BAMT version 0.5 - Easy USB based mining Linux with farm wide management tools

[boozer]

If it is stable at stock and not at higher then it is simply excessive overclock.  Lower memclock won't make it more stable but it does use less wattage.  300 is commonly used number but it is actually a very bad memclock.  It is actually slower than both 280 and 310.

When you find clocks that are stable for 24 hours you likely aren't there yet.  Eventually the rig will crash.  When it does drop the core clock 5 Mhz to 10 Mhz on the affected GPU and reboot.  One by one you will find the stable clocks.  Now the rig may run 15 days or so and crash.  You can either accept that or drop clocks another 5 Mhz or so.  Eventually you will find the speed that runs 24/7 for 90+ days.

Thanks.  The weird thing was that gpu0 always seemed to be the victim of the overclock (even if it was not OC'd), which was confusing me.  I ran stock just fine, then OC'd... and cores 2-5 would appear to be fine as if it rebooted, it set gpu0 to noOC.  So I set gpu0 and 1 to stock, but still rebooted, always adding gpu0 to the noOC ACTIVE directory, which made me think it was a bad card (since it looked to be causing reboots at stock clocks)... but everything has been stable so far (a lot longer than normal) with all cards at stock.  So it seems that OC'ing 2-5 caused 0-1 to become unstable (even if 0-1 were running stock) and that was throwing me for a loop.

[1714016028]

1714016028

[1714016028]

1714016028

[lodcrappo]

Is there any possibility that a failure on "any" GPU gets reported as GPU0 issue, regardless which gpu it was?  I removed the card i thought was bad... so only had 4 GPU's, but the new gpu0 again started having the same problem.  

So I added the card I thought had issues back into the mix and set every card to stock and it has been running fine.  So before, gpu0 running stock would die, but with all other gpus at stock, everything seems to be fine.... its only been 30 minutes, but thats longer than it used to last, lol.  Just thought I would ask about the gpu0 theory, but maybe they both had problems.... I'll check in the AM to see if it stayed up the rest of the night.

It is possible, though I haven't heard of it.  Anything is possible when you are locking up hardware.  It is not a situation where there is much control.

When you add up the time lost in mining from people locking up their GPUs or rigs and the value of the time a person puts into overclocking, I think most people come away with a net loss.  Had they just left everything stock, they would have been solid mining 24/7, probably more actual shares mined anyway, not to mention the time they would have saved.  Don't even get me started on people overclocking for highest mhash number on the screen and not even looking at what they've done to their actual shares submitted rate, or the power per hash ratio.

[malevolent]

Don't tell me you are running your GPUs at stock clocks 

[lodcrappo]

Don't tell me you are running your GPUs at stock clocks  

No, but I don't have the trouble that most people seem to with keeping their cards mining.  If I was losing mining time every day (or every month) due to overclocking, or had to spend more than 10 minutes on it in the life of a rig, I would consider it a loss.

We did some math in the IRC channel..  every 1 Mhash you squeeze out of your GPU is a whopping $1.25 PER YEAR.  (at current diff and btc price).  Not really worth putting much time into.

[boozer]

It is possible, though I haven't heard of it.  Anything is possible when you are locking up hardware.  It is not a situation where there is much control.

When you add up the time lost in mining from people locking up their GPUs or rigs and the value of the time a person puts into overclocking, I think most people come away with a net loss.  Had they just left everything stock, they would have been solid mining 24/7, probably more actual shares mined anyway, not to mention the time they would have saved.  Don't even get me started on people overclocking for highest mhash number on the screen and not even looking at what they've done to their actual shares submitted rate, or the power per hash ratio.

Good points.  I was able find stable OC's on my other rig very easily... not sure why this one is giving me so much grief, lol.  I just wasnt checking the other OC's as gpu0 was always indicated as the one with the issue, now that I disregard that, I should have more luck I'm guessing.  I just got stuck on that fact that gpu0 kept getting returned to stock.... now that i know that I should not always use that as an indicator, I bet I'm good to go as revisit my other OC's (once its run stock stable for awhile).  Still going strong.

[lodcrappo]

It is possible, though I haven't heard of it.  Anything is possible when you are locking up hardware.  It is not a situation where there is much control.

When you add up the time lost in mining from people locking up their GPUs or rigs and the value of the time a person puts into overclocking, I think most people come away with a net loss.  Had they just left everything stock, they would have been solid mining 24/7, probably more actual shares mined anyway, not to mention the time they would have saved.  Don't even get me started on people overclocking for highest mhash number on the screen and not even looking at what they've done to their actual shares submitted rate, or the power per hash ratio.

Good points.  I was able find stable OC's on my other rig very easily... not sure why this one is giving me so much grief, lol.  I just wasnt checking the other OC's as gpu0 was always indicated as the one with the issue, now that I disregard that, I should have more luck I'm guessing.  I just got stuck on that fact that gpu0 kept getting returned to stock.... now that i know that I should not always use that as an indicator, I bet I'm good to go as revisit my other OC's (once its run stock stable for awhile).  Still going strong.

Well I don't know if you can disregard GPU0 being flagged or not.  If it is being flagged, it *is* locked up.  But it may be that having one batshit crazy GPU in the system can cause the others to lock up.  I do not know.  BAMT flags the first locked up phoenix process it finds, starting at 0 and working up.

[boozer]

Don't tell me you are running your GPUs at stock clocks  

No, but I don't have the trouble that most people seem to with keeping their cards mining.  If I was losing mining time every day (or every month) due to overclocking, or had to spend more than 10 minutes on it in the life of a rig, I would consider it a loss.

We did some math in the IRC channel..  every 1 Mhash you squeeze out of your GPU is a whopping $1.25 PER YEAR.  (at current diff and btc price).  Not really worth putting much time into.

More good points...  I'm just starting and learning... probably a little too eager to get the max hash/share OC (no overvolt).  The first rig went so easy, I guess it made me stubborn on this one, lol.  But I am tired of messing with it, so I'll just let it run at stock for awhile and then try some minor overclocks when/if I feel like messing with it again.

[boozer]

Well I don't know if you can disregard GPU0 being flagged or not.  If it is being flagged, it *is* locked up.  But it may be that having one batshit crazy GPU in the system can cause the others to lock up.  I do not know.  BAMT flags the first locked up phoenix process it finds, starting at 0 and working up.

Ahh gotya.  Has anyone had a problem where raising clocks on one card affect the stability of another?  'Cause it kinda seems to be what happens in this case then if GPU0 *is* definitely getting locked.

[jamesg]

Well I don't know if you can disregard GPU0 being flagged or not.  If it is being flagged, it *is* locked up.  But it may be that having one batshit crazy GPU in the system can cause the others to lock up.  I do not know.  BAMT flags the first locked up phoenix process it finds, starting at 0 and working up.

Ahh gotya.  Has anyone had a problem where raising clocks on one card affect the stability of another?  'Cause it kinda seems to be what happens in this case then if GPU0 *is* definitely getting locked.

If you raise your clocks too high on one card, it could bring down the entire rig. This has happened to me before.

[lodcrappo]

Well I don't know if you can disregard GPU0 being flagged or not.  If it is being flagged, it *is* locked up.  But it may be that having one batshit crazy GPU in the system can cause the others to lock up.  I do not know.  BAMT flags the first locked up phoenix process it finds, starting at 0 and working up.

Ahh gotya.  Has anyone had a problem where raising clocks on one card affect the stability of another?  'Cause it kinda seems to be what happens in this case then if GPU0 *is* definitely getting locked.

If you raise your clocks too high on one card, it could bring down the entire rig. This has happened to me before.

Yeah this is true.  I never 100% proved it, but I had one rig where if I overclocked GPU2 past a certain rate, GPU3 would lock up (a 6870 and a 5830).  True story, weird as hell.  GPU2 would run at XXX mhz for days, GPU3 would run at YYY mhz for days, but if I stuck them together, GPU3 locks up at YYY when GPU2 is at XXX.  reducing either one by 10mhz made the whole system stable.  go figure.  then the fan on the 6870 died and I never completely proved that theory.

[DeathAndTaxes]

We did some math in the IRC channel..  every 1 Mhash you squeeze out of your GPU is a whopping $1.25 PER YEAR.  (at current diff and btc price).  Not really worth putting much time into.

I have 48 cores.  A 20 MH/s boost would be worth $1200.  It is a good point though.  No sense is running rigs on the bleeding edge.  Still most cards can be comfortably overclocked without being right on the edge.  I mean a 5970 stock clock is 725/1000.  At 800/240 it runs cooler and provides (by your math above) $100 more per year.  Now the difference between 800 and 820 and 820 and 837.5 starts becoming negligible real quick.

Grab the big pieces of meat.

[boozer]

Yeah this is true.  I never 100% proved it, but I had one rig where if I overclocked GPU2 past a certain rate, GPU3 would lock up (a 6870 and a 5830).  True story, weird as hell.  GPU2 would run at XXX mhz for days, GPU3 would run at YYY mhz for days, but if I stuck them together, GPU3 locks up at YYY when GPU2 is at XXX.  reducing either one by 10mhz made the whole system stable.  go figure.  then the fan on the 6870 died and I never completely proved that theory.

That is exactly what I'm seeing. 

I was trying to jump to 800/240 on the clocks as this should normally be pretty easily attainable on the 5970's, but it appears i have one (or more) gpu's that dont like it...   I'll just have to do the very slow increments, one core at a time thing starting at stock speeds instead of 800... which will take forever, but I think a majority of them will run at least at 800, just got to find the card that wont :/

[lodcrappo]

We did some math in the IRC channel..  every 1 Mhash you squeeze out of your GPU is a whopping $1.25 PER YEAR.  (at current diff and btc price).  Not really worth putting much time into.

I have 48 cores.  A 20 MH/s boost would be worth $1200.  It is a good point though.  No sense is running rigs on the bleeding edge.  Still most cards can be comfortably overclocked without being right on the edge.  I mean a 5970 stock clock is 725/1000.  At 800/240 it runs cooler and provides (by your math above) $100 more per year.  Now the difference between 800 and 820 and 820 and 837.5 starts becoming negligible real quick.

Grab the big pieces of meat.

Most people with a decent sized farm don't fall into the overclocking trap, because they simply cannot afford to.  Where you see the big loss is the guys with maybe 5-6 GPUs and they are fighting for the last 5 or so Mhash on them..  in other words hours spent chasing ~$30 a year.  Yet crashing the rig for half a day when they are out = more lost than they'll ever get back even if they do get o/c working stable later.  It's a sad cycle.

Friends don't let friends over invest in overclocking.

[malevolent]

No, but I don't have the trouble that most people seem to with keeping their cards mining.  If I was losing mining time every day (or every month) due to overclocking, or had to spend more than 10 minutes on it in the life of a rig, I would consider it a loss.
We did some math in the IRC channel..  every 1 Mhash you squeeze out of your GPU is a whopping $1.25 PER YEAR.  (at current diff and btc price).  Not really worth putting much time into.

What counts ultimately is the total valid share count, if you extrapolate MHash/s gain to every GPU in a farm you can get a nice sum. I overclock my GPUs 10 MHz below the verge of instability, as for overvolting this is only fine with good cooling and cheap/free electricity.

[NetworkerZ]

Hiho!

Switched with one RIG form Win7 to BAMT! Seems great so far, THX for your work! Is it possible to put one GPU on one worker? If yes, how?

THX and Greetz
NetworkerZ

[DeathAndTaxes]

Friends don't let friends over invest in overclocking.

To each his own.  I haven't seen any higher stability at stock as I have at 800 MHz.  Seems silly to leave $1200 on the table. I do agree the last few % are not worth it and are best saved for single GPU miners who want the bragging rights.

[Splirow]

Friends don't let friends over invest in overclocking.

To each his own.  I haven't seen any higher stability at stock as I have at 800 MHz.  Seems silly to leave $1200 on the table. I do agree the last few % are not worth it and are best saved for single GPU miners who want the bragging rights.

What settings do you run your rig with 5970?

I have 3 5970's in 1 rig at 825/300 getting 382m/h

[lodcrappo]

Friends don't let friends over invest in overclocking.

To each his own.  I haven't seen any higher stability at stock as I have at 800 MHz.  Seems silly to leave $1200 on the table. I do agree the last few % are not worth it and are best saved for single GPU miners who want the bragging rights.

we are saying the same thing.  I don't mean all overclocking is pointless.  I mean chasing that last bit is dumb.  Obviously if you can make $1200 a year for less time than $1200 means to you, and without jeopardizing your entire profits with instability, then it is worth it.  But a guy spending hours of his own time or creating hours of downtime for his rigs all trying to get $10 more a year is just silly.  You don't have to look far on these forums to find people who have lost a lot more playing the OC game then they will ever earn back in a lifetime of mining.  My point is just that you should consider the cost compared to the earnings.  An unstable rig costs you a lot of money and the last 2-3% overclock earns you very little.

[lodcrappo]

Hiho!

Switched with one RIG form Win7 to BAMT! Seems great so far, THX for your work! Is it possible to put one GPU on one worker? If yes, how?

THX and Greetz
NetworkerZ

You mean use different URL for each GPU?  Just make different pools file for each one. 

[lodcrappo]

Its a wide open rig, similar to the one shown in the "build your own" hardware section of this forum.  I think I ran everything everything at stock for 24 hours, but its been awhile, so I'll go back to that and set mem at 240 and see if stock is stable or not.

If it is stable at stock and not at higher then it is simply excessive overclock.  Lower memclock won't make it more stable but it does use less wattage.  300 is commonly used number but it is actually a very bad memclock.  It is actually slower than both 280 and 310.

When you find clocks that are stable for 24 hours you likely aren't there yet.  Eventually the rig will crash.  When it does drop the core clock 5 Mhz to 10 Mhz on the affected GPU and reboot.  One by one you will find the stable clocks.  Now the rig may run 15 days or so and crash.  You can either accept that or drop clocks another 5 Mhz or so.  Eventually you will find the speed that runs 24/7 for 90+ days.

BTW I agree with this 100%.  In my experience (back when I had tons of GPUs), it pretty much always works out as you say.  You'll find a "stable" clock but it will crash maybe once a day or 2-3 days.  Drop back a bit and you'll be stable for maybe 2 weeks.  Drop back another bit and you're stable forever.  These points will often not be the same for even two identical cards.