Topic: Hacking a BFL Jalapeno to 32GH and beyond....(???) - page 19. (Read 54341 times)

Black Friday?!?! lol God no, that was from june 10th.... I think they may have been bluffing to get rid of the old stuff before it's useless. My order for BF is still processing (The first step) we'll see when they actually ship........4 months later. Why did I click that little box oh why?

Drug5bitz, were those 4 jallys ordered during the black friday sale? 

I'm just wondering where they are at with them.

I don't think many want to take others jalapenos into their hands lol. Scared of bricking it. Got my second one done 4 chips total and a nice setup for a bottom heat sink again all in depth pics taken.

That is AWESOME!  I have 2 Jallys coming that I would love to add chips to. I think they would be great as 4 chip units rather than 2!

We just need a service that will do chip adds through the mail...

OK, got my 4 jalapenos today!!!1!!11  Hitting 32 Ghs, aluminum heatsinks are cut away already out of the box. Added a chip perfectly in less than 2 minutes.  Align chip perfect after putting down flux (Hardest part), preheat boad with kitchen skillet to 400F hit it with the gun for a minute at 400C. Reflux, hit the chip and sides another time for one minute. Booted right up, lots of pics I shall share.

And another update:

This evening I got three more chips and tried to put a chip on the unit. I took off the bad chip, prepped the board, lined it up perfectly, and soldered it on. No problems, system fires up.

Fail.

Complete fail. That's insane. Rebooted it, it finds there chips.

Three chips that include chips 2,3,4 (4,5 are the original old chips, 2,3 have never been lit.)

That's the problem. One of the other chips I have put on must have a short on the select lines. As such it is thinking it's another chip and when I try to go beyond 5 it causes the unit to go nuts. The chips I have been putting on were not bad, one of the existing chips was.

I give up. Removed the chip, stopping everything for now. At this point this poor jally has been flowed a dozen times, has wrecked two chips, deballed two more, and is bannanas. I cleaned up the board, put it back together, promptly broke a fan blade *AARRGGHH*, broke a blade on the other side to pretty much balance it, put it together, and it's hashing at 20gh.

I'm going to wait till the new jallies I ordered come in, then put 2 chips on each. So I will leave it to someone else to go past 5 chips; if you can make a 6 chip jally do post.

This has been an excellent learning experience: I picked up a lot of knowledge, and will probably break even on the upgrade (since I spent for 6 chips and only got three working, $200 spent). Then again I boosted this thing more than a new jally, jally and a half, and it cost the same as a new unit. And it's here now. So it was still a pretty good deal.

Such is life. I'll post again when/if I get my new units, then I'll try putting these chips on. My guess is they will work without a problem, until I short the next CSEL line.

C

WARNING: THIS IS IMPORTANT:

Big big stopping point here. I think this is important:

I know people are flashing these things with all sorts of firmware. We know that the newer jallies with the 256mb flash chips need to be defined specifically in order to flash them. And we know that 1.2.5 and 1.2.9 work on the 256mb chips.

HOWEVER Has anyone successfully flashed 1.2.9 onto a jalapeno that has only 128mb chips?

Here is the reason: I got a Jally in for a chip upgrade. No I am not doing this for money, person asked really nicely and agreed to take all risks. Ok.

As is my protocol I change only one thing at a time. First I unboxed their unit and ran it last night. Says it's 1.2.6, kind of weird firmware but it hashes at 7.6gh overnight without overheating or anything.

Then I flashed it to 1.2.9. Failed flash with the ATMEL set to 256mb chip. Reset to 128mb chip, erased and flashed. However the Jally was *DEAD*. Totally, and completely dead. Power light, no front flashing LED, looked like a serial port. Nothing else, game over.

Flashed two more times, verified code, all good. Fails.

Flasked CK's 1.2.5 code release, Jally comes up happy with 15 cores. Now letting it run for another day to verify stable operation.

WHAT THIS MEANS!

If you have a 128mb Jally, I am not sure if you can load 1.2.8 or 1.2.9 code. Given that this code is needed for running the newer chips, it means YOU CANNOT LOAD NEWER CHIPS ON YOUR JALLY.

So, has anyone verified you can load 1.2.9 compiled code on a 128mb Jally?

C

Latest data from the testing frontier:

1) My chip 6 refuses to come up.Heated it to 500F using 465c air at speed low for 2 minutes. Note, lower air speeds mean more heat since the air is transferring heat *to* the object instead of *around* the object. Forgot about that.

2) I put the case back on the unit with the heat pipe heat sink (*NOT* the Al one) and the fan blowing *up*. I did grease the bottom of the sink when I put it on the base for bottom heat transfer.

Case itself gets warm, temps in 60-65C. So a fully enclosed jally with an early style heat sink can run in the case with 5 chips. The Al heat sink will *not* work.

Will leave it alone today and see if it's stable.

Update: after a day it's totally stable. Nice!

C

Makes sense, the technology is basically obsolete (but a nice first shot).

An idea: Take the remaining ones and send them to the chip credit people; that would flood the market, clear that agreement, and allow you to write off the remaining inventory.

C

(Has two chip credits somewhere around there)

Hm. Eligius has been noisy, and my secondary choice for hashing got some work done, so I can't trust the damn stats. But it does look like my performance boost to 8 didn't speed up my overall hashing rate.

I'll let it run through this morning. However going to speed 8 from 7 seems to have driven up my temps by 10c, my power by 10 watts, and my hashing by 300mh/s and possibly down. Might not be worth it.

Hm.

C

We only ordered 500,000 to 600,000 chips (I can't remember how many wafers specifically off the top of my head) and we're down to under 30k left, maybe quite a bit less.  Once they are gone, they are pretty much gone, we are moving to 28nm.

The raw data dumps on my speed runs:

Default: Speed 7 been running this for awhile now without a reboot. <1% errors.

DEVICE: BitFORCE SC
FIRMWARE: 1.2.9
IAR Executed: NO
CHIP PARALLELIZATION: YES @ 5
QUEUE DEPTH:40
PROCESSOR 0: 16 engines @ 271 MHz -- MAP: FFFF
PROCESSOR 3: 13 engines @ 265 MHz -- MAP: EFFC
PROCESSOR 5: 16 engines @ 248 MHz -- MAP: FFFF
PROCESSOR 6: 16 engines @ 264 MHz -- MAP: FFFF
PROCESSOR 7: 15 engines @ 248 MHz -- MAP: FFFE
THEORETICAL MAX: 19693 MH/s
ENGINES: 76
FREQUENCY: 274 MHz
CRITICAL TEMPERATURE: 0
TOTAL THERMAL CYCLES: 0
XLINK MODE: MASTER
XLINK PRESENT: NO
OK

Speed 8:
DEVICE: BitFORCE SC
FIRMWARE: 1.2.9
IAR Executed: NO
CHIP PARALLELIZATION: YES @ 5
QUEUE DEPTH:40
PROCESSOR 0: 16 engines @ 281 MHz -- MAP: FFFF
PROCESSOR 3: 13 engines @ 271 MHz -- MAP: EFFC
PROCESSOR 5: 16 engines @ 253 MHz -- MAP: FFFF
PROCESSOR 6: 15 engines @ 274 MHz -- MAP: F7FF
PROCESSOR 7: 15 engines @ 253 MHz -- MAP: FFFE
THEORETICAL MAX: 19972 MH/s
ENGINES: 75
FREQUENCY: 283 MHz
CRITICAL TEMPERATURE: 0
TOTAL THERMAL CYCLES: 0
XLINK MODE: MASTER
XLINK PRESENT: NO
OK

Speed 9
DEVICE: BitFORCE SC
FIRMWARE: 1.2.9
IAR Executed: NO
CHIP PARALLELIZATION: YES @ 5
QUEUE DEPTH:40
PROCESSOR 0: 14 engines @ 291 MHz -- MAP: FFCF
PROCESSOR 3: 8 engines @ 277 MHz -- MAP: E3A4
PROCESSOR 5: 16 engines @ 261 MHz -- MAP: FFFF
PROCESSOR 6: 13 engines @ 284 MHz -- MAP: D7FD
PROCESSOR 7: 15 engines @ 259 MHz -- MAP: FFFE
THEORETICAL MAX: 18043 MH/s
ENGINES: 66
FREQUENCY: 291 MHz
CRITICAL TEMPERATURE: 0
TOTAL THERMAL CYCLES: 0
XLINK MODE: MASTER
XLINK PRESENT: NO
OK

Also speed 9, ater a re power cycle.
DEVICE: BitFORCE SC
FIRMWARE: 1.2.9
IAR Executed: NO
CHIP PARALLELIZATION: YES @ 5
QUEUE DEPTH:40
PROCESSOR 0: 15 engines @ 288 MHz -- MAP: FFEF
PROCESSOR 3: 8 engines @ 277 MHz -- MAP: E3A4
PROCESSOR 5: 15 engines @ 262 MHz -- MAP: F7FF
PROCESSOR 6: 12 engines @ 284 MHz -- MAP: D7ED
PROCESSOR 7: 14 engines @ 261 MHz -- MAP: FF7E
THEORETICAL MAX: 17528 MH/s
ENGINES: 64
FREQUENCY: 291 MHz
CRITICAL TEMPERATURE: 0
TOTAL THERMAL CYCLES: 0
XLINK MODE: MASTER
XLINK PRESENT: NO
OK

How about speed 6?
DEVICE: BitFORCE SC
FIRMWARE: 1.2.9
IAR Executed: NO
CHIP PARALLELIZATION: YES @ 5
QUEUE DEPTH:40
PROCESSOR 0: 16 engines @ 268 MHz -- MAP: FFFF
PROCESSOR 3: 13 engines @ 259 MHz -- MAP: EFFC
PROCESSOR 5: 16 engines @ 240 MHz -- MAP: FFFF
PROCESSOR 6: 16 engines @ 256 MHz -- MAP: FFFF
PROCESSOR 7: 15 engines @ 240 MHz -- MAP: FFFE
THEORETICAL MAX: 19191 MH/s
ENGINES: 76
FREQUENCY: 266 MHz
CRITICAL TEMPERATURE: 0
TOTAL THERMAL CYCLES: 0
XLINK MODE: MASTER
XLINK PRESENT: NO
OK

Speed 6 again after a fast power cycle?
DEVICE: BitFORCE SC
FIRMWARE: 1.2.9
IAR Executed: NO
CHIP PARALLELIZATION: YES @ 5
QUEUE DEPTH:40
PROCESSOR 0: 15 engines @ 267 MHz -- MAP: FFDF
PROCESSOR 3: 13 engines @ 259 MHz -- MAP: EFFC
PROCESSOR 5: 15 engines @ 240 MHz -- MAP: FFDF
PROCESSOR 6: 15 engines @ 256 MHz -- MAP: FFBF
PROCESSOR 7: 15 engines @ 240 MHz -- MAP: FFFE
THEORETICAL MAX: 18412 MH/s
ENGINES: 73
FREQUENCY: 266 MHz
CRITICAL TEMPERATURE: 0
TOTAL THERMAL CYCLES: 0
XLINK MODE: MASTER
XLINK PRESENT: NO
OK

How about speed 5? Sure, why not.
DEVICE: BitFORCE SC
FIRMWARE: 1.2.9
IAR Executed: NO
CHIP PARALLELIZATION: YES @ 5
QUEUE DEPTH:40
PROCESSOR 0: 15 engines @ 259 MHz -- MAP: FFDF
PROCESSOR 3: 13 engines @ 250 MHz -- MAP: EFFC
PROCESSOR 5: 16 engines @ 237 MHz -- MAP: FFFF
PROCESSOR 6: 15 engines @ 250 MHz -- MAP: FFBF
PROCESSOR 7: 15 engines @ 235 MHz -- MAP: FFFE
THEORETICAL MAX: 18202 MH/s
ENGINES: 74
FREQUENCY: 260 MHz
CRITICAL TEMPERATURE: 0
TOTAL THERMAL CYCLES: 0
XLINK MODE: MASTER
XLINK PRESENT: NO
OK

Hm. Slower, back to 8
DEVICE: BitFORCE SC
FIRMWARE: 1.2.9
IAR Executed: NO
CHIP PARALLELIZATION: YES @ 5
QUEUE DEPTH:40
PROCESSOR 0: 16 engines @ 280 MHz -- MAP: FFFF
PROCESSOR 3: 13 engines @ 271 MHz -- MAP: EFFC
PROCESSOR 5: 15 engines @ 253 MHz -- MAP: FFDF
PROCESSOR 6: 16 engines @ 274 MHz -- MAP: FFFF
PROCESSOR 7: 15 engines @ 256 MHz -- MAP: FFFE
THEORETICAL MAX: 20022 MH/s
ENGINES: 75
FREQUENCY: 283 MHz
CRITICAL TEMPERATURE: 0
TOTAL THERMAL CYCLES: 0
XLINK MODE: MASTER
XLINK PRESENT: NO
OK

SWEET! Let's run with this.

Good luck everyone.

And by the way, some software thoughts:

Now that I have a heat sink that allows me to have the programming plug attached while the sink is on (oh joy!) I have had a chance to fiddle a lot more with the speeds in the 1.2.9 software. Was kind of surprised by the results. In a nutshell:

Speed 5: THEORETICAL MAX: 18202 MH/s
Speed 6: THEORETICAL MAX: 19191 MH/s
Speed 7: THEORETICAL MAX: 19693 MH/s
Speed 8: THEORETICAL MAX: 20022 MH/s
Speed 9: THEORETICAL MAX: 18043 MH/s
Speed 9: THEORETICAL MAX: 17528 MH/s (after a reboot)

The difference on speed 9 is that one of the chips (Processor 3, the crap processor) only fired 8 engines at 277 instead of 13 at 271. The other processors are more sad as well. Interesting, it seems it could not slow down with the others being fast. I'll post the specs below.

However I am now running at speed 8 instead of 7, will see how it hashes. Note also that letting the unit cool down after running will always produce slightly higher hash results, and unplugging/immediate plug in will cost about .5-1gh.

The moral here though is that maybe some of the Jally clockers who are seeing crap speeds with 1.2.9 are using Tarkin's code, which runs everything at 9. I think it will cause the weaker chips to run at super-slow speeds, while speed 8 will provide the possibility of coming up. Speed 7 seems to be the best balance of power and performance, but note that speed 6 is still pretty damned good. And speed 5 is still *BETTER* than speed 9.

Given the lower power usage and longer life, I would recommend that if you go past 5 chips, you might want to run speed 6. If you stuff 8 chips on go for speed 5. That might give you the most efficient power usage per chip.

For the moment, I'm going to run at speed 8 for awhile and see if it makes a difference in the software. Oddly enough I might be best off tweaking things manually, putting some chips at 9 and others at lower speeds (like the aforementioned chip 3) specifically to optimize performance. I think that's next :-)

C

In the meantime, I'm waiting on three chips to arrive, still thinking about reballing/replacing these two chips I have, and my jally continues to hash happily with 5 chips live (and the one dead one) onboard.

I couldn't refuse the BFL offer so I picked up a total of three jallies (one for a friend) and an ATX cord. I think I'm going to put one more chip on my board, and set the other four chips on the other two jallies. So I'll go to six chips.

What we need are more chips here, anyone got any thoughts/leads? I'm wondering if BFL locked out sales so they could ensure they had enough for their jallies and singles; apparently they sold out of JP's so there might be some left overs once the sale ends today.

C

Those look much better than mine; I actually really blew it and pulled some of the pads off the chip. I was running too cold.

Anyway, the one on the right might reflow if you put flux on the bottom, then heat it. Both might actually, give it a quick try but at lower temps like 350 or so.

C

So the one on the left I don't really remember how I screwed it up... think I got nervous and lifted it thinking no bond had occurred and well, you see the results, the one on the right slid unevenly when heated. I was using 438C @ medium air flow.

As for the Jally orders, my BTC were sent at 08:38am so it was still kinda early.

Naah. I ordered three. Two for myself, one for a friend mainly because I wanted that ATX cable thingie. When did you order (I got mine in at 12:01; web site was sloooooo).

As for reballing, can you post pics of your chips? And how did you screw up; my two screwups were because the chip moved off center and I panicked for one, and because I  hit the chip with the nozzle and caused a board short on the other. Getting better on this.

Actually maybe one of us should by the $70 BGA reballing kit on Ebay, then send it around to people.

C

So I was successful in adding a single chip to each of my two Jally's by following the most excellent pioneer work found in this thread.

I did however manage to screw up two of my chips in the process. I have the reflow equipment to reball, but don't know what size solder balls to order or if there's a good pre-existing template for the 12x12 array on the chip. Any ideas? I'd like to try to get my two bum chips going.

Call me a sucker, but I ordered 4 more Jally's using BFL's Black Friday Sale... guess we'll see if they're really ready for immediate delivery (the price for 4 after shipping and discount was less than $900)

Ah the lulz Ive read through that one already, abd that's how I'm here now.   Thanks for the heads up on the file I didn't dig through the code to much yet but will.