No.

Seek Thermal. $199.
http://www.thermal.com/

Yes, it's because you're on 208V. Notice that your machine is configured to use 1360W max per PSU, but the software is artificially limiting it to 1250W. That's because these PSUs are less efficient (i.e. generate more heat) on lower voltages, so they can't deliver as much power, so Zvi hard-limited them in order to reduce the chance of PSU failure. Spondoolies hasn't done much testing with 208V, so they don't know what the PSU limits should be, but they know 1250 is safe. If you want to try to squeeze more performance out of it, you can bypass the 1250W limit with Just make sure to reduce your PSU limits in the Settings page first. My suspicion is you'll be able to get about 1280W to 1330W out of those PSUs at that voltage, but I've never tried. 1360W is probably too high.

I got a new toy. Thermal photos taken at Toomim Brothers Bitcoin Mining Concern:

Rack of SP30s, viewed from behind:


Looking inside an SP30. Top view, exhaust fans on the left. This shows some deficiencies of the camera. It's miscalibrated (reported temps are about 60°C above actual temps -- it appears this was fixed with a new software version), and it does a poor job with the aluminum heatsinks due to their low emissivity/high albedo. The heatsinks are actually much hotter than this image shows. Moving the camera angle caused the heatsinks to appear to change temperature, suggesting that much of what we're seeing on the heatsinks is reflected IR from surrounding objects, including the PCB and the walls/ceiling of the room.


225 amp 3-phase cable bundles in our overhead cable tray, viewed from beneath. Notice the bundles on the top of the image are warm (about 30°C in 11°C ambient) but the ones on the bottom are cool. That's because we're barely running any current through the bottom bundles, but the top bundles are running at full capacity.

I got a new toy.

Rack of SP30s, viewed from behind:


Looking inside an SP30. Top view, exhaust fans on the left. This shows some deficiencies of the camera. It's miscalibrated (reported temps are about 60°C above actual temps -- it appears this was fixed with a new software version), and it does a poor job with the aluminum heatsinks due to their low emissivity/high albedo. The heatsinks are actually much hotter than this image shows. Moving the camera angle caused the heatsinks to appear to change temperature, suggesting that much of what we're seeing on the heatsinks is reflected IR from surrounding objects, including the PCB and the walls/ceiling of the room.


225 amp 3-phase cable bundles in our overhead cable tray, viewed from beneath. Notice the bundles on the top of the image are warm (about 30°C in 11°C ambient) but the ones on the bottom are cool. That's because we're barely running any current through the bottom bundles, but the top bundles are running at full capacity.

If you're using 220V or greater, you should change the PSU limits on your SP31. If you have Murata power supplies, you should also ssh in and run this: Depending on your ambient temperature, voltage, and power supplies, you may be able to set your PSU limits to somewhere from 1300 to 1370. That can improve performance by as much as 200 GH/s. Keep in mind that Murata power supplies don't recover from overload as nicely as Emersons do, so if you have the Murata type, make sure to give them a bit more headroom from the overload point.

This issue was fixed for SP10s in 1.5.10. I tested it over a week ago, and SP10s work fine on 1.5.10 in temperatures down to at least -10°C.

I recommend users of SP20s and SP3Xs try to keep their intake temperatures above 5°C, as we've noticed reliability issues at low temperatures. The current hypothesis is that thermal expansion coefficient mismatches in the ASIC package and hashboard PCB is causing solder joint fractures and separation at very low ambient temperatures.

Relevant: https://bitcointalk.org/index.php?topic=521520.msg7983457#msg7983457

It can go lower than the initial value.

There are a couple of bugs in SP30s that show up at temperatures below about 8°C.

One causes individual ASICs to become nonfunctional and starts to occur around 8°C. It appears to be a hardware problem, and it affects a small minority of ASICs (about 1% or less). Spondoolies is actively investigating this issue. I expect they will be able to fix it for new machines, but it's uncertain whether they'll be able to fix it for old ones.

Another issue caused machines to shut down if the intake is below 0°C. This appears to have been fixed by firmware 2.5.13.

Except for these bugs, performance of Spondoolies gear improves considerably as temperature drops. On an SP30, you can expect about 200 GH/s higher hashrate for a 20°C drop in temperature.

horcoff, want me to test this with a miner?

We at Toomim Bros Bitcoin Mining Concern are accepting for hosting non-rackmountable machines like SP20s, Antminer S3s, and ASICMiner Prismas now. We also can make and sell server PSUs for miners that require PCIe connectors (e.g. SP20) for about $100.

Edit: $100 PSU offer no longer available.

I've confirmed that version 2.5.13 will usually continue to work below 0°C, whereas 2.5.12 and before will go into "thermal shutdown" mode because -1°C gets misread as 255°C by the firmware. However, there are also some intermittent noise or ASIC reliability issues at low temperatures that 2.5.13 does not fix. I suggest that people try to avoid very low temperatures for another week or two while we look into those noise issues. Once an ASIC has an issue at low temperature, it makes the whole machine stop working until you isolate and disable that ASIC. While the ASICs usually recover after you bring the temperature back up, they don't always. This headache is really not worth the extra 30 or 50 GH/s you'd get from the lower temperature.

Something like that. I haven't had a chance to meter any SP20s yet. I expect there's a wider range of efficiency/performance options available with the SP20 than the SP30, so how much we charge may end up depending on how it's configured and how efficient the PSUs we end up using are. (I hope they'll be around 92% efficient.)

This happens on occasion. Unless the ASIC is physically damaged, it should recover if you restart minergate. Firmware 2.5.12 has this problem less frequently than 2.5.6, but they all can get it for different reasons. If it's a regular problem, you can set up a script or cron job to restart minergate once a day or something like that.

Yes, but I can't promise that it would withstand return shipping, as I was considering just soldering the PCIE wires directly onto 1200W server PSUs. If you want to sell it back to us when you're done with it, I'd probably buy it back for around $50. I would also be willing to lease them for $10/month.

No fee for transfer of ownership. We will ship it to you or your next hosting provider at cost. We should still have the packaging for these machines, so at-cost would probably just mean postage. The paper padding that Spondoolies uses tends to wear out with repeated use, so we might have to get some extra padding material. Our rates are published on our website, and we don't deviate from them. Month-to-month terms are $99/kW per month. Three-month terms are $93/kW.

The 1200W rating that is listed on Spondoolies's page is for 12V DC power. In actuality, most SP10 PSUs are capable of supplying 1260W, and some can go as high as 1265W. Since the PSUs the SP10 uses are about 91% efficient at that load, that means around 1260/0.91 = 1384W of AC draw.

Horcoff requested metered billing for these machines. We measured these SP10s as consuming 1.314 kW each at the wall, which is lower than we typically see for SP10s in our datacenter (typically around 1.38 kW). However, it was much warmer then, and most of his SP10s were not hitting their PSU limits in 12V DC power. Right now, our machines are getting intake temps around 10°C, and it looks like horcoff's SP10s are hitting their PSU limits as a result (and hashing faster).

Toom.im is now accepting SP20s for hosting. If you need a power supply, we can supply one for $100.

Some machines work fine at low temperatures. I've seen a few operate just dandy at intake temperatures as low as 2°C. However, others will show problems at intake temperatures as high as 8°C.

When a machine has trouble at low temperatures, it's often specific ASICs that show the problem (often #2 and #17). It's not always the ASICs in the front that have issues, though it seems more common to be those. Once an ASIC has trouble at low temperatures, it sometimes has trouble even when warmed up. Sometimes it has to be disabled permanently; other times it can be enabled once the machine is warmer.

The 1626 is calculated based on the clock speeds of the ASICs, and should be the actual number of calculations per second your ASICs are performing. Your machine only reports wins back to the pool server, or hashes with a value below the current difficulty setting for that share. The second value is an estimate of the number of hashes your machine must be doing based on the number of wins you have sent back to the server over the last 5 minutes. It's noisier and will bounce around a lot, but this is the number that the pool will reward you for.

Basically, the second number is accurate, but the first number is precise. Both numbers have their uses, which is why both numbers are displayed. They should be within about 20% of each other if your machine has been running for 5 minutes or more.

It's looking more certain that what we've been seeing in our DC is a poor reaction of SP30s to temperatures below about 6°C. We haven't seen any issues with SP10s yet at these temperatures.


This is what we're doing.

I designed our system with the intent of using 100% fresh air -- air would go through our miners once then be exhausted outside. With the revelation that SP30s have problems below about 8°C, we've had to revise that system, as it's currently -4°C outside, with a forecast low for tonight of -11°C. Over the last few days, we have set up a system whereby we can control the amount of fresh air vs. recirculated air we get by turning on and off fans. It's a bit rough still, and produces hotspots when we recirculate too heavily, but it works.

The tricky thing is getting the air to mix where you want it to mix. Our hot air containment system was designed with the intent of having the interior be hot and low pressure. In order to get air to recirculate, we have to make it hot and high pressure. This means we get hot air leaking out at every hole or crack that we have in the containment. Unfortunately, the most common cracks and holes we have in the system are the rack mounting bars for the SP30s, which means we tend to get uneven air temperature within a single SP30. Air on the side tends to go through the SP30, then circle around and come back out the front by the mounting bars, and then go through the SP30 again. On the sides of the machine, the air will make several circuits through the SP30, but in the center, the air goes through only once. We're working on sealing off those little cracks. There are other problems too, like making sure that the eight corners of our pod all are exposed to roughly the same temperature, and making sure that our intake and exhaust air streams get mixed well.


You can click on "Firmware Upgrade" then "Manual Selection" to access several firmware versions which Zvi currently thinks are decent. You can download all of the old firmware at http://storage.googleapis.com/spond_firmware/spon_2.X.YY.tar, where X and YY are the version numbers of the firmware you want to use.