Numbers up front:
832 MH/s - I suspect that will increase with a new firmware load in the future. See below as to production schedules.
Consumes between 85 and 90w at full load.
Draws .77A
Exhaust temperature was between 93 - 100 degreees F at the hottest point. Pointing my laser thermometer at various components inside the unit, I saw a maximum temp of about 110F, but it was kind of hard to hold it on some of the smaller components and snap a pic, so there's no pic of the 110F reading. Ambient temp was about 80F.
The case was cool to the touch after 24H mining. Exhaust temp never wavered during my spot checks.
Sound was minimal and my PSU, which is fairly quiet, was louder. I don't have a decible meter, but I would estimate 26 db or so with both fans at full, maybe as high as 30, but I don't think so. In either case, it's a bit less noisy than an Acellero Extreme HSF, that is to say, it's not the least bit offensive to the ears; you could sleep with one of these in your room.
So, the BFL single I got my grubby mitts on is a loaner. I wanted to clear that up righ away. It is not "my" unit, as I don't have a unit. To address some of the tinfoil hats, I offered to pay BFL, in cash, for a number of units and they declined, due to lack of units. If they were running a scam, I suspect they'd have jumped at the chance to take my cash.
Anyway, on to the unit itself. It's larger than the pre-production unit I had access to before. In the pictures, you'll find listed below, I used an 80mm fan as a reference item, along with a US $100 bill. The unit is 4.25" square, and 2.5" tall, not including the tacked on bottom fan.
Let me talk about that bottom fan for a minute. I'm going to say right up front it's not pretty, and the cut out bottom part to of the housing to acommodate both the fan and extra heat sinks are a little ghetto and I suspect some of you aren't going to be pleased. That said, here's what I think. As I've stated before, I think BFL was unprepared for a number of things, first and foremost the rabid nature of online communities like Bitcointalk. They were also unprepared for something like bitcoin mining that does not follow the same power and usage curves found in pretty much all other FPGA applications of this type. They made estimations and promises based on base experience that would be accurate under just about any other circumstance. Does this relieve them of responsibility? No, but it does explain much of the process that has happened up until now and should be somewhat forgivable for basically pioneers in the field - since you can't really argue against the fact they have the fastest, nicest looking unit money can buy currently. Far outpacing any other FPGA offering.
Ok, so that brings us back to the case. They have pushed the deadline repeatedly and many of you have given them heaps of shit for it, perhaps some of it is justified, but I think much of it not. Regardless, I think what we see here in this first batch is them delivering the performance (revised) promised in a timely fashion. They could have pushed the date back again and retooled the board, case, etc... to be cleaner and you guys would have given them even more shit. Again, rightly or wrongly. In either case, for me personally, I would rather have an ugly box right now that is hashing than a pretty box 4 - 6 weeks from now; I can always order another prettty box then. So what you have here is a box and board that is rigged to provide the performance in stable configuration; it's not as pretty as it could be, but it works as advertised. I am satisfied with that.
That said, once we get past the tacked on HSF on the bottom, you could conceivably still stack these, I'm not so sure it would be a huge issue - the primary exhaust seems to be out the sides of the unit, thus stacking these is still going to draw the "cooler" air in from the top and bottom. Given the temps I saw, I don't think it would be an issue unless you're in a severely enclosed/hot environment. You could even increase the standoffs on the bottom fan by another inch and get that much more airflow.
The revision of the board sees quite a few changes to the power subsystems, which are positive. I don't have an ocsillicope here, so I can't measure the power, but I would imagine it's a lot cleaner, looking at the design. The power brick is also beefed up quite a bit, but it did get a bit warmer than I'd like to see, peaking out around 126F. An end user could probably supply a different power brick if it becomes an issue. The power connector is a standard barrel typ connector (3.5mm maybe? I didn't measure) - not sure if it's center positive or negative. I would also like to see a multi-unit power brick as an option. I can't imagine wanting to have 20 of these bricks sitting around at 125F each, but it's a small price to pay for replacing noisy GPUs.
There is basically one light on the unit to tell you it's on and one light inside to tell you it's hashing (actually two, one for each FPGA I think, but they both come on at the same time). Hahsrate is dead on target, using cgminer v2.2.3 I ran these tests in Windows, since it was the most convenient box for taking pictures and measuring power. It worked first time, with no drivers to install or anything. It was actually kind of amazing... absolutely zero headaches getting it going. It's only a couple steps away from a completely plug and play mining solution. One thing I did see, which I am a little confused about, but I have no addressed this with BFL as of yet - my observed effeciency was around 500%, sometimes much higher, but rarely much lower. So my ultimate question is, why is this unit so much more efficient at finding candidate nonces than a GPU? The shares are not being rejected by the server and appear to be valid, so they are legitimate - so I'm not sure what is going on there. So literally, this thing seems to find, on average, 5 valid nonces per getwork. One theory is it's using rollntime and because it's so fast in a single thread, it can actually get through several iterations before the time is out. Just a half baked theory and I have not had a chance to either investigate or enquire.
This is a solid hashing unit, it never hiccuped, got hot, fluctuated power or did anything bad, other than sit on my desk and spit out 830 MH/s for almost 36 hours. Here are some pictures of the unit in operation as well as a naked unit:
https://eclipsemc.com/bfl/flyingpigs/