The lower end of the gpu-engine range is there for a good reason: if cgminer detects overheat that it cannot control with fan alone, it starts clocking down the gpu engine speed (before getting to the hard cutoff speed). Sure 300 is overkill, but it should never get there unless a fan seizes (which one of mine did on a 6970!). I suggest for your GPU2 you drop the speed a little since it's getting hardware errors. As for power consumption, the hardware and OS and therefore software have absolutely no idea about it and only external tools can tell you.
Solo mining is a lousy idea with less than 1% of the total network (lots written about this as to why small solo mining is bad). Yes I am suggesting you currently need 150GH to make it worth solo mining. I suggest you go mine with ozcoin instead. Config file always goes first. Unfortunately it's a curly part of the code which is painful to fix and I don't care enough to do so. Just delete the config file if it's a problem or don't give it parameters...
Topic: linux cgminer quad hd7970 system up (Read 1764 times)
Thanks for your suggestions, overall significant improvement!
For comparison:
--gpu-engine 300-1150 --gpu-powertune 5 --gpu-memdiff -150,-150,-150,-150 --temp-target 70 --temp-cutoff 80 --auto-gpu --auto-fan
Rate has been as high as 2.9Ghash/s but I'm just cruising right now until my high volume case fans arrive in the mail. That will ideally bring me down a few degrees and then I can push a little more.
A couple more questions: on the --gpu-engine option, my GPUs never used the lower end of the range selected. Would there by any reason not to give a single value e.g. --gpu-engine 1150?
Also curious about how cgminer prioritizes command line input vs. config file. It seems if config file exists, then command line parameters are ignored... ? Easy workaround, but maybe I am doing something wrong?
And a random thought: I haven't taken a peek into the source, but cgminer seems to have low level control over the GPUs. Would it be trivial to output real time GPU power consumption?
Oh yes, I did know this! I am grateful to get such helpful input! I have been mining solo but may go pool to get some BTC to tip with in the meantime!!!
Quote
ckolivas:
These are my current parameters:
--gpu-engine 300-1105,300-1145,300-1150,300-1200 --gpu-powertune 5 --gpu-memdiff -150,-150,-150,-150
These are my current parameters:
--gpu-engine 300-1105,300-1145,300-1150,300-1200 --gpu-powertune 5 --gpu-memdiff -150,-150,-150,-150
For comparison:
--gpu-engine 300-1150 --gpu-powertune 5 --gpu-memdiff -150,-150,-150,-150 --temp-target 70 --temp-cutoff 80 --auto-gpu --auto-fan
Code:
GPU 0: 73.0C 5220RPM | 614.0/631.3Mh/s | A:0 R:0 HW: 0 U:0.00/m I:10
GPU 1: 73.0C 5072RPM | 645.6/664.8Mh/s | A:0 R:0 HW: 0 U:0.00/m I:10
GPU 2: 73.0C 5085RPM | 644.5/666.5Mh/s | A:0 R:0 HW:60 U:0.00/m I:10
GPU 3: 69.0C 4113RPM | 678.8/678.7Mh/s | A:0 R:0 HW: 0 U:0.00/m I:10
A couple more questions: on the --gpu-engine option, my GPUs never used the lower end of the range selected. Would there by any reason not to give a single value e.g. --gpu-engine 1150?
Also curious about how cgminer prioritizes command line input vs. config file. It seems if config file exists, then command line parameters are ignored... ? Easy workaround, but maybe I am doing something wrong?
And a random thought: I haven't taken a peek into the source, but cgminer seems to have low level control over the GPUs. Would it be trivial to output real time GPU power consumption?
Quote
ckolivas:
Note that I wrote cgminer, so much of the GPU code is designed around the way I use it on my own hardware
Note that I wrote cgminer, so much of the GPU code is designed around the way I use it on my own hardware
Oh yes, I did know this! I am grateful to get such helpful input! I have been mining solo but may go pool to get some BTC to tip with in the meantime!!!
Yes, I increase the engine clock speed by 5 at a time from the cgminer menu while it's running and then find the upper limit, defined as the speed where it crashes instantly (goes SICK in cgminer), then I take note of that engine speed and usually it's 20-25 lower that it runs stable at. Then I use that setting from then on. If HW errors start to show up without a crash, I decrease it by 5 further every time, until the HW errors go away. The memory can only be set to 150 lower than the engine speed in linux on 7970s, yet every little bit you decrease the memory speed counts in terms of heat and power saved, so I use the cgminer --gpu-memdiff option to do that.
These are my current parameters:
--gpu-engine 300-1105,300-1145,300-1150,300-1200 --gpu-powertune 5 --gpu-memdiff -150,-150,-150,-150
Note that I wrote cgminer, so much of the GPU code is designed around the way I use it on my own hardware
These are my current parameters:
--gpu-engine 300-1105,300-1145,300-1150,300-1200 --gpu-powertune 5 --gpu-memdiff -150,-150,-150,-150
Note that I wrote cgminer, so much of the GPU code is designed around the way I use it on my own hardware
@ ckolivas: I like the internal layout on yours better... with the cards pointing lengthwise to the bottom of case. Unfortunately my setup is with cards parallel to bottom of case so I am guessing heat is accumulating as they stack. GPU 3 is the topmost one, and it usually runs hottest and its fan spins up the fastest. I just ordered a couple more high volume case fans to replace the stock fans that the case arrived with, am hoping that will do the trick.
Or maybe I just need to commandeer the fridge and park the machine there.
I am going to borrow the idea for the spacers.
Interesting... so you turned up the engine speed from default 925 to 1110-1200 and then turned down the memory speed from default 1375 to ~960-1050? Did I understand you correctly? It looks like you reversed the clock settings for the engine and memory.
Comparing with your numbers here, overall I am getting ~770Mhash/s less than you are. That's significant. I am using a comparable board: GA-990FXA-UD7 in terms of slot speed.
If it's not a bother, please confirm I understand your parameters above and I will give it a shot.
Or maybe I just need to commandeer the fridge and park the machine there.
I run 4x7970s plugged directly into the 16x PCI slots on a GD70 motherboard. However they're in a raven3 case which vents upwards, has 18cm fans blowing directly onto the GPUs, I have the side off the case, and I use spacers between the cards for better airflow. The spacers are half clothes peg...
I am going to borrow the idea for the spacers.
Quote
I overclocked on cgminer till I found the upper stable limit for each card. They vary between 1110 and 1200 engine speed and I use the gpu-memdiff option in cgminer with -150. The default settings in cgminer work very well on 7970s since I wrote them specifically for that card.
Interesting... so you turned up the engine speed from default 925 to 1110-1200 and then turned down the memory speed from default 1375 to ~960-1050? Did I understand you correctly? It looks like you reversed the clock settings for the engine and memory.
Code:
GPU 0: 73.0C 3889RPM | 666.5/666.4Mh/s | A:27915 R:337 HW:0 U: 9.21/m I:11
GPU 1: 74.0C 3750RPM | 686.2/685.5Mh/s | A:28575 R:347 HW:0 U: 9.43/m I:11
GPU 2: 73.0C 3397RPM | 692.6/691.3Mh/s | A:29045 R:345 HW:0 U: 9.58/m I:11
GPU 3: 73.0C 2927RPM | 717.6/718.2Mh/s | A:30073 R:352 HW:0 U: 9.92/m I:11
Comparing with your numbers here, overall I am getting ~770Mhash/s less than you are. That's significant. I am using a comparable board: GA-990FXA-UD7 in terms of slot speed.
If it's not a bother, please confirm I understand your parameters above and I will give it a shot.
Have you considered water cooling? I know many who have had success with that.
-MassMaster
-MassMaster
I run 4x7970s plugged directly into the 16x PCI slots on a GD70 motherboard. However they're in a raven3 case which vents upwards, has 18cm fans blowing directly onto the GPUs, I have the side off the case, and I use spacers between the cards for better airflow. The spacers are half clothes peg...
Case:
http://ck.kolivas.org/pictures/Mining/IMG_1402.JPG
Closeup of spacers:
http://ck.kolivas.org/pictures/Mining/IMG_1403.JPG
I overclocked on cgminer till I found the upper stable limit for each card. They vary between 1110 and 1200 engine speed and I use the gpu-memdiff option in cgminer with -150. The default settings in cgminer work very well on 7970s since I wrote them specifically for that card.
Case:
http://ck.kolivas.org/pictures/Mining/IMG_1402.JPG
Closeup of spacers:
http://ck.kolivas.org/pictures/Mining/IMG_1403.JPG
I overclocked on cgminer till I found the upper stable limit for each card. They vary between 1110 and 1200 engine speed and I use the gpu-memdiff option in cgminer with -150. The default settings in cgminer work very well on 7970s since I wrote them specifically for that card.
Code:
GPU 0: 73.0C 3889RPM | 666.5/666.4Mh/s | A:27915 R:337 HW:0 U: 9.21/m I:11
GPU 1: 74.0C 3750RPM | 686.2/685.5Mh/s | A:28575 R:347 HW:0 U: 9.43/m I:11
GPU 2: 73.0C 3397RPM | 692.6/691.3Mh/s | A:29045 R:345 HW:0 U: 9.58/m I:11
GPU 3: 73.0C 2927RPM | 717.6/718.2Mh/s | A:30073 R:352 HW:0 U: 9.92/m I:11
The system:
CCC 12.6
AMD APP SDK v2.7
ADL SDK v4.0
CGMINER v 2.4.2
on Deb 6.0.5
I'm getting 500MH/s/GPU with default settings... but GPU fans are winding out at 4-5000 RPM to keep things cool (~70C average).
Would be interested in hearing what others are doing on Linux systems with same GPUs to:
1) optimize hashing capacity while
2) keeping GPU temperatures low and
3) minimizing overall cooling expense (idealized but realistic... unfortunately I live in a temperate zone, not at the N/S85 parallel and have no control over my kwh utility prices)
I have 8 fans in the box already... but I am realizing that the ones that it came with are trash in terms of volume airflow. I am thinking I need to buy a couple more in the 80+CFM category.
Anyone have any suggestions for cgminer parameters that may help (pls recall above system parameters)? Will sort through the forum in the meantime.
CCC 12.6
AMD APP SDK v2.7
ADL SDK v4.0
CGMINER v 2.4.2
on Deb 6.0.5
I'm getting 500MH/s/GPU with default settings... but GPU fans are winding out at 4-5000 RPM to keep things cool (~70C average).
Would be interested in hearing what others are doing on Linux systems with same GPUs to:
1) optimize hashing capacity while
2) keeping GPU temperatures low and
3) minimizing overall cooling expense (idealized but realistic... unfortunately I live in a temperate zone, not at the N/S85 parallel and have no control over my kwh utility prices)
I have 8 fans in the box already... but I am realizing that the ones that it came with are trash in terms of volume airflow. I am thinking I need to buy a couple more in the 80+CFM category.
Anyone have any suggestions for cgminer parameters that may help (pls recall above system parameters)? Will sort through the forum in the meantime.
Jump to: