Please be careful with that. Someone could hurted their finger. At least display some WARNING: HAZARDOUS MOVING PARTS signs nearby.

Stop mining with tachyons, be sure to use electrons in the future.

I'd throw some in a SEM and do some elemental analysis.. but I have a damned thesis to write!

Hey Canary and yes I have abused your nice little miners a little bit since I got them but they are still alive

You are technically correct that the grey thermal pastes usually contain silver particles which are themselves probably conductive, however a large percentage of the material is actually oils and binder materials (the exact composition is usually proprietary though you can work it out with some acetone, filtering equipment, and precision balances). These binders are generally non conductive so you end up with something that is a less conductive than a single piece of solid metal.

Of course, alumina and other metal oxide based (white) pastes are generally assumed to be entirely non-conductive because most metal oxides are excellent electrical insulators.

Just how conductive is silver based thermal paste? Well I did a quick test, I took some Arctic Silver 5 and smeared it over some perfectly smooth plastic in about that kind of thickness you would use for any CPU/GPU. Then I took some aluminium metal and made two conductors on either side. The conductive path of paste was 50 mm long and 10 mm wide and then I used a multimeter to measure the resistance across the paste:

For the measurement I used my digital multimeter, which caps out after 40 megaohm. Just how much is 40 megaohm? Well its not allowing much conduction at all. I managed to get 7 megaohm measuring between my fingertips on each hand, through my body, pressing very hard on the probe pins. Pushing lightly with my fingertips, I could raise the resistance until it reached 40 megaohms and then finally went over its limit.

The 50x10 mm strip:

• 

However, it did not give any measurement, indicating the resistance was >40 megaohms! I even pressed down firmly on the aluminium to ensure it was in good contact with the paste on both sides.

Not one to give up, I reduced the area to 10x10 mm. Again, no reading was possible as the resistance was far too high.

So I reduced the measurement area to about 10x 0.3 mm, perhaps the spacing of pins you would find on a chip. In the photo, you can barely see the gap between the aluminium conductors its so small:

• 

Again, it was suprisingly still >40 megaohms.

I raged, and simply pushed out a glob of AS5 and dug the multimeter pin probes straight into the paste:

• 

Still, there was no conductivity. I found that even if the metal probes are lightly touching each other inside the paste, there is still no measurable conductivity.
 In the picture above the red probe is actually resting ontop of the black probe, but still no measurement can be had. Only by pushing the tips together fairly hard within the glob can you get a reading (and as you'd expect, when the metal tips actually touch, it goes to <0 Ohm ie. the other limit of the DMM's measurement range.)

This is kind of suprising to some extent but then I realise that I have seen plenty of CPU and GPU where the onchip SMD capacitors have been drowned in thick silver based pastes but still functioned perfectly. In fact in my experience this is an amusing 'problem' that used to be seen a lot with Pentium 3 and 4 equivalent Xeon chips in severs (especially from Dell!). This is kind of like how you can probe pins on a running PCB circuit with digital components using an oscilloscope: because of the high probe inductance/resistance it usually won't effect the circuit/devices operation. I realize that at kHz and above frequences other electrical effects in seemingly non-conductive materials become considerable though I think in this case you can get away with it thanks in part to the design tolerances of the components and signal:noise will probably still be pretty good.

tl;dr - Getting silver thermal paste (at least Arctic Silver 5) on pins / exposed parts of CPU/GPU/ASIC has no effect on the performance.
Edit: In the overclocking thread I have discovered that capacitive effects from AS5 actually DOES cause problems with the USB miners. However I found that Ceramique 2 compound worked fine, although it too would have some capacitance so YMMV. The optimum solution seems to be using 2 sided tape.

PS: I scraped some 'dried' thermal paste off some scrap sever CPU heatsinks and measured the resistance of that, and again it was the same. The conductivity doesn't really increase in any measurable way when the paste has set/dried.

(Disclaimer: A few manufacturers, including Arctic Silver inc, discourage you from getting it on exposed pins/components because of the other electrical effects I mentioned above. Don't assume from my shitty 10 minute tests that your $$$ device won't implode if you're messy with paste.):

AS5 instructions:

Although this argument is all kind of irrelevant... because its easy to not make a mess with the thermal compound and avoid getting it on the pins, but also because the ASIC die is apparently positioned upside down relative to how die are usually set in epoxy on a CPU/GPU. As a result, cooling is actually superior through the PCB and into the heatspreader rather than putting a tiny heatsink on the topside of the ASIC. See my link above in the overclocking thread for more info. That's why the manufacturers went with a heatspreader under the PCB instead of shipping them with a tiny heatsink ontop.

I've also measured mine with a 'laser' (IR) 'thermometer' (pyrometer), but stopped when the top of the ASIC and the heatspreader on the bottom both exceeded 70 deg C (ambient ~26 deg C too and without any fans). The operating temp limit of some of the components on the PCB is just 85 deg C, getting too close for my comfort. On that topic, does anyone know if AM have released technical information about their chips with electrical data and some recommended temp limits? I fear that this information may be out there but in Chinese.

It is actually pretty hard to use a IR thermometer at close range because the laser beam (which has nothing to do with the temperature measurement and is intended as a crude indication of the measurement position) and the location that the thermopile sensor 'sees' and takes its measurement on are several centimeters different at close range (<6", usually), because they are designed to operate usually at distances of 6" or more and measure relatively large surface areas. For example my 12:1 optics model reads an average from a 1.5" diameter spot size at 12" distance. Most have fixed optics too so the best you can do is guesstimate what component is being measured at any time and if you end up measuring a site 50% over the top of the chip and 50% over the PCB, the sensor is giving an average reading which is going to mislead you and the actual component temps could be much higher than the averaged value shown on the display.

A while ago when I got my erupters I had planned to borrow one of my work's colour FLIR thermal imaging cameras but I have found that one has been taken offsite for unknown period of time and other fails cal and shuts itself down.

I have come up against this problem with trying to cool high power VRM's with poorly fitting waterblocks on cards like 5970's. The simple fact of the matter is that those grey silicone(?) thermal pads (which often tear extremely easily) have 'good' thermal conductivity but as only as far as soft compressible materials go. THEY ARE COMPLETELY INFERIOR TO PROPER HEAT CONDUCTORS LIKE ALUMINIUM, COPPER, SILVER METAL by at least a few orders of magnitude. In my experience they are also inferior to good quality thermal pastes.

If you absolutely must use those thermal adhesive pads, try to get ones that are as thin as possible, more heat will pass through it that way giving you better temperatures. Some thermal adhesive tapes are extremely thin but offer good stickiness, these are what you want to use.

Just like with those pads, even thermal paste (and thermal adhesive) can be very problematic if you apply it too thick. There is a very fine line between using too little and too much but hey, it's still better than having air gaps (air being a gas is one of the best thermal insulators).

Is this your own data (from the debug log or something) or is this from a site? If the later can you provide a link, thanks.

I have managed to find 4 blocks in the last 3 days with ~2500-2800 PPS on one machine, at current market that is about $55 USD, but the CPU itself only uses ~100 watts on the AC side so that's about $0.60 USD per day in power where I live.

This is pretty outlandishly good mining compared to say, mining BTC or LTC or any other altcoin with GPU's where the profits seldom outweigh power costs by any substantially large margin. With power costs where I live I can barely break even mining the most profitable altcoins, so mining primecoins seems very odd to me.

Something will have to give sooner or later. Either the difficulty is going to explode as more people come to mine or the market will change substantially. But of course that is just my opinion.

Unfortunately heat transfer rate and absolute temperature differential are not the same things and do not have the same impact on cooling performance. It could be in the freezer at -18 deg C but as soon as you apply power to it, that power has to go somewhere as heat through all the materials that surround the ASIC die. It's entirely likely that the die is still sitting at 50+ deg C.

I'd be willing to bet that you could remove more heat effectively with watercooling at ambient temps than running a half-assed sub-ambient cooled system in a freezer at -18 deg C, even if it was actively cooled inside the freezer (with fans).

At sub-ambient temps condensation is a real threat to electronics and I'd be willing to bet if you just put an erupter on a USB extender cable* and slide it into your freezer it will be destroyed within seconds. If you went to the effort of sealing the overclocked erupter in an almost perfectly dry isolation vessel, with heatsinks on both the top of the chip and on the PCB underneath, with a fan inside it to circulate the air, you might have good enough heat transfer to cool the chip better than with just a crude watercooling setup. But generally in that setup you are going to be limited by two things: The thermal conductivity of air is very poor, and the heat must still pass through the isolation vessel itself.

The difference with water cooling is that you have a very high heat capacity medium (water) flowing past a metal surfaces (waterblock) at a high flow rate. The metal itself being copper (or brass in my idea above) or aluminium in most cases also has exceptional heat conductivity that is orders of magnitude better than air.

On the other hand, if you had large enough heatsinks removing the heat into the freezer and the metal directly conducted heat from the ASIC chip top/bottom then you would get superior cooling. But I think it would be a huge pain in the ass. It's times like this where die-mounted temperature monitoring like in CPU and GPU would be great, but considering how much the AM ASIC chips cost it's to be expected that they don't have this feature.

If the erupters had built in temperature monitoring on the ASIC die I would have already tried putting on in the freezer just to see what kind of temps can be obtained. However because monitoring the external chip temp is almost meaningless there is little incentive to try. Don't even get me started on actually trying to get high accuracy reproducible measurements of surface temps with thermocouples or IR pyrometry...

I suppose some good news is that as far as I can see none of the components on the erupters are highly susceptible to damage from sub-ambient temperatures. As many people have found out trying to do 'extreme overclocks' on graphics cards, the electrolytic caps don't enjoy being frozen solid. It changes their electrical properties too much. The spec sheets of the major components do state minimum operating temps and many of them are -25 deg C although I wonder if the manufactures actually do failure temperature tests to determine that. I could understand them testing maximum temperature ratings of course, but minimum temp is a bit unusual in my mind and probably not worth their time.

If you wanted to submerge the unit in oil for your testing it will probably be alright as the only component on the circuit that could conceivably be affected is the choke, because the epoxy could be broken down and the magnet wire insulation degraded over time. Which can and does lead to some failures in large oil cooled power transformers.

*FYI, I have already tested running erupters off USB extender cables. I am currently running two erupters each on a 2 meter USB 3.0 rated cable, on a netbook. Interestingly I measure no voltage drop over the 2 meters and the netbook happily outputs 5.06 volts both at its hub sockets and also at the end of the cables while under load. USB 2.0 cables are usually rated to much lower current carrying limits so those might not be able to handle it. Then again 2.5 watts on a unmodified erupter is pretty small amounts of power to begin with.

I was thinking about heat removal from these little things and because of their size it presents some unique problems. If you are serious about trying to reach such high frequencies then I might suggest watercooling.

But of course, where to find such a small waterblock??? Well I was thinking that the best idea might simply be to buy some copper or brass plumbing fittings from a hardware store, which has a surface around the right size to sit over the PCB, by resting the side of the nut against the ASIC. I was thinking maybe the nut of a screw threaded fitting, then you can just run the coolant through some matching pipe that is joined to the fitting. You could repeat this on both sides of the ASIC since the underside will remove quite a lot of heat too (ie. in the standard design most heat is sinked through the base of the chip anyway).

Of course it probably isn't practical by any means... but if you really want to say "I DID IT!"...

https://i.imgur.com/Di9TtVG.jpg

Point the cards to the MilkyWay@HOME project, or one of the other BOINC projects that support GPU. They often require double bit precision cards though. To find out if your card supports double bit precision or just single bit, refer to the ATI/AMD GPU list on wikipedia.

http://en.wikipedia.org/wiki/Comparison_of_AMD_graphics_processing_units

Of course though its donating cycles, electricity and operating costs to charity and for science, so you won't see any money from it. Sometimes there are prizes for competitions on some projects but to my knowledge none of them provide reimbursement for the work done (at least none that would approach break-even).

They should work for any pool that you can mine SHA-256 algorithm coins on. You can even solo mine with them, but that would a bad idea for BTC unless you have hundreds of gigahash/s of erupters.

The hubs will either work with your OS and mining software or they will not. The pool doesn't really have much to do with it as far as I can see.

I might be getting some hardware next week that I can run several instances of primecoin on, with 1M, 2M, 4M sievesize etc. Then I can do some statistical analysis on the results. But it will be one of those "the longer you leave it, the better the results" kind of things so it might take a while :

This. But sadly many manufacturers (especially the cheaper ones) will not provide this information. And it would be tedious to produce the data for the curve yourself. To get a DC variable load, a dedicated geek might want to buy some cheap 100+ watt variable wirewound ceramic cored resistors from one of the cheap Chinese vendors (dhgate, ebay, uxcell etc) although it might still be annoying trying to simulate a realistic DC load spread out across the different rails.

example: http://www.uxcell.com/100w-500-ohm-ceramic-tube-wire-wound-variable-resistor-p-143091.html

Or you could just make your own such resistor with some even cheaper furnace element wire and a brick

Lol can we get a video of it mining diff 1 shares in the dark???

Apparently for you (Josh) there is no grey area between "the customer is always right" and "fire your customer"...

... How about: Don't be a rude to your existing customers - or potential customers (newbies sub-forum after all!) - and especially don't get into arguments with them on a public forum.

The best thing you could do right now is ignore this thread from now on. You are just going further down the rabbit hole by responding. You come off as rude and arrogant and this doesn't reflect well on the company. I don't know how you came to be a part of BFL nor how you are positioned within BFL at the moment, but I am amazed they still let you work there given your interactions with some of the people here.

EDIT: Oh right you are a COO?

All motherboards still rely on IRQ. The reason you think they don't is because now it's all handled automatically. You can sometimes still control their allocation manually and make changes if you need to.

The fact of the matter though is that some motherboards just can't run all slots full of cards, presumably because designers of the boards never expected that scenario: their typical customers probably use only 1 or 2 cards, gamers will use more but miners push things far beyong this. The designers and manufactures rely on dual-PCI-slot GPU's covering up other slots - especially the 1x and 4x slots - inherently limiting the number of devices that can fit on a board in 'normal situations'. Cryptocoin miners are unusual in that a computer case is not nessesarily required and we like to use PCI risers. Motherboard limitations are separate to any maximum GPU limitations imposed by the OS.

Was that using 64 bit high performance mod?

SIW (System Information for Windows) can probably show the CPU temps of those xeons (and everything else) even if you are supposed to use the server manufacturers monitoring software for that. (That software is usually annoying to use).

On the sensors tab.

You can assume all electrical energy going into a computer is converted to heat. The TDP of a 7950 is 200 watts. It will be more if you overclock. It might be less...

http://en.wikipedia.org/wiki/Comparison_of_AMD_graphics_processing_units

http://en.wikipedia.org/wiki/Thermal_design_power

I'm sure some people here can tell you their own experience with how much a 7950 uses, most likely measured on the AC mains side of the PSU. So it will vary a bit with the efficiency of the PSU too.

200 watts = 0.2 kW = 0.2 kJ/s {since 1 watt is 1 J/s}

Is a normal mined block 12 XPM?