Topic: BFL BitForce SC Firmware source code - page 3. (Read 28001 times)

I would just like a good picture of it. I see Jal board as a problem and LS and S board as a solution... And I would like to be sure about the solution...

Jal use shortboard and LS and Single use doubleboard but LS use only half of doubleboard.

Little Single should be identical to jalapeno. Single is two of them.

Is there any picture of a single board out there? Or little single?

Thanks for being polite.

I saw that. Long term cryo is total bs when you need cost effective. Cryo unit itself is $7,500 to low of $2,500 if you buy $125M worth of them and only makes about 5W of cooling out of 100W of power. Compressor is cheaper, but doesn't get that cold.

Understood. 

I didn't mean to dis you; it's only that there was (is?) a hardware scam claiming cryo-cooling in production, which, after I stopped laughing, pissed me off as I'd read people who were actually buying into the notion.   

Sometimes it's hard to tell the AC from the DC, if you know what I'm saying 

I know how to handle liquid nitrogen on electronics and using it for a test wasn't a point of a long term use, but a bit of hyperbole to emphasize that I can get it very cold.

Anyway, I think chip's 61 Vcc bumps are limited to 200-250mA, so 12.2W-15.25W Max before chip starts to quickly loose it's longevity. Unless packaging house gave different number on mA per bump. Question is how many power bumps are on the die itself and what is their max mA spec? Because on packaging there is only 61 for Vcc ?

I would like to see if voltage could be dropped and frequency increased.

If performance is the cooling issue consider it solved, if it's chips design and materials used limitations then there is not much that could be done unless redone.

Just sayin', here:  IMO, you'll probably want to stick to water.  Liquid nitrogen in 24/7 could have huge infrastructure and support issues, is potentially hazardous at 100 below zero to both silicon and carbon based life forms, and can crack or kill components on a circuit board that has a near 200 degree trough to peak temperature profile.

And, re overclocking in general.  This isn't a game, where you want to get a little quicker action now and again.  You're planning for a business with components which will ideally have very short and infrequent down times.

I think 4 will not be a problem if you have new board, 6 might be but probably not, more I do not recommend. You need just to load last firmware to get 2GB upgrade... Overclocking is a bit more tricky... Some modification to firmware would be necessary since SetFrequency is not used. They determine best frequency at startup. Put it in cooler before start... That will "overclock" it

I guess I can make a water block for shortboard 5Gh board, but not sure if it will make financial scene in a long run with just 2 chips on it. Hay, anyone want to send me their miner to try to take accurate measurements in dimensions and temps, to see what else needs to be cooled?

 You can still mine on your own account from my lab, I just want to see how well it could be cooled. And if you want it back, I can pay for evening shipping and overnight 10:30am delivery from you and back so each way you loose less than a day.

But it maybe possible to overclock 5Gh or put more chips on it and WB will help a lot in all cases.

I am working on water block system to go on BFL long board. Packs units in a stack 16" deep

Now I am confused:

I thought that the full 50/60GH board has 16 chips? Edit: Ah, you were talking about shortboard
which mosfets are being used and to which are you planing to go? I want to see if they would be under powered for extreme cooling on the units that were already released.

I can have it ready in 2 weeks if I get exact board fitting.

Edit: Does long board also using PH7030L and PH3230S ?

I would be very interested to see a water block that covers chips, 1850 and mosfets.  I think it would change the nature of performance quite a bit and it's been something I've been itching to try, but haven't had any time to even start on some sort of project like that.

I have cooled the mosfets and 1850 with a block of ice and it does increase stability, but it was pretty ghetto, so doing a proper coverage WB would probably yield even better results.

If someone comes up with a full coverage WB for a short board and wants to send it to me, I can try it out on a 4 and 8 chip shortboard, although I'd need to swap out the mosfets on that board.  All the 8 chip short boards we have are a couple revs behind... hmm, I wonder if I can get a couple current rev short boards in with more chips. 

Is anyone working on a WB?

I can cool MOSFET too with water block or nitrogen, which once does BFL 50/60GH unit is using?

Josh said :

Some PC overclockers have suffered MOSFET damage when watercooling a CPU or GPU due to the MOSFETs being arranged to take advantage of residual airflow from the air cooling solution. So IMO, providing some airflow for the MOSFETs would be wise.

My current mental model of the frequency/power situation, looks a lot like this...



Where we might consider the BFL ASIC as somewhat equivalent to a multicore CPU, due to the large number of engines on the die, all working at once.

So expect a graph that's about the same shape as the blue line above and expect that as shipped they are running in the crook of that curve. Depending how steep that increase is, they may need double the power to run 100Mhz faster, or even just 50 if it's really steep... When initially tested an engine at a time, I suspect they were in a situation similar to the purple line at the bottom, everything looked in spec and fine and dandy and fairly linear out to 500Mhz plus.

Anyway, that points to it being plausible, that with highly efficient cooling, that you could get a little single on a longboard to a bit over 40GH by overclocking, if you happened to get the best grade of chips on it, but by then it would be sucking as much power as the 60, and thus you'd burn the MOSFETs if you went much further... and by much further, I don't mean a lot, because of the steepness of the curve you would be on there, it would be very easy to go too far...


(BTW to spitball the numbers there on the graph to something in the ballpark of bitforce ASICs, mentally divide the watts by 20 on the left and the frequency by 10 on the bottom)

Now how to overclock/crack jalapenos?
How to fix p2pool?

Well it wasnt working with 3.2.1 for some reason i cant remember. But when i switched to 3.2.2 it was mining right away without any errors (except for a bunch of hw errors).

Currently at 14829 accepted shares diff 4 and 2006 hw errors. Think that translates to 3.3% hw errors?

On overclocking 50/60GH units,
lets say I cool the chips with a water block or liquid nitrogen so that heat is not an issue anymore and change the frequency, would power regulators will be the only issue left?
What is the maximum wattage that system can deliver to 16 chips?
Could I get to 500MHz or more with very advanced cooling?

Does that work with cgminer?
It says 1.0.0 but the GetInfo output contains 1.2.x information.
I hope it really is acting like a 1.0.0 otherwise you'll get errors galore with the work queue replies.

Wow whole new terrain for me there. Had to install Java on my Ubuntu machine and enable API in cgminer. But i did manage to get the info.

(added some BR to make it readable)

So if im not mistaken my Jala has 2 full processors, but one is of better quality than the other?

If you blow your chips due to overclocking to ridiculous amounts or failure to cool your chips, there is no warranty.