Show Posts
|
Pages: [1] 2 3 »
|
Apologies, I've been meaning to post this for a while but have had little free time. Yes, you can cross-compile and remove the fan checks. Below I think misses some required bits (specifically, installing the _linux toolchain from below, setting environment variables and installing the x86 compatibility stuff) but might set you on the right track. If I get time, I'll rerun through this from scratch. You need: bmminer-mix gcc-linaro-arm-linux-gnueabihf-4.7-2012.11-20121123_linux.tar.bz2 gcc-linaro-arm-linux-gnueabihf-4.7-2012.11-20121123_runtime.tar.bz2 zlib-1.2.11.tar.gz (Google to find) git clone https://github.com/Zwilla/bmminer-cgminer492apt-get install build-essential cd bmminer-cgminer492/ chmod 755 autogen.sh apt-get install autogen apt-get install autoconf apt-get install libtool autoconf automake ./autogen.sh CC=/usr/bin/arm-angstrom-linux-gnueabi-gcc NM=/usr/bin/arm-angstrom-linux-gnueabi-gcc-nm AR=/usr/bin/arm-angstrom-linux-gnueabi-ar --enable-bitmain-c5 --with-system-jansson --host=x86_64-pc-linux-gnu make
|
|
|
Hello
Unnecessary. Nothing was changed at the BMMiner. Greeting
Interesting... In which case, can you supply the source for the modified cgminer which you are running instead of bmminer? (There appears to be no way to do per-chain settings external to the bmminer binary...) Cheers, Allan.
|
|
|
Voltage is controlled inside the bmminer code, as is fan speed. I don't see any active way to modify them short of a recompile of bmminer, which isn't too hard.
However, you can skip all this downgrading shenanigans if you want, since the rootfs is ubi and mounted rw. --fixed-freq is the required command line for bmminer, which is called by single-board-test (statically in the binary). If you create a shell script (/usr/bin/bmminer) that calls the original bmminer executable, it will be run and you can intercept and change the parameters.
|
|
|
Just in case anyone is wondering, yes, it is possible to reflow a whole board as a last ditch attempt for boards where the source of the fault is extremely tricksy. I don't have a proper oven which can cope with the maaaaaasive thermal mass of all of the heatsinks, so this puts the board *way* outside of the reflow parameters (it takes about 15 minutes to get up to reflow). ... Take off the electrolytics and the power connectors, clean board (I use isopropyl dip), apply thinned down flux to the topside only, ensure that you are heating on a solid surface. On a completely unrelated note, the black thermal adhesive copes remarkably well with heat. The white stuff... Not so much. Completely degrades and a slight touch will knock off the heatsink. So, I now have a completely bare of heatsinks board and am desperately trying to think of a way to properly heatsink it. I may just end up oil bathing the damn thing (it's a 54 chip board, so there are no non-SMD electrolytics). If you leave the connectors on, they will survive (foil wrapped) but look unhealthily yellowed. Subtext: Don't do this, there are *far* better ways to spend your money and time.
|
|
|
It's working on R4. Well i'm willing to experiment,is that your problem?
Simple explanation on PS questions would help.
Thx
R4 is a completely different design with a different type of fan. S5 controller/software doesnt support S7 boards, simple as that. Firmware for the s9 is closed source, there is no way to enable an extra board. BMMiner is open sourced. The only closed bit is the bitstream for the FPGA. https://github.com/bitmaintech/bmminer-mixCompilation isn't too hard and the source isn't terrible to read. Just make sure you have the 32 bit compatibility stuff installed for your toolchain and zlib if I recall.
|
|
|
I suppose if they are series(as logic ICs should NEVER be) and they are shorting to ground you're going to get a lot of burnt connectors and I am seeing a lot of those here and on eBay. But slowly going out one by one that has to be completely unrelated to power delivery, unless it's the power supply is sending noisy power then there is no causal relationship between the first going out and any others, they are just all the symptom of the one continuous PWS malfunction. One other option , and that one is causal , when the first chip goes out it doesn't truly die it just goes insane, sending spurious signals throughout the whole power bus killing the others one by one (haha, I should call this the zombie theory ). Since you think I should do some more research maybe you could point me to the data sheets or maybe the schematics, lol thats funny.
You watch ipadrehab and you want schematics. They are not public - not like Mac not public, like "don't exist outside of Bitmain for >S4" not public. The chain is very real and very much how it's implemented. I am unsure of how much University education is worth given the above statements. Research and learn first. Buy your own kit and repair it before offering a "no-win no fee" service which will end in tears. I would seriously suggest not sending boards to people who don't have a proven track record with antminers or at the very least modern ASIC miners.
|
|
|
240 / 220 is almost always slightly more efficient, since you're drawing less current (which means less resistive losses in the wiring). Although this effect is minuscule. It's also far easier and cheaper to get PSUs that deliver the load with 240.
It is technically possible that your PSU can change your mining rate - if you're overclocking and have one of the really old S7s. Apart from that, unless the PSU gets wobbly (way less than 12v output), there's no difference to hash rates.
|
|
|
Filters will most likely reduce airflow and heat up the miner. If they are in your house, try keeping the house clean and you wont have enough dust build up to worry about it.
ouch... I'll tell that to the wife. In all seriousness though - they're in a fairly high traffic route (hallway) so they don't get filthy. its maybe every couple of months i de_dust them, i just figured having a filter would help - when they arrive i'll do some monitoring for a week and see if they have any negative effects. Well, if you have a wife that allows antminers in the living space, I'm sure she will react just fine Flat fan filters - not recommended, especially if you have any kind of pets. You want the filter to be much bigger than the fan inlet if you can do it - box filter, etc. Good luck and let us know how it works out for you.
|
|
|
Hence server power supplies being a great idea - the breakouts or even ebay 'kits' are quite cheap, relatively speaking.
They are designed for 100% load and tend to be just generally nicer than normal retail PSUs.
Very variable on efficiency, though - monsters like the IBM 2880w one are not terribly efficient and louder than an antminer S7 - shop around!
Cheers!
|
|
|
That looks about right. i7z helps you visualise if the 'correct' cores are unloaded. You don't need to split into two processes - affining the process to the cores works fine.
I get ~630h/s from one node of 2 e5-2650l v1s. Not very power efficient, but with rising prices....
|
|
|
You don't need that much cold air in the winter. Recirculate most, vent some. Stay above dew point for your humidity and you'll be fine.
Once you heat the air, it's capacity to hold water rises - so you don't get condensation. It's a bit trickier to actually calculate due to the high static pressure inside the miner body, but just keep away from the high differentials and you'll be fine.
|
|
|
Looks to me like you may have a dual-rail PSU and you are not splitting the cards along rail boundaries. Means that one side of the PSU is providing all of the current and the other is sitting almost idle. Rail 1 - Card 1 + Card 2 Rail 2 - Card 3 + Controller + fans. Other possibility is that this is an older PSU - PCIe power connectors (especially the cheap ones) are not rated for many insert cycles. Or, as said, dust. If the connectors have a layer of dust on before they are plugged in, it's not copper -> tin -> Copper, it's copper -> tin -> random non-conductive stuff -> tin -> copper. Not good. Good luck. If you want, you can forgo this pain and solder the PSU directly to the board - the tracks are quite thick though so use your big tip.
|
|
|
Monero is for GFX now
Wow. That took you so much cognition. No, it's not, if you have hardware sitting otherwise 'idle'. Besides, if you have a GPU, you're better off mining pretty much any other coin (above SM_3.0 capable cards only - Cryptonight is still worthwhile on older cards - 770s, etc) 9 kH/s gets me 0.0033 BTC / day payout on Nicehash which is not insignificant. Make sure you arrange your cores correctly to use the cache for each CPU package - Linux arranges them as (4 cores, 8 threads) Package 1 | Package 2 0 1 2 3 | 4 5 6 7 (HT1) 8 9 10 11 | 12 13 14 15 (HT2) So as an example if you need 6 threads per CPU package to fill your cache, fill all of one HT set on both package, then fill it out with additional HT cores. That would be affinity set to (Package 1) 0,1,2,3, (Package 2) 4,5,6,7, (Package 1) 8,9, (Package 2) 12,13 If you are only mining with the box, you get better heat efficiency by turning off cores in the BIOS + using larger threads (above assumes xmr-stak-cpu) Cheers!
|
|
|
I get the feeling there _might_ be a small issue with the stats. Not that I'm complaining, mind you.
Next payout 2017-10-11 11:03 Payout amount 0.67092707 BTC 2467.63 GBP Daily income 1.10019024 BTC 4046.43 GBP Weekly income 7.70133168 BTC 28325.01 GBP Monthly income 33.00570720 BTC 121392.92 GBP
|
|
|
The chips are laid out in banks of three and there are two heartbeat signals one going beginning to end, the other end to beginning. Its actually pretty easy to chase down the problem with a decent multimeter and some patience. You can home in on the bank of chips that is having the problem and then do some resistance and other tests to further pinpoint the issue. The s9 is VERY similar in its layout and troubleshooting can be done in much the same way using the diagnostic points that are located all over the board. Just be VERY careful not to short a test point to a heatsink or you can kill the whole board rather easily.
Heh. I sense the voice of experience. The other thing which is obvious, depending on your environment, is remove the boards and have a look. I have one board that is unusable - it looks like somehow moisture got in at the fan entry side (possibly a very quickly deceased large bug) and rotted half of the tracks around the "middle" of the chain. If I baby the voltages and speeds, I can get half that board working - but it certainly isn't a DIY effort. These are tiny chips and pretty much every connection needs to work for the board as a whole to work. Send it to a repair shop and you'll have a far more consistent experience than probing with a multimeter (not that I'm saying that Fanatic isn't right - it absolutely is possible to diagnose if you have the time and no perceived value to the boards)
|
|
|
I'd like to learn to fix these myself, is there a good thread around here that covers most issues with these boards? There is ZERO info as to what typically goes wrong. I too would give my eye-teeth to get a handle on it. The only general symptom usually is the Vcore regulator shutting down (red LED light on board near the PCIE sockets not lit). As to *why* -- unknown. Either the regulator itself fails or a chip or bypass cap shorts out shutting it down. This is ALL based on the S7 stuff, so take it with a grain or 100 of salt. I don't believe that, if the s9 is the same, the LED is the regulator. In the S7, the two LEDs are attached to the last chip in the chain's busy / ready lines. These indicate (along with the correct power draw) that at least some data is getting to the end chip. It does *not* mean that the hashes from the chips (or anything, really) is getting back. I don't have any S9 boards to play with, but on the S7 you can check the 14.5v boost converter, which is (due to it being the coldest part on the board) subject to condensation. It's really easy to replace with an ebay special. I have heard people saying that this boost is only to support the end of the chain, but it doesn't seem to - the whole IO chain goes close to 0v if it's not outputting. In terms of the IO regulation, the 14.5v is split over the entire board (3 chip sections) by a resistive divider. This then feeds the approximate IO voltages into linear regulators, which referenced to the ground on each chip appear to provide the *return* (downstream) IO voltage. Easiest diagnostics to do on the S7s is to play follow the clock. You'll need a scope for that and it's an art rather than a science - sometimes, trouble later on down the chain will disrupt the startup of the first chip, so nothing is seen consistently anywhere. No detection but I2C works - shows as 48 ASICs and shows 'timeouts' on the syslog page. Temps are shown. CAN sometimes mine in this situation but be careful - the section of the chain that is not getting results back seems to overheat (I guess it only gets fed a new block to work on when the rest of the string does and busy-waits till then) Nothing at all - no I2C so can't detect a board is there If someone would like to donate a S9 control board and a dead S9 board, I'll have a look, but I can't afford to do anything which costs money right now. Good luck! Cheers, Allan.
|
|
|
He fixes now tens of power inverters (surface mount) from solar farms in EU, damaged by lightning.
He can buy all boards requiring repair or the whole miners, repair them in advance to offer prompt replacement on board 4 board basis.
Not sure if it makes sense to pay 100 euro or more for repair if the price of new miners falls every few months by half.
Sounds interesting. Just be aware that the chips are 0.4mm pitch MLFs with glued on heatsinks and no well done self-centering pads. If the repairer only has experience in power devices hot air / IR reflow, I would suspect he is going to struggle to get the accuracy and paste requirements down for these chips. Good luck though!
|
|
|
|