[ANN] Spondoolies-Tech - carrier grade, data center ready mining rigs

[TracerX]

u all must not watch the hashrate of them for the first few days, they go down a good 1-3GH / day.
I have 4 SP20's...
All of them run under 100C
I do know for damn sure if we could custom tune the frequencies these run at they would be much more effecient!
The BIST is far too agressive, at the first sign of a HW error it declocks them like mad. Its ok to have a few HW error lol!
The firmware behaviour doesnt try to get the highest / stable clock at a given voltage unfortunately it seems.

if you use auto fan and don't unnecessary cap the power to loops, then machine will bump the highest temp to 115C at your max voltage and you will get more speed.
In addition, I find that sometimes it does go down a few GH in comparison with first hour, but then IT ALSO goes up if you wait for longer than 2-3 days, basically fluctuating in a long waves of +/-5 to 10GH or so.

this seems to be true for me.

most of my machines are 1050 to 1250 gh.

@GenTarkin  are your machines at 1350 or higher?

All 5 of mine drop a few Gh/s over several days. For example I have one that started at 1609 leveled out over several hours to 1603 and after 3/12 days is at 1592 and will probably stay around this mark until I restart. Not enough to see on graph but if you monitor Asic stats you will see it.

Mine do that as well, I should have been clear--I incorrectly assumed that we were talking about a more appreciable drop.  This behavior seems normal in my experience.  That said, we're talking about an insignificant loss in hash power; even double digit GH loss would still account for a 0.0x% percent change, no?

[goxed]

Any ideas how to get chip#2, #3 out of its slumber?

[Biodom]

Any ideas how to get chip#2, #3 out of its slumber?

>snip image<

not sure why there is only 1mv difference between starting and max voltage for unit 2.
Maybe increase to 5mv? It is also OK to have max voltage setting up to 10mv higher than your max starting voltage for the highest of your chips starting voltage. the other chips will adjust automatically, in my experience.
are you limiting power as well?

[goxed]

Any ideas how to get chip#2, #3 out of its slumber?

>snip image<

not sure why there is only 1mv difference between starting and max voltage for unit 2.
Maybe increase to 5mv?
are you limiting power as well?

yes, I want to limit power consumption to 550Watt @ wall and maintain ~1.9GH/Watt.

[Biodom]

Any ideas how to get chip#2, #3 out of its slumber?

>snip image<

not sure why there is only 1mv difference between starting and max voltage for unit 2.
Maybe increase to 5mv?
are you limiting power as well?

yes, I want to limit power consumption to 550Watt @ wall and maintain ~1.9GH/Watt.

in my opinion if you limit both max power (too low) AND max voltage, then the machine cannot optimize chip voltage properly and is run (at least sometime) sub-optimally.

[2112]

in my opinion if you limit both max power (too low) AND max voltage, then the machine cannot optimize chip voltage properly and is run (at least sometime) sub-optimally.

You guys should pressure Zvisha and Spondoolies to implement a proper gradient-free multi-dimensional optimization method in their software:

http://en.wikipedia.org/wiki/Nelder–Mead_method

Do not reinvent the wheel!

[Guy Corem]

in my opinion if you limit both max power (too low) AND max voltage, then the machine cannot optimize chip voltage properly and is run (at least sometime) sub-optimally.

You guys should pressure Zvisha and Spondoolies to implement a proper gradient-free multi-dimensional optimization method in their software:

http://en.wikipedia.org/wiki/Nelder–Mead_method

Do not reinvent the wheel!

https://github.com/Spondoolies-Tech

[alh]

in my opinion if you limit both max power (too low) AND max voltage, then the machine cannot optimize chip voltage properly and is run (at least sometime) sub-optimally.

You guys should pressure Zvisha and Spondoolies to implement a proper gradient-free multi-dimensional optimization method in their software:

http://en.wikipedia.org/wiki/Nelder–Mead_method

Do not reinvent the wheel!

https://github.com/Spondoolies-Tech

Is there anything more descriptive than a Git repository URL? When you go there, you are confronted with 8 "things" that I assume are kinda like directories. When you dive into them, there are bits of code and stuff. Wouldn't it make more sense to give folks guidance  on how to achieve the desired goal using settings in the GUI? Am I just being too dense on this?

[Guy Corem]

in my opinion if you limit both max power (too low) AND max voltage, then the machine cannot optimize chip voltage properly and is run (at least sometime) sub-optimally.

You guys should pressure Zvisha and Spondoolies to implement a proper gradient-free multi-dimensional optimization method in their software:

http://en.wikipedia.org/wiki/Nelder–Mead_method

Do not reinvent the wheel!

https://github.com/Spondoolies-Tech

Is there anything more descriptive than a Git repository URL? When you go there, you are confronted with 8 "things" that I assume are kinda like directories. When you dive into them, there are bits of code and stuff. Wouldn't it make more sense to give folks guidance  on how to achieve the desired goal using settings in the GUI? Am I just being too dense on this?

The git URL was posted to allow 2112 to improve our firmware per her/his suggestion.

[TheRealSteve]

Edit: What they said ^

I think the above was just Spondoolies-Tech's slightly unsubtly way of saying: Hey, our software is open source.  Here's our git repository.  Feel free to fork it, make changes - such as the suggested Nelder–Mead approach for N-dimensional optimization, and submit pull requests, and we'd be happy to review the changes and possible integrate them into our baseline - or you can provide your own custom builds for users who would be interested in it.

Code repositories are generally not intended for end-users

[Guy Corem]

Edit: What they said ^

I think the above was just Spondoolies-Tech's slightly unsubtly way of saying: Hey, our software is open source.  Here's our git repository.  Feel free to fork it, make changes - such as the suggested Nelder–Mead approach for N-dimensional optimization, and submit pull requests, and we'd be happy to review the changes and possible integrate them into our baseline - or you can provide your own custom builds for users who would be interested in it.

Code repositories are generally not intended for end-users

Correct.

[Biodom]

in my opinion if you limit both max power (too low) AND max voltage, then the machine cannot optimize chip voltage properly and is run (at least sometime) sub-optimally.

You guys should pressure Zvisha and Spondoolies to implement a proper gradient-free multi-dimensional optimization method in their software:

http://en.wikipedia.org/wiki/Nelder–Mead_method

Do not reinvent the wheel!

Wow. Solid.

[klondike_bar]

just got an SP20 that was producing serial connection errors (not i2c) in loop 0 and would not hash. reseating the cables did not solve the problem.

turns out swapping the serial ribbon cables moved the error over to loop 2 - so the unit works! I just need a new ribbon cable for it. (@spondoolies? I emailed benny already - hopefully i can get a replacement to recover the missing 25% of the unit)

[wh00per]

Edit: What they said ^

I think the above was just Spondoolies-Tech's slightly unsubtly way of saying: Hey, our software is open source.  Here's our git repository.  Feel free to fork it, make changes - such as the suggested Nelder–Mead approach for N-dimensional optimization, and submit pull requests, and we'd be happy to review the changes and possible integrate them into our baseline - or you can provide your own custom builds for users who would be interested in it.

Code repositories are generally not intended for end-users

I wish this was true. You can fork it and use it as you code the changes, but Zvi has limited time to review any pulls. I can still see that the UI on the latest SP 20 still have some errors which I tried to rectify since I got my SP30 .. forked the stuff, did the changes, which were never implemented by SP-T due to the lack of time .. The pull request is still there  .. but I stopped asking, eventually in the end, the units are hashing ..

Nevertheless, code repositories are good, and I'm an end-user. Probably the exception which confirms the rule.

Correct.

 

[alh]

I have found and examined the "SP20 Jackson Quick Start Guide". At the tail end of that 9 page document there is a reference to "SP20 Jackson User Guide" for further details. I have been unable to find the "SP20 User Guide" anywhere. Does this document exist, and if so where?  Is the "SP3x User Guide" useful at all for the SP20? Sorry if this is an silly  question, but I couldn't find anything relevant here.

[RoadStress]

I have found and examined the "SP20 Jackson Quick Start Guide". At the tail end of that 9 page document there is a reference to "SP20 Jackson User Guide" for further details. I have been unable to find the "SP20 User Guide" anywhere. Does this document exist, and if so where?  Is the "SP3x User Guide" useful at all for the SP20? Sorry if this is an silly  question, but I couldn't find anything relevant here.

You can ask here any question that you might have.

[2112]

Edit: What they said ^

I think the above was just Spondoolies-Tech's slightly unsubtly way of saying: Hey, our software is open source.  Here's our git repository.  Feel free to fork it, make changes - such as the suggested Nelder–Mead approach for N-dimensional optimization, and submit pull requests, and we'd be happy to review the changes and possible integrate them into our baseline - or you can provide your own custom builds for users who would be interested in it.

Code repositories are generally not intended for end-users

My not so subtle response to this: Spondoolies hired a single junior/young software developer who had landed far out of his depth with regards to state-of-the-art of the optimization of rather complex analog/digital system. His attempts at one-dimensional optimization problems when searching for the optimal configuration parameters show that he isn't aware of http://en.wikipedia.org/wiki/Golden_section_search known since mid-1950 or mid-1960. And various folks here already started to request optimization in 2 or more dimensions. This is going to end up rather badly with regards to the software quality when the corporation gives his lead developer no time to actually research the problem space. Instead they gave him a mishmash of tasks from the request list that are rather trivial and cosmetic.

So my friendly advice to zvisha is: step away from fulfilling trivial maintenance requests and broaden your knowledge with a quick peek into gradient-free/derivative-free optimization methods that I've mentioned. The payoff to him (personally) and the company and its user base will be much higher.

My not so subtle advice to Guy: you'll need to give time to your staff to do an actual research as opposed to bogging them down with constant development and maintenance requests. Your company's current work on finding the optimal operating point for your miners is equivalent to the state-of-the-art around World War II. Either give zvisha some time off of D and let him concentrate on R (from R&D) or hire a actual mathematical optimization consultant. This should be quite easy in your location in Israel.

I actually don't care if you open source it or keep it as a closed-source module that controls the cgminer through its API. You could probably then charge extra for it or differentiate your product from the competition in marketing.

The above was a free advice. If you don't like it just ask and I will give you a full refund of the money paid.

[klondike_bar]

^I think you are making a trivial issue into a major product fault that it isnt.

The Sp20 is already leading-edge for efficiency, and while theres a little space to improve efficiency with a little bit finer tuning processes, it might only be by 0.02-0.04w/GH

[2112]

^I think you are making a trivial issue into a major product fault that it isnt.

The Sp20 is already leading-edge for efficiency, and while theres a little space to improve efficiency with a little bit finer tuning processes, it might only be by 0.02-0.04w/GH

I think that you are wrong. I'm following this thread from the beginning and observed a very rapid release cycle with multiple regressions which is indicative of the developers thrashing around without understanding what is going on with the product in the field.

I also noticed multiple user complaints about things like "entire ASIC loops disabled by BIST". I believe zvisha has already discovered a while back that GO/NO GO BIST that he had inherited from the Spondoolies' hardware designer is counterproductive and overly cautious. The use of BIST (in the field) should be completely abandoned now, it could be still valuable in the factory or in the pre-final-assembly stages.

I don't know what is the internal personal dynamics of Spondoolies as a corporation. Does zvisha have enough inside clout to say NO! to some rather trivial support requests via Skype/ssh/Teamviewer? Can he influence the hardware designer to give him more information from the chip that just a single go/no-go bit and a temperature?

What is required now is not more of "finer tuning". There's a need to step back and do some real research and appraisal of the entire field and how it is changing to more and more unattended operation. The cute "Thank you for the personal attention!" notices from the customers may be pleasant now, but they are actually indicative of the production problems in the factory.

[Zich]

I have found and examined the "SP20 Jackson Quick Start Guide". At the tail end of that 9 page document there is a reference to "SP20 Jackson User Guide" for further details. I have been unable to find the "SP20 User Guide" anywhere. Does this document exist, and if so where?  Is the "SP3x User Guide" useful at all for the SP20? Sorry if this is an silly  question, but I couldn't find anything relevant here.

Don't worry about the guide. There are plenty of member that will happily help you  
Some of useful guide:
https://bitcointalk.org/index.php?topic=872014.0
https://bitcointalk.org/index.php?topic=886633