it is a  PERIPHERAL DEVICE.
And as such it is  subtracted from the total count of 127.
However for practicable reasons I have yet to see a 'fully loaded'  USB system.

A guy I used to hang out with did do a full on 127 device chain, back when USB1.1 first came out, and it worked. Bogged the system something fierce, though. With the faster throughput on 2.0 and 3.0, I think it would probably be better. However, it's probably safer and EASIER to just keep the numbers down and use multiple hosts. A usb controller isn't exactly expensive. Even this cheap ass acer I'm using has two of 'em.

Ideally you get something like a  PCIe card with several controllers (NOT USB ports), then build the chain off that.
For SBC's there is a big issue in that you need data-sheets and a circuit diagram to see how it is implemented, I think on the first revisions of the PI  it is badly done.
One absolute killer is where they use  a bloody chip for the Ethernet with extra USB ports on, some of these chips do not even have a TCP/IP buffer!!!  so once you get one of these shitty little cutters on your USB chain, you have a chip with multiple  Ethernet ports open, with no buffers... sucking the bandwidth out of your USB chain.

Even worse are the 'pack of lies' controllers which say USB2.0 compatible, but when you pull the data-sheet they are actually USB1.0 , but they are classed as 'compatible' because the 2.0 infrastructure is capable of communicating with the 1.0 device!!!

Only issue is... if you read the USB spec... a USB chain with a 1.0 device on it, defaults the WHOLE chain to 1.0 spec speeds.... (yep they do have some 'hubs' with translators, but  it depends on the chipsets used.)

It is like a massive Microsoft nightmare, where even accomplishing something a simple as plugging a couple of devices into your computer becomes a major research task.

All on Intel Nuc Core i3

https://i.imgur.com/4elG6BO.jpg (http://imgur.com/4elG6BO)

Just get a bunch of controllers, slap on a current gen i5 or i7, 16-32gb of ram all for piece of mind, and add hundreds of those things.  

The mining device latencies themselves are many orders of magnitude larger than the usb latencies so going to usb3 would achieve nothing. It will not be a rate limiting thing even with 1000 devices.

This has been answered SOOOOOO many times before on these forums.

The issue is point of failure. and point to point communication.

Awesome, thanks!

Thanks for posting this.  I'm running 99 of them right now with 10 arctic fans on 11 Anker hubs, but HW error rates have exceeded 10%.  I wasn't sure if I was just hallucinating about the HW errors, but seeing your screen shot confirms what it should look like.

Will have to find a way to curb these "nonce" HW errors.

icarus-timing short is my config with 200 Mhz firmware. Today go trying v 42 with new heatsinks and 92 mm fan. Without, very much HW


https://i.imgur.com/5nOu6RR.jpg (http://imgur.com/5nOu6RR)
https://i.imgur.com/jTDE6r0.jpg (http://imgur.com/jTDE6r0)

Hi,
Be careful with your internet line...
Although it is "theoretically"possible to put more then 127 devices/port , you have to know and understand that in electronics theory is not = to practice .
Stay low with ~100 devices/computer   ( intel atom board + gb of ram + 16gb flash drive) and you should be ok  - USB hubs will also "eat" some of the 127 devices...

I would like to nail a common miss-conception here about processors/memory/speed.

I have had 54 units on 3 hubs running on a p4 2.8Ghz (skt 775) with 2x512 MB memory (DDR2) on PCLINUXOS with a 20GB IDE drive.

All is sweet, CPU hangs around 2-3% usage, I have no idea why people are throwing money at "I" core systems with vasts amount of memory and HDD space.

If it a stand alone miner and you leave it as such, You can use dated hardware and save $$$

That is OSX.. lets see the 'activity monitor' and a double click on the "CPU usage"

I'm thinking I can get atleast 100 on a machine easily.  I just need a usb hub that supports that.  There are a few out there, but it's pricy.  better to buy some smaller and just daisy chain it.

Really?? The screenshot is of OSX, but he's blatantly using TeamViewer to remote into a Windows8 machine that's running 8 instances of CGMiner 3.1.1 to run 76 devices. And that Intel NUC is a 4"x4" little box with a dual-core 1.8GHz CPU. Not exactly beefy.

I know what you mean, my 'rig' has a single 4GB stick of DDR3, and a old i3 530 with one core disable and the other down-clocked to 1.3GHz and 1.05v! Doesn't even need a fan on it and stays around 50*C!

Sorry I imbued him with a little too much intelligence....

Seems you can get osx running on that intel stack.... with a few hacks, I initially thought he had gone that route.


Anyway .... whilst we are on the number of USB devices

Spotted another serious flaw with trying to 'tree up' a USB port to its maximum.. (actually it was my cat that found it)
A defective plug on a USB cable, allowed the cable to operate out of spec and in doing so took down a section of my miner network
(fortunately i'm on multiple controllers)... unfortunately it also burned out an ATX PSU

So maybe that is an argument for multiple cheap computers....