New WUs appeared (%0044L%), which load the GPU much better (better constant load without idle/halfload during WU processing) - reduction in run time ~200sec, increase in average power consumption ~30W, same min/max power consumption. So average is up due to GPU being used fully.
I had a system crash, all settings either lost or doing crazy things and since them I'm again unable to undervolt the GPU.. Tried to reset, reboot, reapply, etc., no joy. Oh, this is a pain...
They're not "good" tasks, they're "short ends". Some of these tasks at low frequencies have a lower work content and so crunch faster. I mentioned them previously. Not all tasks at low frequencies are "short". It's quite variable - luck of the draw.
Quote:
There is also possible variation based on the frequency term. For example, a task named LATeah0042L_44.0_.... might take a different time than one named LATeah0042L_1012.0_.... Finally, at very low frequencies - 4.0, 12.0, 20.0 ... - some of the tasks run considerably faster (like 50-100% faster) than others at the same frequency. These are known as 'short ends' and there is less data to crunch.
Finally got around the problem of not being able to hold core voltages set. WattMan is a huge crap, at least in the current state. It absolutely doesn't hold settings, sometimes after reboot the GPU clock and PStates are totally off.
Used MSI Afterburner 4.4.0 which has Vega support and all is well. Currently running with Core Voltage -150 mV (telemetry reports 0.994 - 1.050 V). Average chip-reported power ~175 W.
Custom fan curve is also much better in AB than the AMD-defined fan levels.
Further reduced VDDC to -168 mV (measured results 0.975 - 1.031 V). Average chip-reported power ~166 W (min 105, max 202 W).
Do you have measurement of power from the wall?
Thanks!
No, sorry. I have multiple machines hooked on the wattmeter, so it wouldn't be accurate. Even with a single machine measuring at wall is not easy to estimate pure GPU power as you need to take several other factors into account too.
What I can do would be to measure the differential between GPU idle / load at wall.
New WUs appeared (%0044L%),
)
New WUs appeared (%0044L%), which load the GPU much better (better constant load without idle/halfload during WU processing) - reduction in run time ~200sec, increase in average power consumption ~30W, same min/max power consumption. So average is up due to GPU being used fully.
I had a system crash, all
)
I had a system crash, all settings either lost or doing crazy things and since them I'm again unable to undervolt the GPU.. Tried to reset, reboot, reapply, etc., no joy. Oh, this is a pain...
-----
That sucks... Current
)
That sucks...
Current status.
Seems the "good" WUs were
)
Seems the "good" WUs were from LATeah0044L_00* to LATeah0044L_99* series.
Back to normal crunch time/utilization/power consumption.
They're not "good" tasks,
)
They're not "good" tasks, they're "short ends". Some of these tasks at low frequencies have a lower work content and so crunch faster. I mentioned them previously. Not all tasks at low frequencies are "short". It's quite variable - luck of the draw.
Cheers,
Gary.
Cool. Sorry, forgot to "save"
)
Cool. Sorry, forgot to "save" this bit of info into my long term memory. Fixed now ;)
Finally got around the
)
Finally got around the problem of not being able to hold core voltages set. WattMan is a huge crap, at least in the current state. It absolutely doesn't hold settings, sometimes after reboot the GPU clock and PStates are totally off.
Used MSI Afterburner 4.4.0 which has Vega support and all is well. Currently running with Core Voltage -150 mV (telemetry reports 0.994 - 1.050 V). Average chip-reported power ~175 W.
Custom fan curve is also much better in AB than the AMD-defined fan levels.
-----
Further reduced VDDC to -168
)
Further reduced VDDC to -168 mV (measured results 0.975 - 1.031 V). Average chip-reported power ~166 W (min 105, max 202 W).
-----
Mumak escribió:Further
)
Do you have measurement of power from the wall?
Thanks!
Trotador wrote:Mumak
)
No, sorry. I have multiple machines hooked on the wattmeter, so it wouldn't be accurate. Even with a single machine measuring at wall is not easy to estimate pure GPU power as you need to take several other factors into account too.
What I can do would be to measure the differential between GPU idle / load at wall.
-----