Author Topic: NEW KNC TITAN FIRMWARE - 'GenTarkin's Mod' OFFICIAL TOPIC  (Read 16567 times)

Offline GenTarkin

  • Jr. Member
  • **
  • Posts: 94
Re: NEW KNC TITAN FIRMWARE - 'GenTarkin's Mod' OFFICIAL TOPIC
« Reply #60 on: October 19, 2015, 04:20:32 PM »
I'm running v0.99d

Over the course of 24 hours, it shuts down over half my dies, and I have to erase the bypass list.  It seems to be a daily cycle of decreasing performance, then me erasing the bypass list and rebooting the miner.

If I bypass health checking on all the dies, is there any advantage to running this firmware over stock firmware?

So, the bypass list DOES NOT shut down dies. It simply will tell the monitoring script to bypass checking the dies for their health beacuse theyve caused too many hard / soft resets to quickly, thats how dies get 'bypassed' for health checking. Which in turn means those dies are effectively on their own once they get listed in the bypass file. "On their own" means if they need a reset, they will go ignored.
So yes, those dies will most likely stop hashing because of their bad behaviour.
This is not the fault of my firmware, its the fault of those dies causing too many hard / soft resets too quickly.
People using my firmware were complaining about bfgminer constantly restarting(constant soft / hard resets) ... an autobypass of those "spammy" dies is the only solution.
What it comes down to is ... either deal w/ constant hard soft resets OR deal w/ the autobypass. There is no in-between I guess.
Initially I was gonna code in an option to enable / disable auto bypass. But most people wouldnt even understand what that means therefore its pointless to have such an option.

In your case, if I had to guess, it sounds like a die(s) end up requiring nonstop resets whether hard or soft, that doesnt fix them. So, in that case, even on stock firmware, you would have to power cycle the entire machine. I cannot fix dies in a state where hard resets wont bring them back.

To answer your question, bypassing ALL dies means your machine is almost gaurunteed to have a decrease in performance over time. Because theres no handling of dies which would come back hashing if they were issued a soft / hard reset.

The stock firmware only issues soft resets to the dies which need reset. So, it will literally just sit there and issue soft resets all day long until u come along and power cycle the machine(thats why I started coding up this project to begin with, so I didnt have to power cycle my miner every day).
Also, u lose the temp protection my firmware offers and all the other features of my firmware, if u run stock.
You could try running v.99b which does not have the autobypass functionality.
« Last Edit: October 19, 2015, 04:28:41 PM by GenTarkin »

Offline Searing

  • Full Member
  • ***
  • Posts: 312
  • Clueless
Re: NEW KNC TITAN FIRMWARE - 'GenTarkin's Mod' OFFICIAL TOPIC
« Reply #61 on: October 20, 2015, 07:36:35 AM »
I'm running v0.99d

Over the course of 24 hours, it shuts down over half my dies, and I have to erase the bypass list.  It seems to be a daily cycle of decreasing performance, then me erasing the bypass list and rebooting the miner.

If I bypass health checking on all the dies, is there any advantage to running this firmware over stock firmware?

So, the bypass list DOES NOT shut down dies. It simply will tell the monitoring script to bypass checking the dies for their health beacuse theyve caused too many hard / soft resets to quickly, thats how dies get 'bypassed' for health checking. Which in turn means those dies are effectively on their own once they get listed in the bypass file. "On their own" means if they need a reset, they will go ignored.
So yes, those dies will most likely stop hashing because of their bad behaviour.
This is not the fault of my firmware, its the fault of those dies causing too many hard / soft resets too quickly.
People using my firmware were complaining about bfgminer constantly restarting(constant soft / hard resets) ... an autobypass of those "spammy" dies is the only solution.
What it comes down to is ... either deal w/ constant hard soft resets OR deal w/ the autobypass. There is no in-between I guess.
Initially I was gonna code in an option to enable / disable auto bypass. But most people wouldnt even understand what that means therefore its pointless to have such an option.

In your case, if I had to guess, it sounds like a die(s) end up requiring nonstop resets whether hard or soft, that doesnt fix them. So, in that case, even on stock firmware, you would have to power cycle the entire machine. I cannot fix dies in a state where hard resets wont bring them back.

To answer your question, bypassing ALL dies means your machine is almost gaurunteed to have a decrease in performance over time. Because theres no handling of dies which would come back hashing if they were issued a soft / hard reset.

The stock firmware only issues soft resets to the dies which need reset. So, it will literally just sit there and issue soft resets all day long until u come along and power cycle the machine(thats why I started coding up this project to begin with, so I didnt have to power cycle my miner every day).
Also, u lose the temp protection my firmware offers and all the other features of my firmware, if u run stock.
You could try running v.99b which does not have the autobypass functionality.


Just to be clear. Shutting a die to OFF just puts in on the bypass list? or is that a whole different process?

My point is if I set a DIE to off and it simply goes to the bypass list and any physical hardware issues like voltage or other flakyness is not effected ..at the machine end I mean....(stableness of the hash/equip/etc) ...then if all the symptoms persist even if the die is OFF perhaps I should revisit them and just let them 'soft/reset to their heart's content..if basically I'm getting the same hardware symptoms either by letting them run (wonky/less stable) and get some hash maybe at some setting or setting them to OFF really does make them more stable (ie OFF is somehow different then the above list)

assuming I have what you guys are talking about correct (I'm sooooo clueless) that is ...again..if I have the thread of what you are saying above correct :)

Offline Mattzees

  • Sr. Member
  • ****
  • Posts: 704
Re: NEW KNC TITAN FIRMWARE - 'GenTarkin's Mod' OFFICIAL TOPIC
« Reply #62 on: October 20, 2015, 05:14:08 PM »

...basically I'm getting the same hardware symptoms either by letting them run (wonky/less stable) and get some hash maybe at some setting or setting them to OFF really does make them more stable (ie OFF is somehow different then the above list)

assuming I have what you guys are talking about correct (I'm sooooo clueless) that is ...again..if I have the thread of what you are saying above correct :)


Have you checked (or swapped out) your PSUs?  A flaky PSU can cause behavior like that.

Offline GenTarkin

  • Jr. Member
  • **
  • Posts: 94
Re: NEW KNC TITAN FIRMWARE - 'GenTarkin's Mod' OFFICIAL TOPIC
« Reply #63 on: October 20, 2015, 06:03:51 PM »
I'm running v0.99d

Over the course of 24 hours, it shuts down over half my dies, and I have to erase the bypass list.  It seems to be a daily cycle of decreasing performance, then me erasing the bypass list and rebooting the miner.

If I bypass health checking on all the dies, is there any advantage to running this firmware over stock firmware?

So, the bypass list DOES NOT shut down dies. It simply will tell the monitoring script to bypass checking the dies for their health beacuse theyve caused too many hard / soft resets to quickly, thats how dies get 'bypassed' for health checking. Which in turn means those dies are effectively on their own once they get listed in the bypass file. "On their own" means if they need a reset, they will go ignored.
So yes, those dies will most likely stop hashing because of their bad behaviour.
This is not the fault of my firmware, its the fault of those dies causing too many hard / soft resets too quickly.
People using my firmware were complaining about bfgminer constantly restarting(constant soft / hard resets) ... an autobypass of those "spammy" dies is the only solution.
What it comes down to is ... either deal w/ constant hard soft resets OR deal w/ the autobypass. There is no in-between I guess.
Initially I was gonna code in an option to enable / disable auto bypass. But most people wouldnt even understand what that means therefore its pointless to have such an option.

In your case, if I had to guess, it sounds like a die(s) end up requiring nonstop resets whether hard or soft, that doesnt fix them. So, in that case, even on stock firmware, you would have to power cycle the entire machine. I cannot fix dies in a state where hard resets wont bring them back.

To answer your question, bypassing ALL dies means your machine is almost gaurunteed to have a decrease in performance over time. Because theres no handling of dies which would come back hashing if they were issued a soft / hard reset.

The stock firmware only issues soft resets to the dies which need reset. So, it will literally just sit there and issue soft resets all day long until u come along and power cycle the machine(thats why I started coding up this project to begin with, so I didnt have to power cycle my miner every day).
Also, u lose the temp protection my firmware offers and all the other features of my firmware, if u run stock.
You could try running v.99b which does not have the autobypass functionality.


Just to be clear. Shutting a die to OFF just puts in on the bypass list? or is that a whole different process?

My point is if I set a DIE to off and it simply goes to the bypass list and any physical hardware issues like voltage or other flakyness is not effected ..at the machine end I mean....(stableness of the hash/equip/etc) ...then if all the symptoms persist even if the die is OFF perhaps I should revisit them and just let them 'soft/reset to their heart's content..if basically I'm getting the same hardware symptoms either by letting them run (wonky/less stable) and get some hash maybe at some setting or setting them to OFF really does make them more stable (ie OFF is somehow different then the above list)

assuming I have what you guys are talking about correct (I'm sooooo clueless) that is ...again..if I have the thread of what you are saying above correct :)

Setting a die to OFF in advanced page will not put it on autobypass list. At least it shouldnt be, if it does then thats a bug and I have to revisit the code. Now, flip the coin... if you have a malfunctioning die that otherwise SHOULD BE set to OFF ... then eventually it will get added to the autobypass list to prevent constant soft/hard resets of that dies(constant restarts of bfgminer).
When a die is added to the bypass list any further health checking of the die is bypassed. Meaning, if that die is misbehaving, it will eventually stop hashing and no action will be taken to get it hashing again. No action means, no soft / hard resets.
The die should really just be set to OFF to keep the miner running well.
If you chose to ignore the die and there was no autobypass functionality, in my older firmware it would cause CONSTANT HARD RESETS(bfgminer would be restarted every 10-15mins) because that die should be set OFF. On stock KNC firmware it will cause constant soft resets which just cause noise and the die wont magically start working after hours and hours of constant soft resets. LOL!

Really, what it means is no matter what firmware you are running, dies like that should just be set to OFF. For all intents and purposes the die is permanently damaged(most likely the DCDC's are shot and the die is completely fine). Set the thing OFF.

The reason I even coded in the autobypass is cuz everyone was complaining about some of their miners constantly having bfgminer restarts.
Well thats cuz a die(s) was having to constantly getting flagged for a hard reset, and they didnt know how to set the die to proper voltage / mhz to keep the die running somewhat properly or how to set the dang die OFF. LOL!
So, I wrote the autobypass code to prevent constant misbehaving dies from causing constant bfgminer restarts.
People thought my firmware was "unstable" because of those restarts. When in fact it has to do more about the condition of their dies but they cant understand that and set it up properly! lol =P
« Last Edit: October 20, 2015, 06:09:07 PM by GenTarkin »

Offline GenTarkin

  • Jr. Member
  • **
  • Posts: 94
Re: NEW KNC TITAN FIRMWARE - 'GenTarkin's Mod' OFFICIAL TOPIC
« Reply #64 on: October 23, 2015, 04:04:39 PM »
OFFICIAL ANNOUNCEMENT!!! v.99e is ready for download!!!

Changelog:
Code: [Select]
FEATURES & FIXES:
 1. groundwork code for multilanguage capability in webgui. Upon reboot after installing upgrade it will ask which language you want the webgui, only working & complete one is English for now
 2. chinese skeleton site, only includes the status page translated
 3. much more accurate wall watts
 4. ability to set your PSU's expected effeciency in SYSTEM tab which is used to calculate wall watts(defaults to 85%)
 5. more accurate per cube power consumption in ADVANCED tab
 6. Added link "Release Notes" to STATUS page
 7. more modular code in monitoring script(which will give way to autotune someday =-) )

If you're a paid member, downloads can be found on ur bitcoinlasvegas membership page login: http://bitcoinlasvegas.net/kncminer-titan-custom-firmware-mod-paid-membership-page/

If ur still a free member, well all the more reason to upgrade =)

If you have my btsync share, the new version will be available once it sync to ur client.

Thanks all for ur continued support & funds & donations!! This firmware is turning out to be quite the gem!

Future releases:
The "Energy Saver Edition" - autotunes voltages per die to as low as possible while maintaining decent hw% & die stability.
*possible tweaks: pool info added to status page.

Offline GenTarkin

  • Jr. Member
  • **
  • Posts: 94
Re: NEW KNC TITAN FIRMWARE - 'GenTarkin's Mod' OFFICIAL TOPIC
« Reply #65 on: October 28, 2015, 12:36:36 AM »
The autotune is...... WORKING!
I think its safe to say, I have the most sophisticated Titan on the planet atm =P LOL

Still in the process of further fine tuning etc and I especially will be fine tuning / tweaking the firmware once my test subjects undergo the firmwares abuse LOL! =P

Offline Mattzees

  • Sr. Member
  • ****
  • Posts: 704
Re: NEW KNC TITAN FIRMWARE - 'GenTarkin's Mod' OFFICIAL TOPIC
« Reply #66 on: October 28, 2015, 02:20:56 AM »
The autotune is...... WORKING!
I think its safe to say, I have the most sophisticated Titan on the planet atm =P LOL

Still in the process of further fine tuning etc and I especially will be fine tuning / tweaking the firmware once my test subjects undergo the firmwares abuse LOL! =P


I would be happy to test if u wanna get it to me.

Offline Searing

  • Full Member
  • ***
  • Posts: 312
  • Clueless
Re: NEW KNC TITAN FIRMWARE - 'GenTarkin's Mod' OFFICIAL TOPIC
« Reply #67 on: October 28, 2015, 06:49:07 AM »
The autotune is...... WORKING!
I think its safe to say, I have the most sophisticated Titan on the planet atm =P LOL

Still in the process of further fine tuning etc and I especially will be fine tuning / tweaking the firmware once my test subjects undergo the firmwares abuse LOL! =P

how does this work exactly ..does it just reboot a bunch of times testing a setting? confused? (no need to reply here if you plan on doing a reply in bitcointalk....but use simple words please)

seems i asked about something similar to this for 'testing' dead dies but i see you are just going for elec eff but still kinda scratching head on the ifs/hows for this mod


Offline GenTarkin

  • Jr. Member
  • **
  • Posts: 94
Re: NEW KNC TITAN FIRMWARE - 'GenTarkin's Mod' OFFICIAL TOPIC
« Reply #68 on: October 28, 2015, 02:54:43 PM »
The autotune is...... WORKING!
I think its safe to say, I have the most sophisticated Titan on the planet atm =P LOL

Still in the process of further fine tuning etc and I especially will be fine tuning / tweaking the firmware once my test subjects undergo the firmwares abuse LOL! =P

how does this work exactly ..does it just reboot a bunch of times testing a setting? confused? (no need to reply here if you plan on doing a reply in bitcointalk....but use simple words please)

seems i asked about something similar to this for 'testing' dead dies but i see you are just going for elec eff but still kinda scratching head on the ifs/hows for this mod

So, as it is...
When user turns on autotune, its gonna assume the machine is stable w/ current user settings and it will tweak starting from whatever clocks/voltages the user has set. Then, its gonna require that bfgminer has been running for a least an hour, then its gonna wait its specified window( maybe another hour, Im testing for optimimum hw% window measurement). Then its gonna iterate each die and query the last 'window  size' worth of hw%, if dies are above 1.1% hw% its gonna raise the voltage to a notch( to a ceiling of -0.0366v), or if the die is under 1.1% its gonna lower the voltage a notch. Next iteration of the autotune(window size later) its gonna change the dies according to the same formula again, this time if the decrease of dies hw% above 1.1% is less than a 50% decrease in hw% its gonna revert the voltage change and consider that die inherently high hw% ... lowering the voltage didnt help much. If the dies that were previously lowered are still under 1.1% , then its gonna lower them another notch. If those dies are above 1.1%, its gonna raise the voltage again then call it quits(those dies are done tuning).
So, in other words dies will be tuned down in voltage each window until they start having hw% above 1.1%, at which point they will be raised a notch.
If any dies experience stability problems(either during tuning phase or thereafter) and have previously been lowered in voltage then the script will raise the voltage a notch(up to the starting voltage when autotune was enabled) ... till they are stable. (autotune will then not tune that die anymore).
If any dies are in a throttled state, autotune will not tweak them till they are no longer throttled, same w/ any other stability problems of dies not caused by autotune. (still working on coding this up.)
Autotune should place entire miner average hw% between .6-1.5% HW error when its done. Up to 2% is acceptable w/o any loss in effeciency.

Thats pretty much how it works atm, =)
A lot of testing will go into trying to code this so it works well for as many titans as possible. If for some reason autotune messes up clocks n voltages and makes someones titan unstable, they can just disable autotune and it will revert to their previous clocks / voltages, prior to enabling autotune.

As it stands, autotune will never set a die to higher than -0.0366v (for safety reasons).

Offline Mattzees

  • Sr. Member
  • ****
  • Posts: 704
Re: NEW KNC TITAN FIRMWARE - 'GenTarkin's Mod' OFFICIAL TOPIC
« Reply #69 on: October 28, 2015, 03:46:09 PM »

Thats pretty much how it works atm, =)
A lot of testing will go into trying to code this so it works well for as many titans as possible. If for some reason autotune messes up clocks n voltages and makes someones titan unstable, they can just disable autotune and it will revert to their previous clocks / voltages, prior to enabling autotune.


Feature requests:
1) Checkboxes to disable auto-tune and monitoring for each die
2) Ability to view or export the auto-tune log (including miner performance) so that we can graph & learn from its experience

I know I have a problematic rig, but again, when I run v1.15, the software crashes and fails to detect any of my dies, the miner will run full speed for weeks with no issues.  Would love to know why this is.

Also willing to beta test for you.

Offline GenTarkin

  • Jr. Member
  • **
  • Posts: 94
Re: NEW KNC TITAN FIRMWARE - 'GenTarkin's Mod' OFFICIAL TOPIC
« Reply #70 on: October 28, 2015, 04:15:17 PM »

Thats pretty much how it works atm, =)
A lot of testing will go into trying to code this so it works well for as many titans as possible. If for some reason autotune messes up clocks n voltages and makes someones titan unstable, they can just disable autotune and it will revert to their previous clocks / voltages, prior to enabling autotune.


Feature requests:
1) Checkboxes to disable auto-tune and monitoring for each die
2) Ability to view or export the auto-tune log (including miner performance) so that we can graph & learn from its experience

I know I have a problematic rig, but again, when I run v1.15, the software crashes and fails to detect any of my dies, the miner will run full speed for weeks with no issues.  Would love to know why this is.

Also willing to beta test for you.
1. This wont be possible, not in a way that I know of, Ill mull over some ideas but it definitely wont be in initial release unless people really need it.
2. Ill consider it, right now it just spits out results to main monitoring script log and its not entirely verbose, it just says volts up or down or no change or w/e =P

I have no idea why ur rig runs like that....

Offline Searing

  • Full Member
  • ***
  • Posts: 312
  • Clueless
Re: NEW KNC TITAN FIRMWARE - 'GenTarkin's Mod' OFFICIAL TOPIC
« Reply #71 on: November 01, 2015, 07:10:04 AM »


 Heh alcheminer II is coming out in SIX MONTHS....so over priced and so too late to the party (for home miners). I like the optimism thou.....would need prob 20 buck LTC again to ROI imho (heh just saying)
 
here is the link to the post on here about such


https://litecointalk.org/index.php?topic=28046.msg288990#msg288990

I can get a 350mh Titan NOW for those prices. In fact I passed on a 1st batch 350mh Titan for 2k on www.litecointalk.org about 3 weeks ago. It just don't make sense at the prices they are asking.

Anyway know this is off topic but it is like a unicorn sighting new HOME asic equipment figured I'd pass it on .so folk (like me) could watch from the sidelines.

https://litecointalk.org/index.php?topic=28046.msg288990#msg288990





Offline GenTarkin

  • Jr. Member
  • **
  • Posts: 94
Re: NEW KNC TITAN FIRMWARE - 'GenTarkin's Mod' OFFICIAL TOPIC
« Reply #72 on: November 07, 2015, 11:53:12 PM »
Little taste of whats to come =)


Offline Mattzees

  • Sr. Member
  • ****
  • Posts: 704
Re: NEW KNC TITAN FIRMWARE - 'GenTarkin's Mod' OFFICIAL TOPIC
« Reply #73 on: November 08, 2015, 12:08:46 AM »
Tiiiiiiiight!

Offline GenTarkin

  • Jr. Member
  • **
  • Posts: 94
Re: NEW KNC TITAN FIRMWARE - 'GenTarkin's Mod' OFFICIAL TOPIC
« Reply #74 on: November 17, 2015, 05:16:44 PM »
Bit of an update on "Energy Saver"   (Automatic voltage per die tuning) ... its now going further testing on other Titans, Im working through the bugs ... fun times!
All in all... Im surprised by how well its working thus far!

Further tweaks on the advanced page:
« Last Edit: November 17, 2015, 05:22:38 PM by GenTarkin »