HashFast BabyJet users thread

[HF-Engineer]

I am here working at the DataCenter working on the BJ's. I have 12, want to put 2/case.

Where do I plug the motherboard power for the 2nd BJ into on the Power supply?

There is one port labeled "MB" on the power supply. It is already taken by the first BJ.

Thanks

On the BJ, there should be 2 sets of PCI-E cables coming out of the PSU, and one 5-pin control cable.  One of the sets of the PCI-E cables along with the 5-pin connector are already connected to the existing (top) board.  The other set is zip-tied up by the PSU.

You'll need a chaining cable (16 pin ribbon) and the screws/standoffs to add 2 boards to one BJ.

Once you have that, simply mount the second board, connect the chaining cable from the top to the bottom board (using the 2 closest 16 pin connectors), MOVE the 5-pin cable to the second board, and then connect the spare set of PCI-E cables to the new lower board.

-Phil

[1715494320]

1715494320

[1715494320]

1715494320

[1715494320]

1715494320

[HF-Engineer]

Hi Phil,
Yes, I compiled cgminer with --enable-hashfast on my xubuntu. It can detect HF device but can not enable the device, that's why I think it's permission issue.

You'll need to follow the instructions for enabling the udev rules if you want to run as a non-root user:

Create a file called "01-hashfast.rules" in the /etc/udev/rules.d/ directory containing the following lines:

Code:

ATTRS{idVendor}=="297c", ATTRS{idProduct}=="0001", MODE="0660", GROUP="plugdev"
ATTRS{idVendor}=="297c", ATTRS{idProduct}=="8001", MODE="0660", GROUP="plugdev"

You will have to be root or sudo'd to create the file.

-Phil

[legendroute]

Also Phil, I have two more questions:
1. I can see some errors in cgminer out put, how to fix them?
[image snipped]
As you can see the hardware error is keep going up, is this normal?
Also the first line of the scroll-able zone there is an error message:
HFA 0 attempted reset got err:(-5) LIBUSB_ERROR_NOT_FOUND 
How to fix it?

2. Second, with default configuration, from cgminer's interface I can see the hash rate is always around 400GH/s. But from pool statistics average is only between 320 to 380. How to explain that? Can I say my BJ does not perform as described in specification?
Thanks you!
[image snipped]

In both these cases, we hope the new firmware will help correct these issues.  It's absolutely normal to have some hardware errors.  We have fixed the source of some of these in the upcoming firmware.  Con Kolivas is also working with us to make many improvements in cgminer, so we will release both together as soon as we can.

You should be able to get about 400Gh/s at the pool (average) though you might have to clock to 600mhz to get it at present.

My test system at http://setup.hashfast.com/rpi/ is running at 600, it's got a bunch of errors, but as you can see from the pool stats on Eligius it's cruising along at about 428Gh/s for the 12 hour average:
http://eligius.st/~wizkid057/newstats/userstats.php/17ao2fT7gbnnKYpMd7x29E8p4oGdE87Ahd

The longer timeframe stats on Eligius tend to be more accurate and useful.  The stuff under 3 hours is hit-or-miss.

-Phil

Hi Phil,
I would like to OC to 600MHZ, but does that void the warranty? Anyway since the warranty is only 10 days(which does not make sense for such an expensive device), do you see at what chance I will need it since it is running smoothly for about a day already?

Or should I call the warranty under the reason device does not perform as described and let Hash Fast "FIX" it for me? If the official fix is to OC it, then I don't think that will void the on going warranty right(even it's not long)? 

[HF-Engineer]

Hi Phil,
I would like to OC to 600MHZ, but does that void the warranty? Anyway since the warranty is only 10 days(which does not make sense for such an expensive device), do you see at what chance I will need it since it is running smoothly for about a day already?

Or should I call the warranty under the reason device does not perform as described and let Hash Fast "FIX" it for me? If the official fix is to OC it, then I don't think that will void the on going warranty right(even it's not long)? 

I'm not allowed to comment on what constitutes overclocking and/or what particular events will void the warranty.  As far as I am aware, (my personal opinion alone, and not constituting legal advice or that of HashFast.) is that "Overclocking COULD void your warranty".

I can tell you from personal experience that I have seen zero boards damaged by anything you might be able to do externally via software.  If you don't physically mess with the board, I don't see how you can harm it.

Sorry, I wish I could give you better answers, but the lawyers will probably then tell me I have to stay out of the forums.

-Phil

[kenpofighta]

,

I was getting great performance when my 3 BJ's arrived. At least on two of them, there was one that seemed a bit flakey and seems to be getting worse. All 3 seem to be getting worse. One of them keeps shutting down and restarting grabbing a new ID each time. I posted a small capture from my monitor so that maybe if others have had this and fixed it they can help. This one that is flaky was delivered in poor condition, all but one screw was sliding around the inside of the box and 4 risers that prop the board up were also floating around. I though the board was damaged and it may be. Worst assembly I have ever seen and I used to build PC's. That point is mute if they all work but they don't.

Here is what I need help answering.
1. How do I get back to my 400+ GH performance. All three of these were pushing +-420GH.
2. Worse case scenario is how do I raise the clock speed to get up to the 400GH range? I was promised by Erin at Hashfast that it would not void the warranty but the delivery paperwork stated otherwise. I will have to deal with that warranty issue separately, I know.
3. I seem to be getting huge error rates, I think it is on the flaky board but not 100% sure that is the only place. 70% to 80% errors are common. Elgius reports I am only getting 900 to 950 GH/s AVG with 3 boards and I should be getting closer to 1200. Not good for profits. Any help would be appreciated. I hate to try and send a board back under warranty since I might be stuck for months. I still am waiting on my 3 upgrades, I would hat to be down to 2 processors if I can avoid it.

Sorry about the loose parts.  The problem wasn't bad assembly, it was the parts coming loose during shipping.  They are using threadlock now to stop that problem.   It's unlikely your boards are damaged if they are hashing on all 4 die, which sounds like they are.

What are your die temps and voltages?  (if you run the "hf" version of cgminer it reports all dies/voltages)

Since your unit obviously had quite a ride during shipping, I'd look at each die temp and see if there is a discrepancy.  You might need to take the cooler head off, re-grease the thermal interface, and re-tighten the cooler.  Apparently some of those earlier orders that had loose standoffs have a high chance of having a loose cooler.

Also, what hash clock speed are you using?   Have you tired other speeds?

-Phil

Thanks for your time Phil, I know this is on your own and a favor to the newly create Hashfast community.

By the way, I would love to try other speeds but I am not sure how to adjust them and its all pretty hush, hush it seems. Before I ordered I was told I would be able to overclock without voiding my warranty but now I am worried about doing it based on paperwork received. I may try it on the machine that is struggling anyway if someone will tell e how. I run the machine headless so there seems to be little access. If you can email me instructions to overclock that would be great. kenpofighta at gmail dot com.

Details from the struggling miner: (You can see it stop an id and start new ones throughout the day, error rates stay around 80%. Seems 2 cores are much cooler than the other 2)
 cgminer version 3.9.0h2 - Started: [2014-02-05 16:45:10]
                                              
 (5s):322.7G (avg):367.5Gh/s | A:1038456  R:7040  HW:13622629  WU:5999.8/m                                                      
 ST: 2  SS: 0  NB: 38  LW: 1772615  GF: 0  RF: 0                                                                                
 Connected to stratum.mining.eligius.st diff 128 with stratum as user 1Bk9VSwaXn9UhpdowTwHEPEWw95i5F7N2_Miner2                  
 Block: c019bc52...  Diff:2.62G  Started: [22:23:14]  Best share: 699K                                                          
                                              
 [P]ool management Settings [D]isplay options [Q]uit                                                                          
 HFA 2: 61C/.23V796C/.23V1429C/.26V661C/.23V OFF   /40.21Gh/s | A: 96447 R:1472 HW: 1362431 WU:  650.5/m                        
 HFA 4:  75C/.80V 77C/.80V 62C/.82V 61C/.82V 368.7G/369.1Gh/s | A:810420 R:4800 HW:10748045 WU: 6021.6/m    

Here are stats from Eligius (I boke them down by naming the miners so that I could better analyse the data about 4 hours ago, Miner 2 is the struggling miner)
Miner2   3 Hours      214.79 Gh/s   540096   
           22.5 Minutes   196.28 Gh/s   61696   
Miner1   3 Hours      400.76 Gh/s   1007744   
           22.5 Minutes   416.19 Gh/s   130816   
Miner3   3 Hours      387.32 Gh/s   973952   
           22.5 Minutes   374.65 Gh/s   117760

I have the stats for the other two below for comparison. These units all sit side by side in the same room and run about the same regardless of how cool I keep the room.

I appreciate that the loose parts may not have been purely an assembly issue but I have never seen a shipped piece of hardware in this condition after being shipped. I hope the threadlock will stop the issue going forward. I am hoping what you say is correct and that the issue is due to loose parts and nothing more.

I am a software architect by profession and eventually I will port some of these features to Windows, I know there are many haters but it is a very stable platform, I am running Windows Server 2012 and can spin up multiple VM's to handle mining as I grow my operation. I can post on how that goes as I get to it.


 cgminer version 3.9.0h2 - Started: [2014-02-05 16:45:32]
                                      
 (5s):440.3G (avg):421.4Gh/s | A:1900827  R:17408  HW:7210  WU:11114.1/m                                                        
 ST: 2  SS: 0  NB: 38  LW: 2047136  GF: 0  RF: 0                                                                                
 Connected to stratum.mining.eligius.st diff 256 with stratum as user 1Bk9VSwaXn9UhpdowTwHEPEWw95i5F7N2_Miner1                  
 Block: 12f30e18...  Diff:2.62G  Started: [22:31:53]  Best share: 551K                                                          
                                            
 [P]ool management Settings [D]isplay options [Q]uit                                                                          
 HFA 0: 77C/.80V 82C/.80V 73C/.79V 72C/.79V 425.9G/421.4Gh/s | A:1899803 R:17408 HW:7215 WU:11117.8/m                          
                                              

 cgminer version 3.9.0h2 - Started: [2014-02-05 16:46:44]
                                          
 (5s):408.1G (avg):415.5Gh/s | A:1883755  R:16256  HW:6779  WU:10806.0/m                                                        
 ST: 2  SS: 0  NB: 38  LW: 2015649  GF: 0  RF: 0                                                                                
 Connected to stratum.mining.eligius.st diff 256 with stratum as user 1Bk9VSwaXn9UhpdowTwHEPEWw95i5F7N2_Miner3                  
 Block: 12f30e18...  Diff:2.62G  Started: [22:31:54]  Best share: 4.79M                                                        
                                            
 [P]ool management Settings [D]isplay options [Q]uit                                                                          
 HFA 0: 81C/.79V 76C/.79V 73C/.79V 73C/.79V 410.5G/415.6Gh/s | A:1884011 R:16256 HW:6780 WU:10808.2/m                          
--------------------------------------------------------------------------------    

[spectre2112]

Once you have that, simply mount the second board, connect the chaining cable from the top to the bottom board (using the 2 closest 16 pin connectors), MOVE the 5-pin cable to the second board, and then connect the spare set of PCI-E cables to the new lower board.

-Phil

So, when daisy-chaining boards together, the 5 pin power connector needs to be attached to the LAST board in the chain?
If multiple boards are being supplied with enough power, how many could be PRACTICALLY chained together?

[HF-Engineer]

To overclock (may void your warranty!), add the following parameter to the cgminer command line:

Code:

--hfa-hash-clock xxx

Where xxx is the speed in mhz you'd like to run.  Valid values are 125 on up, and default is 550.

If you are running a RPI with our image, you can add this line to the "Extra cgminer parameters" section in the settings page of the web interface.

In a perfect world the hashrate is mhz X 768 (the total number of hashing units built into the ASIC).  Each of the 4 die has 96 cores, and each core has two hashing units, for a total of 768.  Note: It's normal to have a few defective cores per die, and it's also considered normal for some cores to produce occasional errors.  This means the maximum possible hash rate is 422.4 Gh/s when running at 550mhz, but in the real-world it will typically be lower due to the reasons I just mentioned.

Some boards will like certain clock rates and some will like others.  If you do experiment with values, and I'm not recommending you do, try small increments and watch the error rates and/or if the ASIC stalls.  When it attempts to draw too much power, the power supply can momentarily dip and cause it to stop sending work.  This is when you'll see the watchdog timer get invoked.

On an average board, the buck converters (take 12V down to core voltage) are usually the limiting factor for ASIC performance.  They put out over 400 amps of low voltage power for the ASIC's cores to operate with.  Right now in the present version of the firmware and cgminer, the regulators cannot be adjusted, but soon we will release firmware that supports adjustment of these regulators as well as independent clock speed adjustment per die.

We are also presently performing lab qualification of the silicon, so I'll soon have some numbers for you guys as to what the silicon is capable of.

The on chip programmable PLL is capable of pushing the clock rate to well over 1ghz.  (You would need about 2X the power and cooling to hit this rate though!)

-Phil

[joshv06]

http://www.ebay.com/itm/2-54mm-Pitch-16-Pin-16-Wire-Rainbow-Color-IDC-Flat-Ribbon-Cable-Connector-/400617283710?pt=LH_DefaultDomain_0&hash=item5d46a6a07eThan

Is this the Cable we need to chain BJ boards together?

Thanks

[HF-Engineer]

So, when daisy-chaining boards together, the 5 pin power connector needs to be attached to the LAST board in the chain?
If multiple boards are being supplied with enough power, how many could be PRACTICALLY chained together?

We have not tested the firmware with chains longer than 3 boards.  This is the configuration present in the Sierra product.

Practically, the limit is getting all the boards powered.  If the Sierra, there are 2 PSU's, and the left side of each board is connected to one PSU, and the right sides are connected to the other.  It's important that the grounding be the same on the left side to ensure reliability.  This will be what ultimately limits how many you can practically chain.

-Phil

[HF-Engineer]

http://www.ebay.com/itm/2-54mm-Pitch-16-Pin-16-Wire-Rainbow-Color-IDC-Flat-Ribbon-Cable-Connector-/400617283710?pt=LH_DefaultDomain_0&hash=item5d46a6a07eThan

Is this the Cable we need to chain BJ boards together?

Thanks

No, that is a 2.54mm pitch connector, whereas the chaining connector uses a 2mm pitch.  I think we use a Samtec 2x8 IDC connector in the cables.

It's the same pitch as the old 2.5" laptop IDE drive cables were.  So if you had one of those laying around, you might be able to cut it down.  (They were 44 pin IIRC)

-Phil

[joshv06]

Thanks for your ongoing input. If anyone finds these online from the US, let me know. They seem hard to come by.

[port119]

anyone in the UK know if a standard 10A power lead will work with the BJ instead of the 13A one provided by HF?
BJ is rebooting itself every few minutes with this error "HFA 0 NOTICE: ######################### WARNING: Work Watchdog Reboot Imminent!",
then the rasberry PI reboots (sometimes)

edit: also getting this error a lot "HFB 0: Unhandled operation code 12"

edit: MinePeon Version:  0.2.4.3hf8
HashFast Firmware : 735ccca5

[SolarSilver]

anyone in the UK know if a standard 10A power lead will work with the BJ instead of the 13A one provided by HF?

Yes, because you are only loading the power supply up to 450 Watts, which is less than 2 Amps

[port119]

thanks - hmm..got a bit of a problem then, BJ is throwing up "HFB 0: Unhandled operation code 12"" in cgminer-3.12 every 20 seconds or so and rebooting.
also not working on Pi

anyone seen anything like that before?
thanks
warren

[port119]

wait - seems to be working now. am thinking it might have been too warm in the room. doh!

[phreakk]

wait - seems to be working now. am thinking it might have been too warm in the room. doh!

I keep mine outside in the garage! 58C on all cores!

[CurcO]

wait - seems to be working now. am thinking it might have been too warm in the room. doh!

I keep mine outside in the garage! 58C on all cores!

What Clock speed are you getting now ?

[phreakk]

wait - seems to be working now. am thinking it might have been too warm in the room. doh!

I keep mine outside in the garage! 58C on all cores!

What Clock speed are you getting now ?

I run it at 581, as it is the most stable ATM. Once I go above that watchdog resetting the unit so often results in <400GH/s hashrate .

Clocked at 615 it only reaches 63C, but watchdog resets it ~3x/min!!

[kenpofighta]

To overclock (may void your warranty!), add the following parameter to the cgminer command line:

Code:

--hfa-hash-clock xxx

Where xxx is the speed in mhz you'd like to run.  Valid values are 125 on up, and default is 550.

If you are running a RPI with our image, you can add this line to the "Extra cgminer parameters" section in the settings page of the web interface.

In a perfect world the hashrate is mhz X 768 (the total number of hashing units built into the ASIC).  Each of the 4 die has 96 cores, and each core has two hashing units, for a total of 768.  Note: It's normal to have a few defective cores per die, and it's also considered normal for some cores to produce occasional errors.  This means the maximum possible hash rate is 422.4 Gh/s when running at 550mhz, but in the real-world it will typically be lower due to the reasons I just mentioned.

Some boards will like certain clock rates and some will like others.  If you do experiment with values, and I'm not recommending you do, try small increments and watch the error rates and/or if the ASIC stalls.  When it attempts to draw too much power, the power supply can momentarily dip and cause it to stop sending work.  This is when you'll see the watchdog timer get invoked.

On an average board, the buck converters (take 12V down to core voltage) are usually the limiting factor for ASIC performance.  They put out over 400 amps of low voltage power for the ASIC's cores to operate with.  Right now in the present version of the firmware and cgminer, the regulators cannot be adjusted, but soon we will release firmware that supports adjustment of these regulators as well as independent clock speed adjustment per die.

We are also presently performing lab qualification of the silicon, so I'll soon have some numbers for you guys as to what the silicon is capable of.

The on chip programmable PLL is capable of pushing the clock rate to well over 1ghz.  (You would need about 2X the power and cooling to hit this rate though!)

-Phil

Thanks Phil, I had actually found how to overclock last night by reading the latest cgminer documentation. Oddly enough I had to underclock the failing unit to get it up close to 400GH/s. I had to take it down to 525. I then bumped it to 535 and errors started back up so I brought it back down. It has been steady since about midnight last night.

I love having this info though because it means these machines are capable of much more.

[spectre2112]

http://www.ebay.com/itm/2-54mm-Pitch-16-Pin-16-Wire-Rainbow-Color-IDC-Flat-Ribbon-Cable-Connector-/400617283710?pt=LH_DefaultDomain_0&hash=item5d46a6a07eThan

Is this the Cable we need to chain BJ boards together?

Thanks

No, that is a 2.54mm pitch connector, whereas the chaining connector uses a 2mm pitch.  I think we use a Samtec 2x8 IDC connector in the cables.

It's the same pitch as the old 2.5" laptop IDE drive cables were.  So if you had one of those laying around, you might be able to cut it down.  (They were 44 pin IIRC)

-Phil

How about these: http://www.ebay.com/itm/10pcs-Flat-cable-connector-2mm-0-079-IDC-SOCKET-16Pin-2x8Pin-/291060588602?pt=LH_DefaultDomain_0&hash=item43c490203a

Then we just need the ribbon cable. A bit of D.I.Y.