Topic: 280 BTC total bets between Micon and mrb (are BFL ASICs real?) - page 4. (Read 46415 times)

I will evaluate the current reports then in all likelihood send the codes for the winning bets.

Still a handful at 350+ we can work on a settlement or leave it and evaluate it at the appropriate date.

To be technical, it was just "who post pix and verify hashrate", not that they needed to verify power usage with pictures.  I think it's pretty clear that BFL has shipped a number of units meeting the requirements of the bet.

Good eyes spotting gigavps' report. Since my 200 BTC bet with Micon was on the same terms, I seem to have won too

In all fairness to Micon, it is unclear whether gigavps measured 30 Watt himself or not. Personally I am totally fine waiting for a few more reports before claiming my 200 BTC from Micon.

Congrats on your win, always nice to see money taken away from Micon

Micon, I believe I have won our bet!

The text that we bet upon:

grnbrg - reports 4.0 GH/s @ 26 watts = 153 MH/J
https://forums.butterflylabs.com/hardware-discussion/1983-grnbrgs-unboxing-teardown-new-bitforce-5-gh-s-sc-miner.html

Codinginmysleep - reports 5.5 GH/s @ 30 watts = 183 MH/J
http://codinginmysleep.com/bfl-jalapeno-unboxing-and-demo/

Gigavps - reports 5 GH/s @ 30 watts = 1.66 MH/J
https://bitcointalksearch.org/topic/m.1962352

Those are three customers who have posted pictures (or video) to verify hashrate, and as far as I know, none of them are employed by BFL or have been to BFL labs personally.

If you agree, please send me the escrow code.

I added another user (hdsolar) who reported receiving his Jalapeno. That's 3 users total so far.

According to Josh in the shoutbox, the 5GH/s Jalapenos (other than grnbrg's 4GH/s unit) are running at reduced speeds with two ASICs mounted. While they obviously haven't posted what voltages they're running at, I would imagine that running at 1/3rd of the original frequency they've already extracted those gains. We'll see what the next rev of boards brings, but they've been tweaking these boards already for over a month trying to get power down. There may be some gains to be had, but that is still a big gap. Also note that you'd need more of a drop from the ASICs as the draw of some things (like fans, the FPGA and MCU, etc.) would not be affected by dropping core voltage.

An individual user might be able to hard mod the voltage down enough just at the edge of stability to get it that much lower, but that's outside the bet.

I'm not hopeful at all there. They were reporting 1.76J/GH on their good example of a die from the wire bonded test, and that wasn't at full speed or all the engines running. Lower temps from not running full out will make that seem better than it is. Even then, add in a 90% efficiency going from 12V to 1.2V, and you're at best case 2J/GH just for the chips as a best case. While it's possible, it would require an almost shocking return to competency from BFL, and they've hardly been all about exceeding expectations.

Anyway, that's not really the point. There was 100BTC added to that bet making it ~140 for and 100 against back when BFL was still claiming ~1J/GH. If you were betting against BFL back when they were supposed to ship in a month and there was a 7 month buffer until the deadline, why wouldn't you double down now that they've gone from having to blow their specs by more than a factor of 3 in order to lose the bet to having to increase efficiency by 50% in the next couple months to win it?

It is not a slam dunk at this point because:
- my bet specifies efficiency at the DC input
- enmaku's 191 Mhash/J at the wall (30W) equates to 238 Mhash/J at the DC input (24W), assuming the power brick has an average efficiency of 80%
- 350 Mhash/J would necessitate a power consumption for 5720 Mhash/s below 16.3W (or 67.9% of 24W)
- reducing power consumption to 67.9% of its current value is trivial to achieve in theory by undervolting the chips to 82.4% of the nominal voltage because efficiency is (edit: inversely) proportional to its square (0.824^2 = 0.679)

So, in theory, BFL could release a rev2 of the Jalapeno any time in the next 2 months by undervolting it by 82.4% (eg. from 1.300V to 1.071V), and attaining 350 Mhash/J.

Will they do it? I don't know. But it is possible. The bet is still very much up-in-the-air IMHO.

I'm surprised that more people aren't taking the No position in your BoB bet. It seems as though that should be almost a slam dunk at this point, you'd be getting almost 1:1 odds that BFL won't improve efficiency by 2x. Trust in BoB must be really low after the BS over the last bet.

I added this to the OP:

As of Apr 25, so far, here are the potential datapoints that could determine the bet winners:

User	Mhash/Joule	Proof
enmaku	191 (at the wall)	5-second average of 5720 Mhash/s while measuring 30 Watt
grnbrg	156 (at the wall)	2-second average of 4065 Mhash/s while measuring 26 Watt

I suppose grnbrg's doesn't really count as an end customer, since he didn't order the unit he received?
https://forums.butterflylabs.com/hardware-discussion/1983-grnbrgs-unboxing-teardown-new-bitforce-5-gh-s-sc-miner.html

Good luck mrb, rootin' for ya

I trust Dave, and love his codinginmysleep blog.  I think he's super smart and not a "Luke-Jr" in that i think he is altruist and cannot be corrupted.

That being said Dave has very close ties to BFL.  He says at the end of this video that he is going to take the thing apart.  I anxiously await that vid.  And to see if this thing doesn't melt his desk.

If we see a few more of these start to arrive with other trusted organizations (bitcoin foundation, Jeff G, Gav) and get similar reports then I will stop wagering on them failing.  If another month goes by without another verified unit then I'm still confident.  

Certainly an uptick for the Pro-BFL bettors.  Looks like their power draw isn't close to what was promised, and it looks like the latest round of bets, if mass shipment is made, will be relatively close on the power rating front.

I heard that the two developers they shipped to were customers, but certainly that should be verified.

Yep, ~165 Mhash/J at the wall. Assuming the power adapter is 80% efficient this is ~210 Mhash/J at 12V.

It seems that Micon will win our small bet (80 BTC), whereas I will win our big bet (200 BTC).

Also, we don't know if the reviewer's single was one with the old, or redesigned (more efficient) power circuitry.

http://codinginmysleep.com/bfl-jalapeno-unboxing-and-demo/

We may have one down with two to go. I'm not sure If David paid for his unit so he may not qualify as a customer. I'll try to get that info from him. If he's been been to BFL, let me know. BTCguild shows 5.5 gh/s avg but David states it is fluctuating from 4.5 to 5.5 on bfgminer so we can avg at 5gh/s. Also this is measured with killawatt after powerbrick so draw should actually be lower. In any case, 5 gh/s at 30.2 brings us to:

165.56 mh/j

Thanks!

MrTeal and I settled our bet, he sent me his code & I collected the 75 coins and sent him the 5 coin settlement as promised.



at the jack, not the wall no problem.

It is a good point that the point of measurement wasn't specified in the bet between Micon and myself as well.  I think it is reasonable to disclude the power supply inefficiencies.  Micon, agree that measurement at the jack and not at the wall is to be used for our bet as well then?

no problems.  Measured at the jack is fine.  I do not anticipate any issues.



This is booked, I will send coins to the hash in the next 12 hrs.  
[edit:  sent]