Basic Instructions for Using the ASICMiner Overclock Kit
OscillatorNear the finger-side corner directly opposite the network jack is Q1, a small silver rectangle labeled 12.000
This is a 12MHz 3.3V self-contained oscillator unit with four pads. Ground, 3.3V, Output and an Enable line. When the enable line is pulled low, the output goes to high-impedance - this was useful on the older-style blades with optional "high speed" because the outputs were tied together and the speed was set by enabling one oscillator or the other.
On the newer units, the second set of pads (Q2, see pictures) may or may not be populated with a 14.318MHz oscillator (for high-speed operation) which is not used on the new blades.
This kit replaces the 12MHz oscillator with a 16.384MHz oscillator. This is the small silver rectangle, basically identical to the one being removed. The easiest way to do this is with a hot-air station. It may help to remove the heatsink from the backside to facilitate heating the oscillator to solder-melt temperature. Make sure the replacement oscillator goes down in the same orientation (see pictures) or it won't function properly. Looking at it from the nearest board edge, the writing should be upside-down.
If you hook up the board and check the config page, it should read all X at this point. This is normal. The chips are operating at too low a voltage to switch properly at the driven rate.
RegulatorsThe next step is to work on the voltage regulators. I'd suggest working over one regulator at a time. The voltage is set on the first regulator (nearest the ethernet jack) by R29, R30 and R33. R30 and R33 are in series between the regulator's sense pin and ground; R29 is the feedback from the voltage output and is what we're currently interested in.
The stock should be a 68.1K 0805 surface-mount resistor, labeled 6812. This code basically means 681 with 2 zeroes - 68100 ohms or 68.1K. We're replacing these with 120K 0805 resistors - they are the small black bits in a paper tape, and should be labeled 124 (or 12 with 4 zeroes, 120000) which will take the VRM's output voltage from stock 1.05 to around 1.25
I'd suggest replacing R29 on the first regulator and then testing out the board. This replacement can be done with just an iron if you're careful. Once R29 is swapped with the 120K, power up and check the config page. If everything's as it should be, the leftmost four chips should show OOOO and the rest still X'd out.
If you're still seeing all X, I'd first probe the output voltage from the first VRM (see picture) and see if it reads about 1.25. If it doesn't, turn it off fast and check your work again. If the voltage is reading in range, consider re-heating the oscillator and making sure all pads are connecting well.
If everything's proper, replace the rest of the VRM's feedback resistors (R41, R53, R65, R77, R89, R101, R113) with 120K and test the board. These resistors are all located in the same place relative to each VRM.
This should bring all chips into O state.
Before long you're going to want to replace the stock fuse with the 15A fuse supplied in the kit, or at least some fuse rated at least 12A. Depending on board configuration, it may be soldered on or in a socket. This fuse is located immediately adjaced the green 12V power jack.
At this point, the blade should be operating at overclock state without any problems. All the essential parts have been changed.
This config page was from a board with unmodified heatsinking and very solid airflow top and bottom. With proper cooling it should converge around 14800-15000 MHS with a 200-205 shares per minute and near 100% efficiency.
OPTIONAL COMPONENTSI noticed on my board that only one of several pad pairs for output capacitors had been populated. The voltage ripple on some of the VRMs was very high, up around 30%, which is hard on the hardware.
Since I didn't trust the one capacitor to source most of about a 12A load for 90% of a switching cycle - the source of a fair amount of ripple. So I added a couple additional capacitors to unpopulated pads (large caps on C166, C167 - see probe points in the Regulator image) which helped share some of the load. These capacitors are the black devices labeled 22-16: 22uF, 16V. The end of the capacitor with a line on it needs to go to the positive side of the output.
The newer round of kits were shipped with 47uF ceramic capacitors (the unlabeled strip) instead of a pair of 22uF tantalum capacitors. These are not polarized, and as such the orientation doesn't matter for installation.
Adding additional capacitance reduced ripple substantially, but I still had some high-frequency switching noise peaking out on occasion. I had on hand some 560pF and 470pF capacitors (the small unlabeled parts with brown bodies), so I laid one down in unpopulated pads for C288. This shunted a lot of the peaks and got it running well enough.
In the second round of kits, these capacitors are in a strip labeled "470pF" to distinguish them from the 47uF capacitors.
The second board I worked on already had another capacitor in C288, and the output ripple on that VRM was really not substantially improved by adding the additional output caps C166 and C167 - but it was already rather low so adding them on the rest of the VRMs was deemed unnecessary.
COOLINGDepending on how well your blade is already cooled, you may not need to change heatsinking at all. The stock heatsink on mine is not very effective, and without chip-side heatsinking and good airflow the hashrate drops out and efficiency bottoms rather quickly because of heat-related errors.
The second board I worked on, I had to remove the heatsink just to free up the oscillator - it pulled heat very well. With no chipside heatsinks and moderate airflow it was running at about 90% output. You'll have to determine for yourself what is required for proper cooling, as there are differences from one board to the next just how well cooling works but I would recommend chip-side heatsinks. Watch out with clearances though, as there are a few capacitors nearby which could be shorted by too much aluminum.
After the first few kits shipped, all kits have been selling with chipside heatsinks supplied - eight 8mmx30mm and four 8mmx15mm heatsinks, as well as 18cm of thermal tape. This may not seem like enough tape, but if used cleverly it is. The tape is actually too wide, so splitting it lengthwise into two 6mm wide strips is just about perfect for covering chips. I cut each strip into eighths lengthwise (by halving, then halving, then halving) and laying each piece across two chips (after peeling off the paper backing). Peel off the green backing and stick one 30mm and one 15mm heatsink end-to-end to completely cover one row of four chips.
Some boards may still run hot, and with reduced efficiency, even with chipside heatsinks unless a lot of cold air is provided. One thing that you may need to do is smooth out and warp slightly the stock heatsink. The large stock heatsinks have a fairly rough surface, and since the screws are all on the edge of the board there's not really any good presure to provide solid contact and good heat transfer from the chips to the heatsink. I'll lay out the process that's given me the most success.
Step 1 - sand the mating surface of the heatsink smooth. All the ones I've encountered were fairly grooved, possibly a poorly made or worn-out extrusion mold. In any case, a few minutes with some 220-grit sandpaper buffed the surface quite a bit smoother, which smooth surfaces make for better contact.
Step 2 - Laying the heatsink down fins-up between two wooden blocks, and using a third wooden block and a hammer, I give the heatsink a slight bend lengthwise down the middle - no more than 1/16" is necessary, checked with a straightedge. Once there's a slight bend in the middle, I put the block halfway between the center and the edge such that this second bend would be right in the middle of the chips. This makes sure that the board has to flex slightly to meet the heatsink, which will force pressure between the two. The second bend makes the curve below the chips more continuous, giving more contact area.
Step 3 - I dabbed some thermal compound on the exposed plated areas on the board's backside below the ASICs, then transferred some of that to the heatsink. I then placed the silicone pad back inbetween (to prevent shorts) but now there's thermal compound helping smooth the interfaces.
This has helped me bring overclocked boards from 30% efficiency to 99% with the same cooling.
If you're interested in purchasing kits, email me (Kittan) for pricing information. Current kits include three oscillators - for 13GH, 14.4GH and 15GH - so you can run the fastest you are able to keep stable. Kits cost 0.025BTC apiece and 0.01BTC shipping within the US for orders of 1 to 12, 0.025BTC shipping for international orders.
Special ThanksTong the excellent hardware resupplier, from which I got the first blade
r00t$ trusting that I could do what I said I could, and his board supplied a lot of the pictures and data for this writeup
|If you like what you saw and you want to contribute to future awesome projects, feel free to donate some BTC: 1GekkosciLeaey8Na9siC8oD5HcMtLnWwd|