Re B580 power use, it's interesting that your card is using a lot less power, and similarly seems to be running slower than the sample earlier in this thread from Mumak. I wonder if he was overclocking or anything?
he seemed to be getting runtimes less than 500s and running ~111W.
Re B580 power use, it's interesting that your card is using a lot less power, and similarly seems to be running slower than the sample earlier in this thread from Mumak. I wonder if he was overclocking or anything?
he seemed to be getting runtimes less than 500s and running ~111W.
I noticed that too and was very curious about that. Can't say i have any explanation as of yet. Going to investigate tomorrow.
Also, I can't remember is A4500 gpus are running 2x or 3x.. too many numbers in my head..
No concrete conclusions about the power usage being so different. I did a simple overclock based on what others were doing/posting online for these GPUs, and I saw a ~10% increase in wattage (peaked ~80w). I know I could push the speeds further but this is what others were commonly overclocking to. I am not trying to overclock when running E@H, I am trying to match the reported wattage earlier in the thread.
What is interesting is that when I do a GPU benchmark, it will hit 130w easily. The GPU is not using the "compute" function of the GPU, just rendering. I then bumped the MeerKAT work to 2x and it still hovered around ~73w.
I am running a few different benchmarks so I can see how the wattage fluctuates. The GPU has two 8-pin power cords. I wonder if one of them is not working? Would the GPU power on with one 8-pin instead of two? I know 8-pin is 150 watts, so this crossed my mind.
At least some cards for at least some connections actively report during boot-up if a connection is not active.
I rather doubt that it would silently cut power consumption because of a failure to connect one of two 8-pin connectors, but I can't be sure.
I don't spot obvious reporting for 2X runs in the tasks list. Maybe those have not been reported yet?
Mumak declined to try 2X on the ground that the reported GPU utilization at 1X was so extremely high as to make it unlikely 2X would add production to 1X.
We've seen very high GPU chip and host CPU capability dependence on the improvement or lack of improvement from higher multiplicity. We'll all be interested to see what happens on your card for 2X on BRP7 running an opencl app under Windows.
You might find it interesting to check GPU utilization. As it may vary all over the place in the short term, I'd suggest using a monitor which can average between button clicks. For example GPU-Z or HWiNFO (which by the way is Mumak's work.
I don't spot obvious reporting for 2X runs in the tasks list. Maybe those have not been reported yet?
Mumak declined to try 2X on the ground that the reported GPU utilization at 1X was so extremely high as to make it unlikely 2X would add production to 1X.
We've seen very high GPU chip and host CPU capability dependence on the improvement or lack of improvement from higher multiplicity. We'll all be interested to see what happens on your card for 2X on BRP7 running an opencl app under Windows.
Definite improvement running 2x. They are starting to show up and validate on the host (I sometimes temporarily disconnect a host so I can just look at completed work units/times in BOINC when trying something new or different). GPU shows consistent 100% utilization when running 2x versus the drops in utilization when running 1x. Highly doubt 3x will show any improvement, but will be trying this tomorrow.
Edit: 3x crashed the gpu... plenty of vram left and temperature headroom, so must have been driver related. It didn't like that at all. However, 2x was far superior to 1x.
Re B580 power use, it's
)
Re B580 power use, it's interesting that your card is using a lot less power, and similarly seems to be running slower than the sample earlier in this thread from Mumak. I wonder if he was overclocking or anything?
he seemed to be getting runtimes less than 500s and running ~111W.
_________________________________________________________________________
Ian&Steve C. wrote:Re B580
)
I noticed that too and was very curious about that. Can't say i have any explanation as of yet. Going to investigate tomorrow.
Also, I can't remember is A4500 gpus are running 2x or 3x.. too many numbers in my head..
Idle wattage is ~7w.
No concrete conclusions about
)
No concrete conclusions about the power usage being so different. I did a simple overclock based on what others were doing/posting online for these GPUs, and I saw a ~10% increase in wattage (peaked ~80w). I know I could push the speeds further but this is what others were commonly overclocking to. I am not trying to overclock when running E@H, I am trying to match the reported wattage earlier in the thread.
What is interesting is that when I do a GPU benchmark, it will hit 130w easily. The GPU is not using the "compute" function of the GPU, just rendering. I then bumped the MeerKAT work to 2x and it still hovered around ~73w.
I am running a few different benchmarks so I can see how the wattage fluctuates. The GPU has two 8-pin power cords. I wonder if one of them is not working? Would the GPU power on with one 8-pin instead of two? I know 8-pin is 150 watts, so this crossed my mind.
At least some cards for at
)
At least some cards for at least some connections actively report during boot-up if a connection is not active.
I rather doubt that it would silently cut power consumption because of a failure to connect one of two 8-pin connectors, but I can't be sure.
I don't spot obvious reporting for 2X runs in the tasks list. Maybe those have not been reported yet?
Mumak declined to try 2X on the ground that the reported GPU utilization at 1X was so extremely high as to make it unlikely 2X would add production to 1X.
We've seen very high GPU chip and host CPU capability dependence on the improvement or lack of improvement from higher multiplicity. We'll all be interested to see what happens on your card for 2X on BRP7 running an opencl app under Windows.
You might find it interesting to check GPU utilization. As it may vary all over the place in the short term, I'd suggest using a monitor which can average between button clicks. For example GPU-Z or HWiNFO (which by the way is Mumak's work.
archae86 wrote:I don't spot
)
Definite improvement running 2x. They are starting to show up and validate on the host (I sometimes temporarily disconnect a host so I can just look at completed work units/times in BOINC when trying something new or different). GPU shows consistent 100% utilization when running 2x versus the drops in utilization when running 1x. Highly doubt 3x will show any improvement, but will be trying this tomorrow.
Edit: 3x crashed the gpu... plenty of vram left and temperature headroom, so must have been driver related. It didn't like that at all. However, 2x was far superior to 1x.