OFFICIAL CGMINER mining software thread for linux/win/osx/mips/arm/r-pi 4.11.1

[jake262144]

There are two kinds of safeguards (fan speed configuration is excluded to simulate a catastrophic fan failure):
(1)  soft, speed throttling is supposed to keep the GPU temperature below target temperature:
       + "gpu-engine" : "400-1005",
       + "auto-gpu" : true,
       + "temp-overheat" : "76",
       + "temp-target" : "72",
       + "temp-hysteresis" : "2",

(2)  hard, a measure of last resort to switch the device off before it goes critical:
      + "temp-cutoff" : "81",

Once your card exceeds (temp-target + temp-hysteresis) two things may happen: the fan will speed up to maximum defined fan speed ("gpu-fan" : "52-57") and the core will be throttled down.
Obviously for this to work, auto-gpu and auto-fan need to be enabled. Gpu-fan needs to have a percentage range defined to change the fan speed.

Once your card exceeds temp-overheat two things may happen: the fan will speed up to 100% (we can forget that in our doomsday scenario) and the card will maximally throttle its core down according to gpu-engine.
auto-gpu is mandatory for the engine to be downclocked.
auto-fan is mandatory for the fan to be speeded up to 100%
If gpu-engine is defined as a static value the GPU will only underclock to its stock speed. Otherwise, the card will underclock as far as gpu-engine allows.

temp-cutoff absolutely depends on auto-gpu being set. It does not require a gpu-engine range to be defined.

Normal fan speeds ("auto-fan" : true, "gpu-fan" : "60-65") are only used to prevent the card from hitting temp-overheat. If gpu-fan is set too low and the card hits the temp-overheat threshold you'll definitely hear the fan going in the self-preservation mode.


Examples: (ver 1.0.42a)
"auto-gpu" : false, "auto-fan" : false - your card will burn if it exceeds critical temperature. All protections are inactive.
"auto-gpu" : true, "gpu-engine" : "942", "auto-fan" : false - the card will start throttling core speed on exceeding (temp-target + temp-hysteresis). On exceeding temp-overheat the core will be throttled down to its stock speed. On exceeding "temp-cutoff" mining threads will be disabled. The fan will keep steady on temp-overheat.
"auto-gpu" : true, "gpu-engine" : "400-942", "auto-fan" : false - as in previous example, except on exceeding temp-overheat core will be throttled down to 400MHz.

"auto-gpu" : false, "auto-fan" : true - your card will burn on catastrophic fan failure. The fan will speed up to 100% on temp-overheat.
"auto-gpu" : true, "gpu-engine" : "942", "auto-fan" : true, "gpu-fan" : "60" - the card will start throttling core speed on exceeding (temp-target + temp-hysteresis). On exceeding temp-overheat the core will be throttled down to its stock speed and the fan will speed up to 100%.  On exceeding "temp-cutoff" mining threads will be disabled.
"auto-gpu" : true, "gpu-engine" : "942", "auto-fan" : true, "gpu-fan" : "55-65" - as in previous example, except on exceeding (temp-target + temp-hysteresis), the card will start throttling core speed and will start raising the fan speed up to 65%.

[1713498886]

1713498886

[DeathAndTaxes]

      + "temp-cutoff" : "79",

The question is will temp-cutoff be enforced even if auto-gpu is false?
What about temp-overheat?

I think they are just would be good to get confirmation from the code before someone relies on that.

[gnar1ta$]

      + "temp-cutoff" : "79",

The question is will temp-cutoff be enforced even if auto-gpu is false?
What about temp-overheat?

I think they are just would be good to get confirmation from the code before someone relies on that.

From what I understand temp-overheat is only used to move the fan from 85% to 100% only when auto-fan is used.  If auto-fan and auto-gpu are set then the gpu clocks will be lowered first before the fan jumps to 100%.  Which would lead me to believe temp-cutoff is not set without auto-gpu. If I set auto-gpu and gpu-engine X (no range) will it keep a static engine speed?

[jake262144]

Damn the haste  
Post fixed now, sorry guys.

[DeathAndTaxes]

temp-cutoff does not require a gpu-engine range to be defined but it absolutely depends on auto-gpu being set.

If that is true I wonder if it should be changed.  Most people looking to set a static gpu speed likely are interested in failsafe protection.

I would imagine having temp-cutoff always active to be the safer solution.  If someone wants to risk their GPU they could always set a temp-cutoff value of say 300.  That would require the user to essentially say "I don't care if it causes my GPU to melt never reduce speed or bring GPU offline".

[bulanula]

temp-cutoff does not require a gpu-engine range to be defined but it absolutely depends on auto-gpu being set.

If that is true I wonder if it should be changed.  Most people looking to set a static gpu speed likely are interested in failsafe protection.

I would imagine having temp-cutoff always active to be the safer solution.  If someone wants to risk their GPU they could always set a temp-cutoff value of say 300.  That would require the user to essentially say "I don't care if it causes my GPU to melt never reduce speed or bring GPU offline".

I agree there is no reason to disable this feature and should be hardcoded. If the mad guys wanna fry their GPUs then let them compile it with off option etc.

[Gabi]

Tried the --gpu-memclock on a 6950, it doesn't work. Memory clock doesn't change when i start cgminer.
While --gpu-engine work perfectly and change the core clock.

Oh well not a problem, i'll still use msi afterburner for changing clocks

Except that, cgminer work without problems.

[P4man]

temp-cutoff does not require a gpu-engine range to be defined but it absolutely depends on auto-gpu being set.

If that is true I wonder if it should be changed.  Most people looking to set a static gpu speed likely are interested in failsafe protection.

I would imagine having temp-cutoff always active to be the safer solution.  If someone wants to risk their GPU they could always set a temp-cutoff value of say 300.  That would require the user to essentially say "I don't care if it causes my GPU to melt never reduce speed or bring GPU offline".

+10 !

I just had a fan failure and my card was cooking. I mistakenly assumed it would cut off at cut off temp, but I had not set dynamic clocks, so to my great surprise, it did not shut down  

[gnar1ta$]

Tried the --gpu-memclock on a 6950, it doesn't work. Memory clock doesn't change when i start cgminer.
While --gpu-engine work perfectly and change the core clock.

Oh well not a problem, i'll still use msi afterburner for changing clocks

Except that, cgminer work without problems.

6950's have a memclock limit of engine -125.  Try setting --gpu-memdiff -125 or just make sure the mem clock is no more than 125 less then engine clock.

[Gabi]

But afterburner let me set it at like 300mhz, without problems. And gpu-z too confirm that memory is at 300mhz. Or both lie?

[bulanula]

temp-cutoff does not require a gpu-engine range to be defined but it absolutely depends on auto-gpu being set.

If that is true I wonder if it should be changed.  Most people looking to set a static gpu speed likely are interested in failsafe protection.

I would imagine having temp-cutoff always active to be the safer solution.  If someone wants to risk their GPU they could always set a temp-cutoff value of say 300.  That would require the user to essentially say "I don't care if it causes my GPU to melt never reduce speed or bring GPU offline".

+10 !

I just had a fan failure and my card was cooking. I mistakenly assumed it would cut off at cut off temp, but I had not set dynamic clocks, so to my great surprise, it did not shut down  

So does this mean that if I say, set the clocks manually to 960 core then my card will fry if the fans goes out while I am not around ?

[DeathAndTaxes]

But afterburner let me set it at like 300mhz, without problems. And gpu-z too confirm that memory is at 300mhz. Or both lie?

It likely didn't.

GPU-z main tab = what card is SET at.
GPU-Z sensor tab = what card is running at.
cgminer GPU display = what card is running at.

I think you will find the sensor tab on GPU-Z shows a different thing.
Alternative if you "set" it 300 Mhz in GPU-Z and then look into cgminer it will show it at stock.

[DeathAndTaxes]

So does this mean that if I say, set the clocks manually to 960 core then my card will fry if the fans goes out while I am not around ?

Yeah hopefully conman can confirm.
a) --temp-cutoff is always active
or
b) it isn't but it can be made always active in next version.

Looking at the code briefly it looks like --temp-cutoff is only checked if --auto-gpu is enabled.

https://github.com/ckolivas/cgminer/blob/master/adl.c

[jake262144]

So does this mean that if I say, set the clocks manually to 960 core then my card will fry if the fans goes out while I am not around ?

This means, if you don't set auto-gpu and your fans fail when you're not around, firefighters might arrive.
See the example configuration

Yeah hopefully conman can confirm.
a) --temp-cutoff is always active
or
b) it isn't but it can be made always active in next version.

If it is b I think as a hack around you could set "auto-gpu" and a "gpu-engine" range of like "799-800".

It's b except you don't need to meddle with gpu-range. auto-gpu and gpu-range "800" will do the trick.
Auto-gpu and gpu-range "400-800" will be even better.

Conman is a kernel hacker, I believe temp-cutoff depends on auto-gpu because he wanted to give the user full control over the hardware and make cgminer bulletproof against devices with failing temperature sensors...  or he just made a mistake nesting all those ifs

[bulanula]

OK. Thank you very much for these revelations jack ! I will apply them right now and hope that this will indeed work as it is supposed to.

Anyone wanna try and jam the GPU fan for a "simulation" so we can have confirmation ?

Thanks !

[gnar1ta$]

"auto-gpu" : true, "gpu-engine" : "942", "auto-fan" : true - the card will throttle itself down to its stock speed on exceeding temp-overheat, mining threads will be disabled on exceeding "temp-cutoff". The fan will speed up to 100% on temp-overheat.

I don't think this is right.  I use auto-gpu and auto-fan on a rig and if fan is at 85% gpu is throttled at target-temp (75 default) + temp-hysterisis (3 default).  I think temp-overheat is just used to move the fan to 100% after other options fail.  If that doesn't work temp-cutoff is the failsafe.

[DeathAndTaxes]

Conman is a kernel hacker, I believe temp-cutoff depends on auto-gpu because he wanted to give the user full control over the hardware and make cgminer bulletproof against devices with failing temperature sensors...  or he just made a mistake nesting all those ifs

I understand the logic but I do believe it is counter intuitive and cgminer would be safer is
a) temp-cutoff is always checked.
b) default temp-cutoff value is some high but safe number like 90C

thus if user wants to push cards higher they have to explicitly set a higher value.  If user doesn't (and thus might be unaware) the miner will protect him from himself.

[jake262144]

OK. Thank you very much for these revelations jack ! I will apply them right now and hope that this will indeed work as it is supposed to.
Anyone wanna try and jam the GPU fan for a "simulation" so we can have confirmation ?

Jam? Easier to just unplug the fan...

[DeathAndTaxes]

"auto-gpu" : true, "gpu-engine" : "942", "auto-fan" : true - the card will throttle itself down to its stock speed on exceeding temp-overheat, mining threads will be disabled on exceeding "temp-cutoff". The fan will speed up to 100% on temp-overheat.

I don't think this is right.  I use auto-gpu and auto-fan on a rig and if fan is at 85% gpu is throttled at target-temp (75 default) + temp-hysterisis (3 default).  I think temp-overheat is just used to move the fan to 100% after other options fail.  If that doesn't work temp-cutoff is the failsafe.

+1.  That is correct ... I think.

Clock is lowered starting at temp-target + hysteria (or higher).
Clock is raised at temp-target - hysteria (or lower).
Fan is maxed at temp-overheat.
GPU is shutdown at temp-cutoff.

Thus if one wanted a gpu to act like it has a static frequency BUT have failsafe you could use setting like this:

Code:

"auto-gpu" : true  <- Required to engage safety features
"gpu-engine" : "800",   <- Desired static clock
"gpu-fan" : "80" <- Desired static fan value
"temp-hysteresis" : "3",    <- Max amount temp can exceed a value before action taken.
"temp-target" : "85",       <- Set this HIGHER than optimal but at what you consider a safe value
"temp-cutoff" : "90",       <- You could set this same as target.
"temp-overheat" : "95",   <- You could set this the same as the target and/or cutoff.

Outcome:
The GPU will run at a static 800Mhz and 80% fan if the temp is 0C to 87.9C.
At 88C the card will downclock to avoid higher temps (85C target + 3C hysteria).
At 93C the card will max fan to avoid higher temps (90C cutoff + 3C hysteria).
At 98C the card will shutdown and not restart to avoid higher temps (95C overheat +3C hysteria)

For simply a failsafe with nothing else set the params to same number.

Code:

"auto-gpu" : true  <- Required to engage safety features
"gpu-engine" : "800",   <- Desired static clock
"gpu-fan" : "80" <- Desired static fan value
"temp-hysteresis" : "3",    <- Max amount temp can exceed a value before action taken.
"temp-target" : "97",       <- Set all 3 values at what you consider max safe temp.
"temp-cutoff" : "97",       <- Set all 3 values at what you consider max safe temp.
"temp-overheat" : "97",   <- Set all 3 values at what you consider max safe temp.

Outcome:
The GPU will run at a static 800Mhz and 80% fan if the temp is 0C to 99.9C.
Then at 100C the card will lower the clock, raise the fan to max, AND shutdown (thus the first two are not really used).
You get a "semi-static" frequency but all the safety of auto-gpu.

I think this is a good workaround but I feel --temp-cutoff should always be active.

Edited for clarification (thanks Jake).

[jake262144]

"auto-gpu" : true, "gpu-engine" : "942", "auto-fan" : true - the card will throttle itself down to its stock speed on exceeding temp-overheat, mining threads will be disabled on exceeding "temp-cutoff". The fan will speed up to 100% on temp-overheat.

I don't think this is right.  I use auto-gpu and auto-fan on a rig and if fan is at 85% gpu is throttled at target-temp (75 default) + temp-hysterisis (3 default).  I think temp-overheat is just used to move the fan to 100% after other options fail.  If that doesn't work temp-cutoff is the failsafe.

EDIT: You were correct. I used poorly-chosen example temperatures as my hysteresis had been set to 4 all the time. 59 + 4 = 63 = temp-overheat

Both core clock and fan speed are being changed at the same time:
[2012-02-02 20:47:46] Overheat detected on GPU 1, increasing fan to 100%
[2012-02-02 20:47:46] Overheat detected, decreasing GPU 1 clock speed

Options used:"auto-gpu" : true, "gpu-engine" : "942", "auto-fan" : true, "gpu-fan" : "52-57", "temp-target" : "59", "temp-overheat" : "63", "temp-cutoff" : "79,67",