Topic: Experimenting with Jalapeno firmware... - page 3. (Read 62632 times)

Well, I definitely understand how chips can be different heights, but in theory, the mounting process should be uniform enough to make the variances stay within a level of tolerance that'd be filled in by design of thermal paste. Pads are extremely poor conductors compared to grease, and my Jalapeno had visibly poor cooling using its stock pads - there were discolorations on the chips from the hot spots! That's why I went with the grease.

The use of an (unlaminated) metal plate on the bottom seems like amateur design suicide... yes, it'll conduct a tiny amount of heat, but at the extreme risk of scraping off insulation of the PCB and shorting out contacts over time. There's no thermal conductive pad there (which itself would be a good idea as well), so it seems entirely unlikely the plate serves any real purpose for cooling.

I like your method of tightening the screws, lightfoot... that sounds like a good plan. Huge, huge kudos for getting 4 chips temporarily running as well. Sounds like something I might try doing myself as well, if I can get the equipment together; I've got some chip credits to burn.

edit:
Disassembled it again and recycled the grease that slopped over the sides - something I'm sure a lot of people would argue with. Then, lightly reassembled using the light press and finger-tightening with the torx bit, and also replaced the cardboard with a thinner thermal conductive pad covering both chips' positions on the board. Sure enough, the plate does get plenty warm too. Back to normal temps again and rolling at 8.6GH/s again, after hashing all day at 7.8GH/s from last night's attempt. Cool deal! I very much believe thermal transfer has a lot more to do with speed than just crapping out at 70C or 80C.

Yep. My method here is to start the screws, then flip the jally over and press on the bottom plate to take up slack evenly on the chips. Then hand tighten the screws down with a torx bit, stopping when they are *just* snug. That way I am not cranking into the board and I know the chips are getting finger pressure.

Remember I am now running three chips (was 4, soon to be six) so my temps are going to be high. Yes this will lower the life expectency of the unit, but after Feb it's not going to be able to mine enough coin to matter, so it doesn't matter. Might as well go for broke :-)

It's possible that is what RUN HEAVY DIAGNOSTICS is. It's also possible that last night I was flashing it without the fan which caused the board to heat up enough to trip the "I'm too hot" sensor. Because it started working again after I went to hit the can with it unplugged. I'll try more tonight, but I really should just hard-wire the damn jtag port so I'm not always pulling the sink.

Running it with bad cores disabled cuts power usage (I think bad cores run 100% *on* and suck full power for zero benefit and take work thus slowing the overall chip down) so I'll re-flash again tonight. I think 1.2.6 might work with the newer chips, and ck has a compiled elf for that so I might give it a try. Or try Tarkin's load. Or just go back to a stock 1.2.9 with speed 7 and little_single. That worked well.

Anyway back to work.

When I was running with a friend's Jally I fixed till a few days ago, my 8gh normal speed, and my current speed at 12. Expect faster speeds "soon" once I get a new pad of that heat sink stuff and my other two chips.

C

By the way, that bottom plate helps transfer board heat to the bottom of the case. Don't put cardboard between that and board.

Ran all night, some errors on one of the new chips, bit odd. Eligus shows me at 11.93gh/hr over last 3 hours. That about matches the performance on the unit from BFG.

Still an annoying amount of errors on that one core, there has to be some mistake I am making in the firmware that is causing the unit to not properly clock the chips; CK's code runs the most nicely; maybe I need to try his 1.2.6 code run and see if it works.

Meantime, time to go to workies!

I don't think you should use paste with multiple chips, they will not all be at the same height, best to use those thermal pads like the ones that come with the Jally in the first place.

God damn it... well, I had 8.5-8.6gh/s, but now after fucking with the grease again thinking I had air pockets in the last application, this "last time I can screw with it before buying more paste" attempt resulted in a nearly-800mh/s reduction in performance - now running a solid 7.9gh/s average. Thermal problems are insanely understated because hardware usually compensates for most of the stupid crap people do, and it goes unnoticed. With the Jally, I think many hardware tweakers are gonna get a crash course in thermodynamics.

And this thing has painfully tight tolerances inside. Even adding that slice of cardboard spaced out the bottom plate such that the board now would flex a little if I tightened down the four corner mounting screws. So now it sits a little loose (and back to open-air again). Sigh... but this time, sleep takes priority.

Damn, you're as persistent as I am. Can't sleep with a problem. >.<

Yeah, with BGA as fucking insanely sensitive as it is, it's a goddamn miracle you got the new chips working. I was just looking at my Jally a few minutes ago as I was putting it back together (getting it back in the case, may be a bad idea) I noticed my (all-aluminum) heatsink had room to completely cover exactly 2 more chips. The heat sink design seems very "not fully thought-out" though, allowing WAY too much room for error in over-cranking the screws and cracking the fuck out of the board. I see a *lot* of people messing that up (and not assuming you have, either). Mechanically, over-tightening the screws wouldn't feel like over-tightening at all, but just barely going beyond the point where the screw head meets the board, and the board will begin to bow, causing the chips to bend away from the heatsink instead of pulling them tighter. Be really careful about that.

The fan control and monitoring in the firmware is a complete, unprecedented joke. There is no control at all - from what I've seen, there's a flag set in there with a code comment that almost literally says "this thing works better without control, so fuck it". The fan-control commands literally have no hook in the code at all, and no way to trigger it from the interface. Basically an initial control is ran to start the fan, doesn't even check to see if it runs, then it forgets it even has one. Really wish I could control mine. I have to feed it external 5V power (when the case is removed, of course) to keep my sanity.

With temps in the 60s and 70s though, you've got issues for sure. Mine maxes out around 55-60 right now with the case on and running the 8.5gh firmware, with freshly applied Ceramique II and screws lightly tightened just to remove the rotating wobble from the board/heatsink, and a thin (credit-card-thickness) slice of cardboard between the aluminum plate and PCB for insulation and padding (WTF was BFL thinking, metal-on-PCB?!).

Error checking shouldn't affect startup though. Maybe the code you're turning on is set to check all (non-existent) chips and craps out trying to test the ones that aren't there. /shrug... random shotgun guess without sifting through the code.

I really. Really. Wish I was a normal person and could just go to sleep with a problem.

Naah.

So I finally got up at 1:30 and took apart the Jally again. Looked at the chips and one was higher on balls. Not good. Tried re-melting it, unit fired up when hot, not when cold.

Bullseye.

Pulled the chip. Turns out it moved a bit and had some solder traces shorted. Would work sometimes, failed when I cranked down the heat sink. Now I'll have to learn to re-ball it.

Put things together, fired up Jally. Fired up with 3 chips, even without fan. Note to all: Fan speed is not monitored, total red herring. Cycled it 5 times, then put some heat sink grease on what's left of the pads, put together, fired up.

Running with some errors because I disabled error checking pretty much everywhere. Crap. But it is hashing at 12gh, which is damn slow for a jally. Temps are 75c, super high. So apart it is, lid off and flipped it and it's down to 67c.

Need a lower temp set here, those dead engines are pulling heat.

Tried reflashing with error checking on, got flashing light after test. Removed, back to normal. 12gh hashing at 65c with the case open; I am declaring victory for tonight and going to bed. Because I have to like... work tomorrow.

I'll put on two more chips when they come in this weekend.

C

Well, now it's consistiently dead. I've tried flashing ck's firmware (came up with 4 processors but flashed fast when rebooted), 1.2.9 with fan stuff commented out, 1.2.9 stock with little_single only, 1.2.9 with the changes recommended by Dana. Once it just came up by itself after a reprogram but silly me I powered it down to put it together and dud again.

At this point it can be one of three things:

1) I've broken the fan sense wire somehow and the Jally flash fails on startup.
2) One of the chips is shorting or something and flash failing
3) I have broken something else on the board.

What is bugging me is why it went bad; I must have power cycled it a dozen times with no problems while testing, then foom dead. Now the only things I can think of were:

1) I did snug down the bolts a bit more before closing it up the first time when it went bad; that might have caused a solder ball to go bad on one of the new chips.
2) I did have that fan thing, but I think that was before it started going totally bad.

It did run without errors for about 20-30 minutes there, it was running well.

So it's either try to re-heat the chips, pull them off one at a time, or pull diags somehow. I don't like the latter option; that really sucks as I will have to reball the things.

Any pointers on how I can use the USB port to pull diags? I tried plugging putty in at com6, but it didn't seem to listen.

Ah well, candle that burns brightly and all that.

On the info side, I got the chips to solder with a pre-heat of 120c on the board, followed by 400c for about 30 seconds per chip with the heat tool. Running at 280c didn't even melt the balls, even with 2 minutes of solid heat. Both look square on the board, I can see the edges with a 4x jewelers loupe, and one is a bit closer to the board than the other.

Any thoughts appreciated. I'll work on it more tomorrow after work.

C

Bedtime.

Houston we have explosion....

Crap. So after fiddling with it a bit I decided to put it in the case. Unplugged it, put it in the case, powered up.

Fast flash.

Crap. So I pulled it out and checked it. Remembered that when I put the fan plug in I heard a small crack, thought the fan plug had become delaminated from the board. No, looks good, but tried reflowing solder anyway.

Now it seems to work at random. For awhile I thought the issue was the fan cable, wiggling it then powering up would sometimes get all four chips up. Now I seem to have a dud.

Power supply is reading 13.6 volts, it seems good. However this thing is not happy. Question: What kind of tools can I use to query this? If it's on a com port can I use putty, and if so what kinds of things can I do to get to the diagnostic messages?

Crud.

C

Good work. Let us know.

*crackle*

Houston, we have a Jally with 15.67gh-16.04gh, error ratio of .76 or so, temp with unit open fan pointing down of between 54 and 61c. Power use is a pig; went from 20 watts with nothing to 40 idle to 95 hashing. So the extra two chips bump the jally from 40-75 watts draw.

Running un-modified 1.2.9 release, speed 7, little single.

Chips themselves: New chips are running at 4.2-4.3gh, older chips are at 3.5-3.76gh. New chips good, old chips suck.

No errors at all on the new chips, all errors on old chips. Moral: Old chips suck cock.

More later.

C

One way to find out....

C

More like are all the components there to support more chips correctly or have they been left off?

I dont think you could just add more chips to a jally. Can the board support that extra load?

Can you PM me your firmware too ? I got really bad luck and non of the binaries out there can work on my jala, neither my self-compiled one...

Gee you lashed out, I only went for th $8.95 programmer on eBay! My main worry will be identifying it to see its a compatible device. I like your i approach of using Linux rather than Atmel studio as that may not work with the drivers that are shipped with the device. I am more comfortable in Linux than windows anyway.

Oh, trust and believe, Arduino is a desperate last measure. ANY, and I mean literally ANY, programmer that labels itself as "JTAG" capable, will be able to program this thing with some tool out there, if I'm able to patch this crap-spree together to make it work. I've got a $20 JTAG programmer on its way from eBay too, and it'll be freeing my Arduino up for other projects as soon as it comes in. So I figure it's best to chronicle my adventures now before I hop, skip, and jump my way gleefully away from 3-hour programming durations