Add --tstop and --tstart option preventing GPU overheating. #1146

StefanOberhumer · 2018-05-24T05:40:25Z

As we're getting summer here:

To be sure not to burn the GPUs the miner stops mining on a GPU
if a specific temperature (--tstop) is reached.
After cooling down and reaching --tstart temperature (default 40)
mining will be (re)started on the GPU.

AndreaLanfranchi · 2018-05-24T06:56:15Z

ethminer/MinerAux.h

@@ -690,6 +729,9 @@ class MinerCLI
 			<< "        Use at your own risk! If GPU generates errored results they WILL be forwarded to the pool" << endl
 			<< "        Not recommended at high overclock." << endl
 #endif
+			<< " Temperature management: (implies -HWMON=0|1)" << endl


Description should be set to

"Temperature management: (implies -HWMON=1)"

Would not work on without hwmon

-HWMON=0 will be automatically enabled (m_show_hwmonitors=true which is what -HWMON=0 does).

AndreaLanfranchi · 2018-05-24T07:00:40Z

ethminer/MinerAux.h

+		{
+			try
+			{
+				m_tstop = stol(argv[++i]);


Use stoul instead of stol

AndreaLanfranchi · 2018-05-24T07:01:05Z

ethminer/MinerAux.h

+		{
+			try
+			{
+				m_tstart = stol(argv[++i]);


stoul instead of stol

AndreaLanfranchi · 2018-05-24T07:04:52Z

ethminer/MinerAux.h

+
+
+		// Sanity check --tstart/--tstop
+		if (m_tstop && m_tstop <= m_tstart)


Implement a check for tstop to have a range from tstart of, at least, 10 degrees
Otherwise a tstop of 41 and a tstart of 40 would be legit but will never start mining I assume

Also do not let tstart be lower than 40.
A tstart of 10 would make no sense.

In your example - it will mine if the temperature equals 40 degrees.

Forcing (tstart >= 40) and (tstart + 10 <= tstop):
Hmmm: I like it when software does not too much automatically - the user should have the control !
I force the two values to be in a range of 30 to 100 and 0 for tstop to disable feature again (eg --tstop 60 --tstop 0) ?

AndreaLanfranchi · 2018-05-24T19:18:53Z

libethcore/Miner.h

@@ -241,6 +282,7 @@ class Miner: public Worker
 	std::chrono::high_resolution_clock::time_point workSwitchStart;
 	HwMonitorInfo m_hwmoninfo;
 private:
+	bool     m_wait_for_tstart_temp = false;


Would make m_wait_for_tstart_temp a std::atomic as we're querying from different threads.

Why it must be accessed from different threads? If that's required make it an atomic_flag.

Cause each miner lives in it's own thread and the value is getting accessed by Farm which lives on a separate thread.

@chfast , @AndreaLanfranchi
As I'm new in this topic (C++, C11) - do we really need this?
As m_wait_for_tstart_temp is just updated (modified) in update_temperature() in Miner.h which is called from the farm and I don't think its a critical section - do I have to use atomic_flag ?

Yes, you have to. Otherwise this will be a race condition and Thread Sanitizer will report that as an issue.

AndreaLanfranchi · 2018-05-24T19:33:57Z

Bottomline ... a good catch.
Highly requested.
Almost ready to merge.

chfast · 2018-05-24T19:54:55Z

libethcore/Miner.h

@@ -241,6 +282,7 @@ class Miner: public Worker
 	std::chrono::high_resolution_clock::time_point workSwitchStart;
 	HwMonitorInfo m_hwmoninfo;
 private:
+	bool     m_wait_for_tstart_temp = false;


Why it must be accessed from different threads? If that's required make it an atomic_flag.

To be sure not to burn the GPUs the miner stops mining on a GPU if a specific temperature (--tstop) is reached. After cooling down and reaching --tstart temperature (default 40) mining will be (re)started on the GPU.

StefanOberhumer · 2018-05-25T08:16:35Z

As suggested to use std::atomic_flag for m_wait_for_tstart_temp:

Unlike std::atomic<bool>, std::atomic_flag does not provide load or store operations.
So I decided to use std::atomic<bool> for m_wait_for_tstart_temp flag.

chfast

Very nice!
Sorry about confusion about atomic_flag vs atomic<bool>. I though they are almost the same.

ali8889 · 2018-09-20T15:37:39Z

Goood

AndreaLanfranchi reviewed May 24, 2018

View reviewed changes

ethminer/MinerAux.h Outdated

{

try

{

m_tstop = stol(argv[++i]);

Copy link

Collaborator

AndreaLanfranchi May 24, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Use stoul instead of stol

AndreaLanfranchi reviewed May 24, 2018

View reviewed changes

ethminer/MinerAux.h Outdated

{

try

{

m_tstart = stol(argv[++i]);

Copy link

Collaborator

AndreaLanfranchi May 24, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

stoul instead of stol

AndreaLanfranchi reviewed May 24, 2018

View reviewed changes

StefanOberhumer closed this May 24, 2018

StefanOberhumer deleted the Add-tstop-tstart-Option--PreventingOverheating branch May 24, 2018 11:16

StefanOberhumer restored the Add-tstop-tstart-Option--PreventingOverheating branch May 24, 2018 11:43

StefanOberhumer reopened this May 24, 2018

StefanOberhumer force-pushed the Add-tstop-tstart-Option--PreventingOverheating branch 2 times, most recently from 0055d4e to bcd9099 Compare May 24, 2018 12:33

AndreaLanfranchi suggested changes May 24, 2018

View reviewed changes

chfast suggested changes May 24, 2018

View reviewed changes

Add --tstop and --tstart option preventing GPU overheating.

b0eb122

To be sure not to burn the GPUs the miner stops mining on a GPU if a specific temperature (--tstop) is reached. After cooling down and reaching --tstart temperature (default 40) mining will be (re)started on the GPU.

StefanOberhumer force-pushed the Add-tstop-tstart-Option--PreventingOverheating branch from bcd9099 to b0eb122 Compare May 25, 2018 08:06

chfast approved these changes May 25, 2018

View reviewed changes

AndreaLanfranchi approved these changes May 25, 2018

View reviewed changes

chfast merged commit 8cc88d9 into ethereum-mining:master May 25, 2018

StefanOberhumer deleted the Add-tstop-tstart-Option--PreventingOverheating branch May 25, 2018 10:35

AndreaLanfranchi mentioned this pull request May 25, 2018

Stop overheated cards function #978

Closed

StefanOberhumer mentioned this pull request May 27, 2018

Display info when --tstop/--tstart temperatures are reached. #1159

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add --tstop and --tstart option preventing GPU overheating. #1146

Add --tstop and --tstart option preventing GPU overheating. #1146

StefanOberhumer commented May 24, 2018

AndreaLanfranchi May 24, 2018

StefanOberhumer May 24, 2018

AndreaLanfranchi May 24, 2018

AndreaLanfranchi May 24, 2018

AndreaLanfranchi May 24, 2018

AndreaLanfranchi May 24, 2018

StefanOberhumer May 24, 2018 •

edited

Loading

AndreaLanfranchi May 24, 2018

chfast May 24, 2018

AndreaLanfranchi May 24, 2018

StefanOberhumer May 25, 2018

chfast May 25, 2018

AndreaLanfranchi commented May 24, 2018

chfast May 24, 2018

StefanOberhumer commented May 25, 2018

chfast left a comment

ali8889 commented Sep 20, 2018



		// Sanity check --tstart/--tstop
		if (m_tstop && m_tstop <= m_tstart)

Add --tstop and --tstart option preventing GPU overheating. #1146

Add --tstop and --tstart option preventing GPU overheating. #1146

Conversation

StefanOberhumer commented May 24, 2018

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

StefanOberhumer May 24, 2018 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

AndreaLanfranchi commented May 24, 2018

Choose a reason for hiding this comment

StefanOberhumer commented May 25, 2018

chfast left a comment

Choose a reason for hiding this comment

ali8889 commented Sep 20, 2018

StefanOberhumer May 24, 2018 •

edited

Loading