Pages:
Author

Topic: [ANN] Bitfury is looking for alpha-testers of first chips! FREE MONEY HERE! - page 5. (Read 176664 times)

KNK
hero member
Activity: 692
Merit: 502
Very nice. It sounds like you're testing each board and moving chips around/replacing them based on their performance and/or whether they're dead or not. Have you considered any methods to remove a dead chip from the chain without actually having to rework the board?
Well ... not exactly. This is the prototype, so it was designed to make tests and moving things around easier. The problem was that i had lots of bad chips (https://bitcointalksearch.org/topic/m.3538767 ) - from 120 purchased the first two boards didn't start at all, so had to replace some chips and the end result is 111 of those 120 working or over 8% waste.

I have considered designing the new board in a way that will allow skipping a dead chip (by changing the resistor diver on SPI as stated earlier), but that makes the board huge and no longer single layer board (with only chip grounds on the bottom), so the next decision was to make shorter strings of 7 chips and be able to skip the whole link until it's fixed. It is not possible to skip a chip with a simple jumper as in the BFSB boards, because here there are different voltage levels at each chip.

As you can see from the pictures above most chips are at 54 clock, but some are at 55 - that's how they are balanced in performance, not by replacing the chips or moving them around.
legendary
Activity: 1274
Merit: 1004
Have you actually played around with this?

Quite a lot. I have a board from 14 chips running from 12V and they can be chained for higher voltage. My initial idea was to make a string for 220V AC directly, but that won't be easy and cheep enough to do properly and in addition it's dangerous, so i gave up on that.

Here is an old setup from November at 12V
And this is a fresh one from few minutes ago at 24V

The bold red lines are missing/nonworking chips - they should be at the end of the chain and replaced with diodes if missing. It is possible to 'skip' a single non working chip in the middle of the chain by changing the SPI resistor dividers, but it's easier to replace or move them at the end.

Most of the HW errors are actually software errors (chip starts working on the old job because the new one has not been sent yet), so i am working on a new driver based on the cgminer's BAB driver, but still have some problems to understand the internal work queue API and workflow and the lack of time lately is adding to the delay Sad

EDIT: and the passive cooled miner itself No fans, no noise. 'Smart aluminium heater' and the back can be populated too Smiley ... if there's chips Sad
Very nice. It sounds like you're testing each board and moving chips around/replacing them based on their performance and/or whether they're dead or not. Have you considered any methods to remove a dead chip from the chain without actually having to rework the board?
KNK
hero member
Activity: 692
Merit: 502
Have you actually played around with this?

Quite a lot. I have a board from 14 chips running from 12V and they can be chained for higher voltage. My initial idea was to make a string for 220V AC directly, but that won't be easy and cheep enough to do properly and in addition it's dangerous, so i gave up on that.

Here is an old setup from November at 12V
And this is a fresh one from few minutes ago at 24V

The bold red lines are missing/nonworking chips - they should be at the end of the chain and replaced with diodes if missing. It is possible to 'skip' a single non working chip in the middle of the chain by changing the SPI resistor dividers, but it's easier to replace or move them at the end.

Most of the HW errors are actually software errors (chip starts working on the old job because the new one has not been sent yet), so i am working on a new driver based on the cgminer's BAB driver, but still have some problems to understand the internal work queue API and workflow and the lack of time lately is adding to the delay Sad

EDIT: and the passive cooled miner itself No fans, no noise. 'Smart aluminium heater' and the back can be populated too Smiley ... if there's chips Sad
legendary
Activity: 1274
Merit: 1004
I'll throw my hat in the ring to test these if needed.
No problem! Send me one reel of chips and I'll send you back 4 chains of 98 chips for testing ... you will need a PSU which gives 80V DC and 10A
From that message, it sounds like you're talking not only about chaining the SPI bus, but also running strings of Vdd similar to what Metabank is doing with their new cards.

Have you actually played around with this? I have a few BF chips still sitting around that I wouldn't mind playing with, and I'm curious how you'd handle fault tolerance and load balancing in such a setup.
legendary
Activity: 4466
Merit: 1798
Linux since 1997 RedHat 4
Can antone tell me how many bitfury's can be chained?
Ive seen the proven 16 bitfury design. metabank has multiple chains.
Nowhere is listed how many you can put on one chain.
No there is no such information and there are no limits from the chips themselves, but ...

There are 2 types of limiting factors involved:
 [1] RPi (or other device) processing power, which applies to both single and multiple chains, but also on software.
 [2] each chip in the chain adds some delay, which results in lower SPI speed required in order to successfully communicate with the last chip in the chain, then at some point your SPI communication will be too slow to walk all the chips in the chain in time before they need new work.

BSB has 4 boards of 16 chips each in a chain, so 64 chips are proven to work. I have read somewhere about 100 chips working in a chain and about 400, but not sure if in a single chain or just on the same RPi.
Going for more than 150-200 in a chain is unreasonable because of [2] and more than 350-450 total because of [1] which is close to those 100 and 400 numbers above

The V2 controller from BlackArrow, with the BlackArrow Fury boards, supports up to 384 chips
The V2 controller has 4 SPI 10pin connectors that allows 6 boards each SPI chain - so 24 boards.
... yes that's running on an RPi
From a software POV it's 4 banks of 6 boards i.e. the same thing as the V2 bfsb code.
KNK
hero member
Activity: 692
Merit: 502
I'll throw my hat in the ring to test these if needed.
No problem! Send me one reel of chips and I'll send you back 4 chains of 98 chips for testing ... you will need a PSU which gives 80V DC and 10A
hero member
Activity: 926
Merit: 1001
weaving spiders come not here
I'll throw my hat in the ring to test these if needed.
KNK
hero member
Activity: 692
Merit: 502
Can antone tell me how many bitfury's can be chained?
Ive seen the proven 16 bitfury design. metabank has multiple chains.
Nowhere is listed how many you can put on one chain.
No there is no such information and there are no limits from the chips themselves, but ...

There are 2 types of limiting factors involved:
 [1] RPi (or other device) processing power, which applies to both single and multiple chains, but also on software.
 [2] each chip in the chain adds some delay, which results in lower SPI speed required in order to successfully communicate with the last chip in the chain, then at some point your SPI communication will be too slow to walk all the chips in the chain in time before they need new work.

BSB has 4 boards of 16 chips each in a chain, so 64 chips are proven to work. I have read somewhere about 100 chips working in a chain and about 400, but not sure if in a single chain or just on the same RPi.
Going for more than 150-200 in a chain is unreasonable because of [2] and more than 350-450 total because of [1] which is close to those 100 and 400 numbers above
sr. member
Activity: 251
Merit: 250
Can antone tell me how many bitfury's can be chained?
Ive seen the proven 16 bitfury design. metabank has multiple chains.
Nowhere is listed how many you can put on one chain.
There's no hard limit, but more chips increases the chance of a single bad chip blocking the communication. Also, the longer the chain, the slower the communication. First generation M/H board design has 256 chips in a single chain, 2nd generation has 4x64.
full member
Activity: 141
Merit: 102
Can antone tell me how many bitfury's can be chained?
Ive seen the proven 16 bitfury design. metabank has multiple chains.
Nowhere is listed how many you can put on one chain.
KNK
hero member
Activity: 692
Merit: 502
Ideally - I really hope in the next generation of bitfury's chips to see a built-in thermometer or at least to add a diode on the die that can be accessed externally and be used for temperature measurements.
+1
As for the ideas - a weird one ...
Temp sensor (edit: thermistor) of few kilo ohms, a small capacitor, triggered from the OE signal of the board and measuring the charge (or discharge) time on a dedicated pin ...
advantages: no ADC, no expensive chips, small and cheap, easy to place anywhere
disadvantage: less accurate, but enough for detecting overheating
vs3
hero member
Activity: 622
Merit: 500
I was actually looking at adding one.

Ideally I wanted to add a SPI one so that I can put it on the same bus as the chips, and that way it could physically be located right next to the last chip in the chain. This would be ideal as the output SPI signals are right next to the 1.8V and that way the thermometer chip can be literally millimeters from the bitfury chip.
Except that I can't find ones that work at 1.8V - all of the SPI ones I've found need at least 2.5V or 3.3V.

There are some that work at 1.8V but they're I2C but I can't put them on the same SPI bus (as some of the signaling will interfere with the RESET sequence).

The alternative is to run extra wiring just for the temp. sensor alone, but that's a whole another can of worms.

And not to mention those chip-sensors are ridiculously expensive.

So, on my to-do list I've added to use a cheap AVR or PIC that works at 1.8V (those in the $0.4-0.5 range will do) and just code my own thermal sensor and make it just enough SPI-compatible so that it can be added on the chain of bitfury chips.

Or if anyone else has other ideas - please share...

A low-cost NTC and a small MCU running Steinhart-Hart would do the trick:)

intron


yup - exactly what I had in mind.

There are going to be some interesting design challenges (like "how accurate is that AVR's ADC at different temperatures") but nothing that can't be relatively easily solved.

Ideally - I really hope in the next generation of bitfury's chips to see a built-in thermometer or at least to add a diode on the die that can be accessed externally and be used for temperature measurements.
sr. member
Activity: 427
Merit: 251
- electronics design|embedded software|verilog -
I was actually looking at adding one.

Ideally I wanted to add a SPI one so that I can put it on the same bus as the chips, and that way it could physically be located right next to the last chip in the chain. This would be ideal as the output SPI signals are right next to the 1.8V and that way the thermometer chip can be literally millimeters from the bitfury chip.
Except that I can't find ones that work at 1.8V - all of the SPI ones I've found need at least 2.5V or 3.3V.

There are some that work at 1.8V but they're I2C but I can't put them on the same SPI bus (as some of the signaling will interfere with the RESET sequence).

The alternative is to run extra wiring just for the temp. sensor alone, but that's a whole another can of worms.

And not to mention those chip-sensors are ridiculously expensive.

So, on my to-do list I've added to use a cheap AVR or PIC that works at 1.8V (those in the $0.4-0.5 range will do) and just code my own thermal sensor and make it just enough SPI-compatible so that it can be added on the chain of bitfury chips.

Or if anyone else has other ideas - please share...

A low-cost NTC and a small MCU running Steinhart-Hart would do the trick:)

intron
vs3
hero member
Activity: 622
Merit: 500
I was actually looking at adding one.

Ideally I wanted to add a SPI one so that I can put it on the same bus as the chips, and that way it could physically be located right next to the last chip in the chain. This would be ideal as the output SPI signals are right next to the 1.8V and that way the thermometer chip can be literally millimeters from the bitfury chip.
Except that I can't find ones that work at 1.8V - all of the SPI ones I've found need at least 2.5V or 3.3V.

There are some that work at 1.8V but they're I2C but I can't put them on the same SPI bus (as some of the signaling will interfere with the RESET sequence).

The alternative is to run extra wiring just for the temp. sensor alone, but that's a whole another can of worms.

And not to mention those chip-sensors are ridiculously expensive.

So, on my to-do list I've added to use a cheap AVR or PIC that works at 1.8V (those in the $0.4-0.5 range will do) and just code my own thermal sensor and make it just enough SPI-compatible so that it can be added on the chain of bitfury chips.

Or if anyone else has other ideas - please share...
newbie
Activity: 56
Merit: 0
Hello
May I have samples chip?
I'm developing bitfur hardware mining
Thanks
KNK
hero member
Activity: 692
Merit: 502
Hint for those who can't think of the obvious ... put one on the board somewhere at least ...

It's pretty easy for BF using RPi to add a separate I2C sensor on the board 'somewhere', but with several hashboards .. on which one? There are 16 boards on a full rig and only 3 bits (8 addresses) for the sensors.
One, somewhere? Then use the RPi internal temp from /sys/class/thermal/thermal_zone0/temp - it will give you 'some' temp to work with (in my case more than enough as it's on the same heatsink, but not for other designs)

I had 8C difference between chips on the same board (because of loose thermal connection on some of them), so even per board sensor may not be enough - You need per chip info to be accurate in your decisions.

Many people think they got better cooling with BFL hardware, because the sensor is close (but separated from) the heatsink - a fan blowing down shows much less heat, while the truth is a bit different, so placing 'something' 'somewhere', just 'to have one' is a bad hint IMHO. I definitely prefer no info than a false info
legendary
Activity: 4466
Merit: 1798
Linux since 1997 RedHat 4
... any software accessible temperature sensors ......... ?

None built-in as far as I know. At least in Gen-1 chips. Hopefully in the next generation such thing is added ... or at the very least just a diode on the die that can accessed externally and could easily be turned into a thermometer (sort of, not the most accurate but good enough for our purposes).

My only guess as to why there isn't one in its current incarnation - those chips just LOVE heat! From what I've seen - all other chips will stop working, burn, melt, explode, fall apart, etc., you name it, before bitfury's stop working Smiley

...
I guess it's winter over there.
Summer here, and hot days I've had my BA board HW suddenly hit 100% for no reason at all ... and no temp sensors.
Restart it (when it's had a minute to cool down) and it's all happy again.

Going on more than 2 years since I brought up the issue about having temperature sensors somewhere in mining hardware and people still make the same mistakes ... sigh. They always know better ...

Hint for those who can't think of the obvious ... put one on the board somewhere at least ...
KNK
hero member
Activity: 692
Merit: 502
So if this works then what's the point of having the hardware reset at all? I wonder if we're missing something here...

For a single chip you don't need reset, but for a chain you do ... or it will be difficult to address chip 5 when you don't know the state of the previous chips ... does chip 3 forward or not? if it does you skip one fasync, but if it doesn't you need to send fasync
vs3
hero member
Activity: 622
Merit: 500
Quote
few NOPs and a brake should do same

That's what the brake will do right?

That's an interesting idea. I haven't tried it (as all sample code used the hardware reset) but if it works it should permit even simpler designs!

So if this works then what's the point of having the hardware reset at all? I wonder if we're missing something here...

Anyways, I'll take a look at the code examples that I've found and see if there are any details why should or shouldn't the reset stuff be done one way or the other.
KNK
hero member
Activity: 692
Merit: 502
there is no other way to get it out of it

Quote
few NOPs and a brake should do same

That's what the brake will do right?
Pages:
Jump to: