Topic: Experimenting with Jalapeno firmware... - page 13. (Read 62632 times)

True, I have also upgraded units and I don't think any of them has both chips A grade.
Most likely binned just to reach 7GH.

I guess my problem is that I consider Josh to be a pathological liar, and automatically disbelieve any statement issued by, or in conjunction with, or regarding, the smelly swamp that is BFL that he utters.  

I said nothing about believing or not believing, there were only two quotes without any other comment from me, I don't understand what your problem is.

And you believe what Josh says because ...

I'm saying it is not A in upgraded unit...

No ? No......what? I am just relaying some quotes from Josh here.

It is the same as Jala... Only difference is that you need to change this part as well

/*************** Product Model *********************/
//#define __PRODUCT_MODEL_JALAPENO__
#define    __PRODUCT_MODEL_LITTLE_SINGLE__
//#define __PRODUCT_MODEL_SINGLE__
//#define __PRODUCT_MODEL_MINIRIG__

Change to:

/*************** Product Model *********************/
//#define __PRODUCT_MODEL_JALAPENO__
//#define    __PRODUCT_MODEL_LITTLE_SINGLE__
#define __PRODUCT_MODEL_SINGLE__
//#define __PRODUCT_MODEL_MINIRIG__

So has anyone flashed a single yet? Mine stays at almost exactly 60GH even when I turn off the AC and let the ambient temp rise. Makes me think it is locked and there is headroom.

It seems silly to me that the firmware needs to be upgraded using an external device and all. As if they were going to get the firmware 100% perfect on the first try leaving everyone else SOL. Not looking to push it but if I did I would have it well cooled. Let's face it these devices have a limited useful lifetime like no other device.

Just got my dragon today.
Did a flash based on the source code 1.2.5 compiled by me, speed went from 5.5GH/s to 7.6GH/s. That jally has a grade B and a grade C chip. HW error is at a alarming rate of ~8%... Will run it 24 hours to see the average then update here again.
Getinfo gives

No... I have B and D in it... So this is not the case... And had 4% HW error stock... But interesting fleshing did put them down to 2,5%... And give 700MH exstra... So it is 7,3...

And again a quote from Josh:
https://forums.butterflylabs.com/pre-sales-questions/4031-2gh-upgrade.html

The problem was indeed thermal with the single board single.
ckolivas had a play with it and managed to get cgminer to work ok with it, but it spent a % of time idle cooling down and lost a few GH/s due to that.

That's a good question -- up until now it seemed like everyone could flash to 8+ GHz, but now there's someone who couldn't get better than 7.  The problem is that upgrading to 7 doesn't ensure that you will be able to reflash to 8.

I have an order for two jallies (5GH/s) and was planning on flashing them, do you guys think upgrading to 7GH/s to get better chips would be worth it?

Odd numbers of chips do apparently work, the major objection to doing that is mechanical, it's hard to get even heatsink mounting pressure on odd number of chips. Back in the day when overclockers were pulling the heatspreaders off K6-2/K6-3 chips to get the last 50-100Mhz out of them, the solution to stabilise the sink and keep from crushing the core was either to cut a shim from plastic or to use high density foam "stick on feet" or similar on the corners to prevent wobble. Methods similar to that may work, but even so be careful mounting the sink.

Another point there, is that additional ASICs might not sit as conformally in the same plane as they would have done if installed at the factory, due to variation in mounting process, slight amount more solder etc. Even BFL is expecting some variation and is using the thermal pads to combat minor variance. This could be greater with chips you mount yourself or get a "reflow" tech to do for you. So inspecting fit very carefully before initial power up would be recommended.

I believe the thermals weren't looking so great cased up, and the power circuitry was a little strained to offer a little single product on that board. In a brief chat session with Josh, he seemed to indicate that it was a near thing that they didn't do the little single on that, and he thought it was capable of running 8 chips, and I guess with re-doing things for the big single and not being able to use the original jally board, it made more sense to up the jallies to those boards and do a one board solution for the others. There only seems to be a small fraction of little single orders compared to bigs and jallies.

Anyway, this is what Luke-Jr was originally running, the infamous "April 1st delivery" an 8 chip little single on that board.

Indications are that if it runs that single firmware, it will recognise more chips.

There was also an early running change from "hot mosfets" to a part with either a better spec, or that actually met the spec, so 2nd week onward jallies might be more capable of running 8 chips long term than the earlier ones. (That's a spitball of where the change was, might have been a tad earlier, but that's delivered in 2nd week, with 1/3 shipping plan some 3rd or 4th week orders might have got early production) However, we are not quite sure if anything else other than the ASICs are missing off the jally boards, like maybe some resistor packs or surface mount caps, if Luke-Jr is in possession of that original little he might be able to compare with a jally, but not sure if he is. Also not sure how much things are changing as production continued, if they had a stack of populated 8 chip boards they are using those first, but if it comes to using unpopulated boards, or re-ordering they might have more stuff left off.

So, actually several ways to go...

If you've got many jallies and are okay with straining the original PSUs, you might add just 2 chips and get somewhere between 12 and 16GH on them, this probably allows reuse of heatsinks also. May be capable of running cased with minor mods.

If you've got a medium number of jallies and can furnish 100W of PSU (Actually many older ATXes and AT PSUs will give you 10A on 12V so you should be able to find 100W for free practically) stick 4 more chips on for 6 total... Now, if you've got grade A chips on there, you might be able to run uncased on original heatsink with upgraded fan, if on the aluminum sink, but the copper/heatpipe sinks that were "used up" early on may do better. You should see over 20GH on these, if you have a good sink and PSU you may be able to push them to 30GH BUT if trying to get them high clocked use a IR thermometer on those mosfets and make sure they don't get too hot.

If you've got a handful and get a lot of chips, you could go for broke and fill all 8 pads, you would need an excellent heatsink, probably a minimum 150W PSU, maybe more depending on how inefficient the mosfets are getting by this time. Try to be gentle first power up flash it to low speed (Best done in advance). Here's where I think YMMV greatly because I think they blew a lot of boards in the lab before they got a handle on what was happening. Stick a killawatt on it or an amp clamp on the DC and kick it up one speed level, and see how much the watts/amps jump, if they jump 50W at that point, I think you're gonna have to stop there and be happy with what you get which might at that point be on the low side of 30GH. Also check temperature of mosfets... if you get lucky with your chips, you might get to kick it up one speed step at a time until it's mid 30s. If you get really lucky and you like living dangerously, you might get 40GH out of it. With 8 chips I think you're going to have to hand tune every one, keep a close eye on cooling, upgrade it as much as possible and generally "baby" it somewhat. Large scale miners might not want to bother with 8 chips. In my estimation the power usage per speed level is gonna be somewhat exponential, my best guess is that you want to stop at 200W and expect double the last increase... meaning if the last bump was 30W and took you to 190, then the next might be more like 60 and take you to 250W, which could be magic smoke liberation time.


This is all best guess based on following development closely and needs to be proved or amended by experiment.

EDIT: Oh another point to note, keep an eye on that barrel connector at power levels over 100W make sure you use a barrel connector that fits perfect, since there might be several that are "close" otherwise it's going to spark/melt or weld itself together. You may want to add additional or upgraded power connectors. A possible deal for Canadians, these http://www.sunforceproducts.com/product_details.php?PRODUCT_ID=148 have been on clearance at local Canadian tires here, something like $1.99 a set, the polarised 12V connector there, 3rd from left on pic, is actually on the end of all 7 of those leads, so you get 8 of them in there. Known as an "SAE connector". Not quite sure what max is on that, but I'd stick 2 on an 8 chip just to be sure.

That's great.Ive done a lot of BGA rework so this chip looks  easy compared to ps3 gpu. Was there any reason way BFL didn't populate the rest chips? Could you tell me what else need to be done to get additional chips going? Will the firmware recognize the chips automatically? Also any idea on any blf chips being delivered to customers?

Damn, I wish there was some sort of "one-click." method for this that went through the USB. I swear every single time I try to do something like this I end up failing miserably. I always have horrible luck, especially when there are guides and everyone else seems to have no issue.

My Jalapeno is running at 5650 MH/s and although that's fairly awesome seeing as I though I was getting 4500, I still would love to have 7000 - 9000. I doubt I will be doing it though, maybe after I make ROI or if/when the difficulty is high enough that I don't feel bad having my miner not running (I know someone in here said 0.12 BTC was nothing, but I'm running at 0.09 per day, and at 9+ dollars a day that's quite a bit, haha.)

Isn't it an awful feeling to feel as though no matter what you do, you risk a tonne of money and time? lol.

I'm definitely saving this thread, it was awesome to read though.

 

  yes, you can add more but not too much  and not easy job to put BGA chip on board.