Block Erupter USB - Overclocking/ hacking ?

[Bluestreak66]

I don't believe it's the missing color layer that improved your thermal conductivity, i believe it's more from not using the original screws.
The layer would be ~30-50µm thick, the 0,5mm silicone layer has around 0,5-1 k/W, a layer ten times thinner could not make that much difference even if it is simple lacquer, don't you think?
Also, it has perfect fit on the pcb side, because it was liquid when applied, and the other side has less bumps than your now blank copper side, i think.

Mine look like they bend from the pressure of the silicone layer, increasing resistance. A Screw in the middle would help

Anyone knows if epoxy or CA will withstand the temps?

I had the exact same setup as above on both BLock Erupters with the solder mask on. I can tell you there is a massive difference in temp. I only did this to one so I can compare them. They both have the same heatsink from intel motherboards. The bare copper transfers heat much more efficiently. The solder mask also has a cross hatched pastern on it so it effectively insulates the pcb by not allowing direct contact. I noticed this while sanding it off. I'm going to desolder the usb plug off the board and use the data lines from a usb cable then power the board from a external supply. I may bolt this to a water block and see how far I can crank it up. I may even pull off the core voltage supply regulator and substitute my own off- board. One thing I would like to know is where does the 3.3 volt supply come from on the board? Measuring the voltages on the board I get 4.73 on the usb and header pins, 3.4 on the vcc and 1.07 on the regulator output.

Edit: On further inspection it appears the silabs chip has and internal regular the outputs 3.3 externally.

Another thing to keep in mind is that the original heatspreader is attached with a silicone thermal pad. Those are not really known for their amazing heat conductance. Of course they are relatively good compared to say, air, but they are are pitiful compared to silver and alumina based thin layers of thermal paste or compound.

Also instead of modifying the USB plug that is soldered on, it might be easier to get a spare socket or the socket from a male-female USB extender cable.

There has to be some form of electrical insulation between the bare copper and the heatsink otherwise it will short vcc to ground, any type of paste can allow this to happen. I think the silicone sheet is good for now. I would like to note to those putting heatsinks on the top of the chip that it is almost useless. That will not help the core temp at all your just cooling the latent heat radiating up from the board and base of the chip the thermal conductivity is to low. It would be like trying to cool a cpu from the bottom, sure the heatsink will get hot but the core temps will be scorching, its not designed to be cooled like that. The die is fliped in these chips and basically insulated on top by the plastic. With the setup I posted above the top of the chip is barely warm but the heatsink it quite hot without a fan, before it was the opposite way round.

[1715025023]

1715025023

[Trillium]

So I removed the solder mask from my two units today.

Before:

• 

Like Bluestreak66 I first used sandpaper and got about half the mask off

It was pretty slow and tedious, even using different sandpapers. Part of the problem is that its hard to hold or support the device when it's upside down... I didn't want to damage any components while working on it.

I found that my Dremel kit had a rough polishing tip that worked really well. It was MUCH faster and since I could easily hold the board while working on it there was less risk of damaging it. It only took me about 2 minutes to do the whole second miner using this tool.

• 

The heatsink I chose is from a server, it has a copper heatspreading plate bonded to aluminium fins, for 90 W TDP Xeons. Slight overkill...

• 

To conduct heat from the copper on the PCB to the copper heatsink I decided to use Arctic Silver's Céramique 2 (oxide based) on one and Arctic Silver 5 (silver based) on the other. As I pointed out in another thread yesterday, risk of conduction through the paste is non-existent, however the warning about capacitive effects remains fair enough.

Undeterred, I applied the pastes on the back of each PCB and pressed them against the copper heatsink plate. Connected to USB, I noticed that the one with Arctic Silver 5 acted strangely as it was slightly moved on the copper underneath, it was malfunctioning and looked like it was restarting itself. Cgminer reported that it had become a faulty avalon device. I assume these are the kinds of capacitive effects mentioned earlier. The one with Ceramique was fine though, and it was actually better since it has better adhesion, so I removed the AS5 with absolute ethanol in a agitation bath (different to a ultrasonic bath), and replaced it with Ceramique. In a perfect world, I'd have some double sided tape and would have used that... but I was out.

So instead, I put down a piece of packing tape across the heatsink block, avoiding bubbles underneath, to use as a insulator 'just in case'. The tape is only ~0.01-0.02 mm thick, much better than the original silicone sheet which is ~0.32 mm. Because heat conduction rate is

Code:

Rate = k*A*[(T1 - T2)/d] where k = a constant, A = surface area, T = temps, d=distance

clearly if d is smaller by roughly one order of magnitude then heat transfer will be improved the same, assuming all other things are kept constant, and that the silicone and the tape+compound are roughly the same thermal conductivity.

Also following the rubber band trend, I put them next to a server vane fan running low volts so it's silent. It lives on the top shelf of my computer desk now, away from harm.

• 

Fan cooled, stock heatspreader, 22 deg C ambient temp: Maximum ASIC topside temp = 46 deg C
Fan cooled, custom heatsink, 22 deg C ambient temp: Maximum ASIC topside temp = 34 deg C  NOT BAD!

The whole copper block sits 1.5 deg C above ambient while mining and the air going through increases 0.7 deg C.

I have some very small heatsinks on the topside of the ASIC though I agree it is more or less ineffectual and cooling from the base through the PCB is the way to go. I will get some 2 sided tape soon and replace the Ceramique+tape I have used above, I expect even lower temps.

Definitely recommend doing something like this if you guys plan to try overclocking it.

[amigaman]

Okay, if your tests say so...
how to mount them professionally on a aluminium heatsink?
Rubber bands will not work for longer amounts of time...

so my idea was to glue one or two glimmer pads for TO247 or the like to the board, thinner and isolating, as a replacement for the silicone pad.
then glue the sink to the glimmer.
or use the screw holes and 2 glimmer plates and lots of arctic silver, or will the silicone pad suffice?

[Trillium]

Okay, if your tests say so...
how to mount them professionally on a aluminium heatsink?
Rubber bands will not work for longer amounts of time...

so my idea was to glue one or two glimmer pads for TO247 or the like to the board, thinner and isolating, as a replacement for the silicone pad.
then glue the sink to the glimmer.
or use the screw holes and 2 glimmer plates and lots of arctic silver, or will the silicone pad suffice?

A good quality and really thin 2 sided tape is going to be the best thing unless you go one step better and also drill holes into the cooler block to mount it using screws. I would have really like to do this today but the heatsink fins are too closely spaced to fit holes/screws in there. I suppose if you have the right tools you could tap your own threads for some machine screws. I'm just going to change mine slightly one day so it has 2 sided tape, I've adjusted the USB extender cables on mine so there is nil pressure applied in any direction to the miners: they won't be going anywhere.

[Trongersoll]

Okay, if your tests say so...
how to mount them professionally on a aluminium heatsink?
Rubber bands will not work for longer amounts of time...

so my idea was to glue one or two glimmer pads for TO247 or the like to the board, thinner and isolating, as a replacement for the silicone pad.
then glue the sink to the glimmer.
or use the screw holes and 2 glimmer plates and lots of arctic silver, or will the silicone pad suffice?

A good quality and really thin 2 sided tape is going to be the best thing unless you go one step better and also drill holes into the cooler block to mount it using screws. I would have really like to do this today but the heatsink fins are too closely spaced to fit holes/screws in there. I suppose if you have the right tools you could tap your own threads for some machine screws. I'm just going to change mine slightly one day so it has 2 sided tape, I've adjusted the USB extender cables on mine so there is nil pressure applied in any direction to the miners: they won't be going anywhere.

Trillium, are we going to see you overclock soon? extra cooling is cool, but has it improved your hash rate? or did you just turn good enough into great?

[amigaman]

cooling does not improve hash rate, nor does it for the errors.
the asics are fixed on 28x the frequency as hash rate, so the only way to increase hash rate is to replace the 12.000 oscillator with one with more frequency, 14,31818 being most common.
that'll give you 390+ instead of 336, a massive 16% increase.
Also the maximum fc has stated for the chips to be stable with.

but that needs more voltage, that in turn will increase heat, so better cooling is the first step to increase hash rate.

[Bluestreak66]

Okay, if your tests say so...
how to mount them professionally on a aluminium heatsink?
Rubber bands will not work for longer amounts of time...

so my idea was to glue one or two glimmer pads for TO247 or the like to the board, thinner and isolating, as a replacement for the silicone pad.
then glue the sink to the glimmer.
or use the screw holes and 2 glimmer plates and lots of arctic silver, or will the silicone pad suffice?

A good quality and really thin 2 sided tape is going to be the best thing unless you go one step better and also drill holes into the cooler block to mount it using screws. I would have really like to do this today but the heatsink fins are too closely spaced to fit holes/screws in there. I suppose if you have the right tools you could tap your own threads for some machine screws. I'm just going to change mine slightly one day so it has 2 sided tape, I've adjusted the USB extender cables on mine so there is nil pressure applied in any direction to the miners: they won't be going anywhere.

I drilled and taped two holes in the intel cpu heat sink I had and used the silicone pad between them. The glue and tape is going to have higher thermal resistance than the silicone pad. I stopped using the cpu cooler because it was to bulky with my hub so I used the rubber bands and smaller heat sinks. (The rubber bands have been on there over two weeks running 24/7 BTW) If i have good results I probably will just bolt them to a spare water block and plumb them into my pc water cooling. From what I've seen mounting it with screw does not warp the board enough (if any) to cause a reduction in heat transfer neither doe the silicone pad. I have some artic silter two part epoxcy but i don't think i want to chance ruining one by epoxying it. I'm going to look for my Ir thermometer to show some data.

Trillium, are we going to see you overclock soon? extra cooling is cool, but has it improved your hash rate? or did you just turn good enough into great?

cooling does not improve hash rate, nor does it for the errors.
the asics are fixed on 28x the frequency as hash rate, so the only way to increase hash rate is to replace the 12.000  with one with more frequency, 14,31818 being most common.
that'll give you 390+ instead of 336, a massive 16% increase.
Also the maximum fc has stated for the chips to be stable with.

but that needs more voltage, that in turn will increase heat, so better cooling is the first step to increase hash rate.

I've got some 16mhz and 18mhz oscillators on the way as well as a few hundred 0805 resistors

[Trillium]

Sorry my electronics skills are not so great when it comes to oscillators and things. I think there is too high of a chance I would destroy it. I will leave this up to the more skilled enthusiasts and electronics engineering types among you.

So yes I turned good enough into great, I suppose this might increase the lifespan of the devices in the very long term, but impossible to say for sure.

Hopefully we've shown though that it is easy enough to cool the ASIC itself, without too much work (2 minutes with a rotary tool per miner isn't hard).

In regards to the error rate, I have looked at this before, and I have a bunch of logfiles recorded over 24 hours+ with the miners setup in different ways. Although I have not done much statistical analysis on the logs yet, a quick summary would be:

• Miners are not effected by standard levels of EMI (electromagnetic interference) that you'd find in a home or office. I improvised a substantial Faraday cage-like enclosure and the HW error rate is unchanged.
• HW error rate is not reduced by improved cooling. It is possible that having no fan on the device (very high temps) might increase HW error rate, but I haven't specifically looked at that.
• HW error rate is not increased by using up to 2 meter long USB 3.0 current rated extension cables. But using USB 2.0 rated cables (lower current handling capability) is not recommended from my experience: it causes intermittent dropouts or restarts if the miner can even detect it's there.
• I have tentative results that show the error rate from the Dlink DUB-H7 hubs is higher than running from computer USB ports or other hubs. The reason for this seems to be that the DUB-H7 - at least the one I have - outputs a voltage of ~4.8 volts instead of ~5.0 volts you'd expect from USB. Many of the PC ports I've tested output at closer to 5.1 volts. HOWEVER the difference in error rate is very small, you need to test both cases for days to get enough data to prove it statistically.

There are not too many other things I can think of testing to figure out why HW errors might occur: there is of course the most realistic scenario in that it's just because of the circuit design / component selection and/or the ASIC chips themselves (perhaps some small flaw that ASICMiner thought it wasn't worth worrying about because it's only <1% error rate anyway. I had even speculated that HW errors might caused by cosmic radiation, because we don't know if ASICMiner engineered any of the usual technologies found in processors today (CPU/GPU) that are designed to detect and mitigate problems caused by cosmic rays. In my own experience with CCD detectors (which I realize are not the same as die) a 10x10 mm CCD will have an collision with cosmic rays at least a couple of a times per hour. The ASIC die is obviously smaller than that and roughly corresponds to the error rates that I see. Of course, shielding cosmic rays is almost impossible which is why processors are usually engineered to detect and recalculate erroneous values instead.

[Bluestreak66]

Sorry my electronics skills are not so great when it comes to oscillators and things. I think there is too high of a chance I would destroy it. I will leave this up to the more skilled enthusiasts and electronics engineering types among you.

So yes I turned good enough into great, I suppose this might increase the lifespan of the devices in the very long term, but impossible to say for sure.

Hopefully we've shown though that it is easy enough to cool the ASIC itself, without too much work (2 minutes with a rotary tool per miner isn't hard).

In regards to the error rate, I have looked at this before, and I have a bunch of logfiles recorded over 24 hours+ with the miners setup in different ways. Although I have not done much statistical analysis on the logs yet, a quick summary would be:

• Miners are not effected by standard levels of EMI (electromagnetic interference) that you'd find in a home or office. I improvised a substantial Faraday cage-like enclosure and the HW error rate is unchanged.
• HW error rate is not reduced by improved cooling. It is possible that having no fan on the device (very high temps) might increase HW error rate, but I haven't specifically looked at that.
• HW error rate is not increased by using up to 2 meter long USB 3.0 current rated extension cables. But using USB 2.0 rated cables (lower current handling capability) is not recommended from my experience: it causes intermittent dropouts or restarts if the miner can even detect it's there.
• I have tentative results that show the error rate from the Dlink DUB-H7 hubs is higher than running from computer USB ports or other hubs. The reason for this seems to be that the DUB-H7 - at least the one I have - outputs a voltage of ~4.8 volts instead of ~5.0 volts you'd expect from USB. Many of the PC ports I've tested output at closer to 5.1 volts. HOWEVER the difference in error rate is very small, you need to test both cases for days to get enough data to prove it statistically.

There are not too many other things I can think of testing to figure out why HW errors might occur: there is of course the most realistic scenario in that it's just because of the circuit design / component selection and/or the ASIC chips themselves (perhaps some small flaw that ASICMiner thought it wasn't worth worrying about because it's only <1% error rate anyway. I had even speculated that HW errors might caused by cosmic radiation, because we don't know if ASICMiner engineered any of the usual technologies found in processors today (CPU/GPU) that are designed to detect and mitigate problems caused by cosmic rays. In my own experience with CCD detectors (which I realize are not the same as die) a 10x10 mm CCD will have an collision with cosmic rays at least a couple of a times per hour. The ASIC die is obviously smaller than that and roughly corresponds to the error rates that I see. Of course, shielding cosmic rays is almost impossible which is why processors are usually engineered to detect and recalculate erroneous values instead.

The internal regulator on the silabs chip regulates the usb voltage to 3.3v powering all chips on the board. The data sheet shows it will work fine down to at least 4v. LDO regulators usually require ~200-300mv of voltage drop to regulate without ripple so it should be fine to 3.6v on the usb port without issue. However as voltage decrease more current is needed for the core supply, so with a lower voltage you would exceed the 500ma rating of the usb spec and the port would be disabled. My powered hub delivers 4.73v at the connector and there is no increase in error rate vs directly in my pc. The error rate is more likely due to noise coupling into the miner or the hub interfering with the signals. There is always going to be errors due to the internal design of the asic no design is perfect. Less than 1% errors is pretty good. Cosmic rays and moon phases probably have no measurable effect. 

[amigaman]

The HW errors are in the design of the ASICs and should be below 1%.
Higher values usually indicate Problems with powering the sticks, but they work well with as low as 4V, as long as this is stable e.g. the voltage drop is from cabling, not from overloaded PSU.
Regarding the frequency: the higher the internal frequency is, the higher the supply voltage needs to be.
400MHZ (14.x osc) needs 1,15-1,2V, 336MHz (the stock 12MHz) will work with 1.05V.
There are some borders you can't cross:
- Higher voltages cause problems inside chips, they have an actual max they can't exceed and will break if you try. FC does not state where this is, afaik.
- Higher Voltages cause higher currents, generating more heat. The absolute number is limited by the package. Even the best cooling will not lower the chip-to-case resistance, and chip temp is what counts. FC also does not state what is abs max voltage/current consumption.
- The regulator (AZ1021) is limited in its capacity to deliver current. Leaving the other chips out of the math just for convenience: the stick uses 500mA+ on 5V, making 2.5W. The regulator roughly outputs 1V, making it 2.5A. So we can safely assume we are at 2 to 2.5A load on the AZ1021, which is close to max. I'd say on 1.15 or 1.2 max you reach the safe 3A. And you don't want the parts to be on 100% load or over for extended periods of time.

My conclusions on this are:
- the stick is limited to roughly 400MHz with the 14.31818 osc (some seem to have this fitted on their non-blue-non-v3-sticks).
- more may be possible, but you'll need some preconditions: better cooling, and replacing the built in regulator with some 5A capable 1.3V-range
- even then you'll be limited to lets say 14.7, 15Mhz if you can get some osc's for that.
- i don't believe 16 or 18MHz will be even working or stable for longer amounts of time.

I'd say fixing them up to 1.2V and 14.31818 or 14.7x and not have them blown up heatwise will be great succes, but expect them to have 700mA on the USB side, so you'll need beefed up hubs or even direct supply via the pads at boards end.

[Bluestreak66]

The HW errors are in the design of the ASICs and should be below 1%.
Higher values usually indicate Problems with powering the sticks, but they work well with as low as 4V, as long as this is stable e.g. the voltage drop is from cabling, not from overloaded PSU.
Regarding the frequency: the higher the internal frequency is, the higher the supply voltage needs to be.
400MHZ (14.x osc) needs 1,15-1,2V, 336MHz (the stock 12MHz) will work with 1.05V.
There are some borders you can't cross:
- Higher voltages cause problems inside chips, they have an actual max they can't exceed and will break if you try. FC does not state where this is, afaik.
- Higher Voltages cause higher currents, generating more heat. The absolute number is limited by the package. Even the best cooling will not lower the chip-to-case resistance, and chip temp is what counts. FC also does not state what is abs max voltage/current consumption.
- The regulator (AZ1021) is limited in its capacity to deliver current. Leaving the other chips out of the math just for convenience: the stick uses 500mA+ on 5V, making 2.5W. The regulator roughly outputs 1V, making it 2.5A. So we can safely assume we are at 2 to 2.5A load on the AZ1021, which is close to max. I'd say on 1.15 or 1.2 max you reach the safe 3A. And you don't want the parts to be on 100% load or over for extended periods of time.

My conclusions on this are:
- the stick is limited to roughly 400MHz with the 14.31818 osc (some seem to have this fitted on their non-blue-non-v3-sticks).
- more may be possible, but you'll need some preconditions: better cooling, and replacing the built in regulator with some 5A capable 1.3V-range
- even then you'll be limited to lets say 14.7, 15Mhz if you can get some osc's for that.
- i don't believe 16 or 18MHz will be even working or stable for longer amounts of time.

I'd say fixing them up to 1.2V and 14.31818 or 14.7x and not have them blown up heatwise will be great succes, but expect them to have 700mA on the USB side, so you'll need beefed up hubs or even direct supply via the pads at boards end.

High voltage is required to stabalize the chips at higher frequency. Current draw does not go up with voltage it goes up with frequency. It is well know and documented that these chips run fine on 1.15v @ 392mhz. These chip were designed to operate at much higher "theoretical frequency" but was lowered to 336mhz due for current draw and thermal reasons. The design thread for these chips state that the max voltage is 1.8 per the foundry. I don't plan on going anywhere near that. 448mhz does not seem like much a stretch from 392mhz. 504mhz, yea thats probably a stretch with the current design. There are at least 2 threads on here from when these chips where being designed its worth a good read.

[r3wt]

The HW errors are in the design of the ASICs and should be below 1%.
Higher values usually indicate Problems with powering the sticks, but they work well with as low as 4V, as long as this is stable e.g. the voltage drop is from cabling, not from overloaded PSU.
Regarding the frequency: the higher the internal frequency is, the higher the supply voltage needs to be.
400MHZ (14.x osc) needs 1,15-1,2V, 336MHz (the stock 12MHz) will work with 1.05V.
There are some borders you can't cross:
- Higher voltages cause problems inside chips, they have an actual max they can't exceed and will break if you try. FC does not state where this is, afaik.
- Higher Voltages cause higher currents, generating more heat. The absolute number is limited by the package. Even the best cooling will not lower the chip-to-case resistance, and chip temp is what counts. FC also does not state what is abs max voltage/current consumption.
- The regulator (AZ1021) is limited in its capacity to deliver current. Leaving the other chips out of the math just for convenience: the stick uses 500mA+ on 5V, making 2.5W. The regulator roughly outputs 1V, making it 2.5A. So we can safely assume we are at 2 to 2.5A load on the AZ1021, which is close to max. I'd say on 1.15 or 1.2 max you reach the safe 3A. And you don't want the parts to be on 100% load or over for extended periods of time.

My conclusions on this are:
- the stick is limited to roughly 400MHz with the 14.31818 osc (some seem to have this fitted on their non-blue-non-v3-sticks).
- more may be possible, but you'll need some preconditions: better cooling, and replacing the built in regulator with some 5A capable 1.3V-range
- even then you'll be limited to lets say 14.7, 15Mhz if you can get some osc's for that.
- i don't believe 16 or 18MHz will be even working or stable for longer amounts of time.

I'd say fixing them up to 1.2V and 14.31818 or 14.7x and not have them blown up heatwise will be great succes, but expect them to have 700mA on the USB side, so you'll need beefed up hubs or even direct supply via the pads at boards end.

first time asic user i have a question, how do you detect hardware errors? i have one of the rev 3.00 boards and the only error i seem to get is occasionally cgminer says "connection to stratum pool 0 interrupted or "stratum pool 0 requested work restart"

[amigaman]

It's the HW value from CGMiner.
should be max 1% of the A + R value, i think.
Some HW's are ok, but should get not too much.

@Bluestreak:
Higher voltage leads to more heat produced (It's V x A, remember?), so you usually only upvolt to increase clock.
I'm waiting for your report of having a sufficiently cooled stick running on 450 with not that much HW errors.
Please make a step-by-step with pictures.
I have ordered some 14.7 oscillators and plan to use them, only cooling is my problem.

[r3wt]

It's the HW value from CGMiner.
should be max 1% of the A + R value, i think.
Some HW's are ok, but should get not too much.

I got

Code:

A: 146 R: 0 HW: 19

thats not to bad in my opinion, but i'm not expert

[Trillium]

It's the HW value from CGMiner.
should be max 1% of the A + R value, i think.
Some HW's are ok, but should get not too much.

I got

Code:

A: 146 R: 0 HW: 19

thats not to bad in my opinion, but i'm not expert

You need to do the calculation with diff 1 shares. If your shares are not diff 1 then you need to convert them to diff 1. If your pool is set to 4, 8 ,128 diff for example, you need to do the math in that case.

[pidge]

re: cooling of stock USB erupters, I've posted pictures of my "ghetto cooling duct" here: https://bitcointalk.org/index.php?topic=253749.msg2794024#msg2794024

[amigaman]

The pictures are black X'es.
Only me?

[notlist3d]

The pictures are black X'es.
Only me?

For me no pictures at all.

[amigaman]

FYI:
i've got 2100+ "A"ccepted, 3 "R"ejected and +40 "HW" Errors after around 15hrs, on CGMiner 3.3.1, Win7Pro x86, Diff auto/mostly on 2 (because i have 10 plus 2 GPUs on my rig (they output 280-320 each at the moment, up to 650 when winter)).
Cooling is up to 36°C room temperature (from outside), no fans, just slight air movement.
Sticks are "as sold", 2 10port hubs (logilink UA0096, i believe) modded to "without backpower", 5 sticks each every other slot.

Is that good? I believe Yes.

[r3wt]

FYI:
i've got 2100+ "A"ccepted, 3 "R"ejected and +40 "HW" Errors after around 15hrs, on CGMiner 3.3.1, Win7Pro x86, Diff auto/mostly on 2 (because i have 10 plus 2 GPUs on my rig (they output 280-320 each at the moment, up to 650 when winter)).
Cooling is up to 36°C room temperature (from outside), no fans, just slight air movement.
Sticks are "as sold", 2 10port hubs (logilink UA0096, i believe) modded to "without backpower", 5 sticks each every other slot.

Is that good? I believe Yes.

i'm jelly