GekkoScience BM1384 Project Development Discussion

[hurricandave]

Why, when USB1.1 has enough bandwidth? The only benefit to USB3 is increased power available, which admittedly is handy, but a solution to that is making the miner adjustable so you use whatever power you can.

Just bought three new mobo's and each of them only has two USB 2.0 ports. Type-C is coming and the vast majority of common USB devices are on the 3.0 already.

[sidehack]

Is 3.0 still backward-compatible?

[hurricandave]

Is 3.0 still backward-compatible?

I question that all the time. How many times have you read comments in this forum where, a user says 'I just got this new yada yada miner and can't get it to work'. And how many times does the initial responses come back with.... Make sure you are plugging it into a USB 2.0 port.

[sidehack]

Well unless the device end is backward-compatible with USB2.0, the same problem would happen if we build 3.0 miners and people used 2.0 ports. All the good crappy hubs folks want to buy are 2.0, and so are a lot of the decent ones. If 3.0 ports aren't backward-compatible with 2.0 is fairly stupid, as they'd no longer merit the "U" in their own name.

For an approximation of my thoughts on that "the vast majority of common USB devices are on the 3.0 already", consider that I own exactly one motherboard with any 3.0 ports, (one pair), and I own zero 3.0 devices.

[2112]

Is 3.0 still backward-compatible?

I question that all the time. How many times have you read comments in this forum where, a user says 'I just got this new yada yada miner and can't get it to work'. And how many times does the initial responses come back with.... Make sure you are plugging it into a USB 2.0 port.

It is all because lack of the real USB3.x bridge chips that interface serial/SPI/I2C.

All SiLabs bridge chips are USB 2.0 at most.

All useable FTDI bridge chips are also USB 2.0.

FTDI publicly has only two USB 3.0 chips (FT60{0,1}) that serves only parallel FIFOs (16&32-bit wide).

Do you know of any useable native USB 3.0 chips serving serial/SPI/I2C ? Not an entire ARM-based controller like Cypress FX3/FX3S that will require extensive programming?

[sidehack]

Admittedly the interfacing is more Novak's department, but I don't know of any USB3 chips offhand.

[PlanetCrypto]

I prefer SPI versus UART. Just cuz' that's what I worked with the most.

As far as I know there are no USB 3.0 to SPI/I2C chips.
Most USB 3.0 stuff is targeted for the storage world. Hence the need for speed and are implemented to interface with other fast interface standards (SATA, PCIe, etc.). SPI/I2C/UART aren't, to the best of my knowledge, multi-gigabit protocols. Hence interfacing something that runs @ Mbps to something that runs @ GBps poses some challenges.

As far as I know USB 3.0 implementations are backwards compatible with 2.0 from a data/signaling standpoint.

But the point is why spend the expense for 3.0 when 2.0 has all the bandwidth required (for hashing) and it's butt loads cheaper. (redundant?)
Although, moving to 3.0 adds a contributing reason to put an MCU on a hash board, cuz' there are many MCU choices that have SPI/I2C/USB 3.0 as interfaces (some have ethernet too).

If I were doing a board, I'd use SPI onboard and USB (whatever) for off board comms. If that meant an MCU (due to a decision to implement USB 3.0 compatibility) then the more the merrier.
But it'd be a "smart" hash board (cuz' ya can w/ an MCU) and likely wouldn't be something inexpensive for the "home" miner.

In my mind, affordable = USB 2.0 and versatile/sophisticated/more expensive = USB 3.0

"Speed costs money, how fast can you afford to go."  -- Stroker McGurk

I have 2 machines with USB 3.0 add-in boards and rarely (if ever) use the ports.

BTW, The Synopsys IP SHA-256 cells come in 32 bit and 64 bit flavors.

Gotta' eat, be back l8r.

[PlanetCrypto]

"I was thinking about ethernet working around collisions in the mid 70's but that's because I don't know WWII radio relay history. Nice tidbit of knowledge I'll have to look up now."
I know this cuz' I'm an old fart who did networking with smoke signals. 

Pretty sure it was Bell Labs and not HP or DEC cuz' they each had their own protocols (ARCNET and DECNET).
IBM of course had "choking ring" (token ring).

[AJRGale]

I wonder how hard it'd be to integrate level shifting for string comms?...

didn't newrbox do this this? ..wait, or was it just multiplexing? dang it, i don't have one in front of me, i could look up that silly chip..
(there is a little dual lead flat pack just down and to the side of the controller, next to it is a few resistors in a little network on a mount for another one of them chips. there is 2 of these, one set is also below the fan headers).

[hurricandave]

I guess part of what I felt was possible with USB 3.0 is the potential flexability from the secondary TXRX channels being meant as power control and sync of the slave/host relationship while on the fly. Real time adjustability at chip freq and voltage. But after digging in even TI, USB controller still would require use of FIFO and an ARM chip (AM355X to an A8 or such). 

[sidehack]

Realtime adjustment of frequency and voltage should still be possible if the code is done even halfway decently.

New R-box has no chained comms. I think the Prisma is the only BE200 string, and it had external level shifters for comms - on the Rev2 they were ASICMiner's proprietary PMS01 chip that integrated node-level current regulation and SPI level shifting into a single device. When I said level-shifting, I meant level-shifting to a different ground potential, which is required for string devices. On the S5, that's what the little set of resistors and diodes adjacent to every other chip does. The diodes are forward-biased into the data output and the next node up's data input is taken from the top of the diode, which mirrors the bottom node's data output plus forward drop (generically 0.7V), which steps it up to a valid range since the next node's ground potential will be about 0.8V above. I used mostly the same setup on my two-chip string test boards. With parallel comms, you don't have the advantage of going from one node directly to the next (so only shifting up 0.6-0.8V at a time); you get one IO line and you have to bump it all the way up. I mean it's possible to do it piecewise but then you have to worry about cascaded propagation delays and such. Better to have an independent shifter at each node all acting in synchrony (if, like SPI, it's a synchronous data transfer protocol).

[PlanetCrypto]

Maybe not the right place to ask this question so am apologizing to Gekko et. al. in advance for posting here.

What would the community think about a wired ethernet stickminer?

i.e. exactly the same thing as a USB stickminer but interfaces via RJ45/48 male plug w/ a 6 pin PCIe power plug on the opposite end.
Could have a minimal web interface to access parameters (CoreV & Clock), miner status, pools, temps, and w/ port 4028 api access.
Thinking it would be a low-count hash-chip stick (2-4) made long and skinny.
DHCP or static IPv4.
Run an abbreviated Linux distro.

Used 12,16,24,48 port managed/unmanaged hubs/switches are cheap ($50-$200).
So on the low end a guy could plug 1 into a port on his/her router, on the high end 48 plugged into a 48 port switch and anything in between.

[sidehack]

One problem with that is cost. If you're integrating a full Linux controller for only a few ASICs, your support circuitry will end up costing substantially more than the mining chips themselves. It's also less efficient, as you're putting the overhead of a controller on each stick - so if you ran 48 sticks you'd have 48 CPUs instead of one (or a few).

Another problem is mechanical - what kind of torque are you seeing imposed on the ethernet jack by the weight of the device and its associated cabling?

[Mikestang]

But the point is why spend the expense for 3.0 when 2.0 has all the bandwidth required (for hashing) and it's butt loads cheaper. (redundant?)

Is there any advantage (related to stick miners) to the slightly higher voltage a 3.0 port puts out vs a 2.0 port?

[philipma1957]

One problem with that is cost. If you're integrating a full Linux controller for only a few ASICs, your support circuitry will end up costing substantially more than the mining chips themselves. It's also less efficient, as you're putting the overhead of a controller on each stick - so if you ran 48 sticks you'd have 48 CPUs instead of one (or a few).

Another problem is mechanical - what kind of torque are you seeing imposed on the ethernet jack by the weight of the device and its associated cabling?

mechanical   a spice rack can solve.   or some mounting setup.  


Where does the power for the stick come from?  Power over ethernet is not so cheap.


edit:
missed the pcie jack for power.

[philipma1957]

But the point is why spend the expense for 3.0 when 2.0 has all the bandwidth required (for hashing) and it's butt loads cheaper. (redundant?)

Is there any advantage (related to stick miners) to the slightly higher voltage a 3.0 port puts out vs a 2.0 port?

yeah  in theory:

   the USB 2 will give only .5 or .6 amps at 5 volts

  the USB 3 gives .9 amps

These power numbers vary greatly hub to hub.

These sticks will need very good hubs to do high amps for freq 250 and up.

My test hub is bridging 2 .9 amp usb2 ports and can supply 1.8 amps to the bridge. 

[novak@gekkoscience]

Is 3.0 still backward-compatible?

I question that all the time. How many times have you read comments in this forum where, a user says 'I just got this new yada yada miner and can't get it to work'. And how many times does the initial responses come back with.... Make sure you are plugging it into a USB 2.0 port.

It is all because lack of the real USB3.x bridge chips that interface serial/SPI/I2C.

All SiLabs bridge chips are USB 2.0 at most.

All useable FTDI bridge chips are also USB 2.0.

FTDI publicly has only two USB 3.0 chips (FT60{0,1}) that serves only parallel FIFOs (16&32-bit wide).

Do you know of any useable native USB 3.0 chips serving serial/SPI/I2C ? Not an entire ARM-based controller like Cypress FX3/FX3S that will require extensive programming?

The reason that there are no USB 3.0 serial chips is that USB 2.0 is 480 Mb/s.  You can't run a serial protocol (probably, at least reliably) over a few tens of megabaud, and it's doubtful that you want to run it that fast.  You certainly don't need that sort of speed for a bitcoin miner.  USB 3.0 is still reverse compatible with USB 2.0, so just use a 2.0 chip.  There's no advantage to having a USB 3.0 chip at all, although a port with extra power is handy.

--
novak

[Mudbankkeith]

Plenty of 49 port usb2 "block erupter" hubs about.
I believe they can host up to 1.5 amps per port.

EDIT:- with a cheap atx psu then probably no more than 150w    approx 30amp  ...   so that gives about 20 sockets@1.5amp

[novak@gekkoscience]

Plenty of 49 port usb2 "block erupter" hubs about.
I believe they can host up to 1.5 amps per port.

If you have a power supply over about 370W then yes, otherwise you can't fill every port.  Typically the actual USB power jacks are only rated for about 1.5A so that becomes something of a practical upper bound- though I haven't seen one fail.

[sidehack]

Add that to the cheapo 40A fuses those came with, and the fuse sockets that I don't completely trust to 40A let alone 49*1.5= 73.5A load without bursting into flames (a fairly common sight on those hubs).