I've done some more tests and it seems temperature was the critical factor (as mentioned so many times elsewhere). So the improved run times were due to reduced temperature - raised fan speeds and lower power limit. Apparently HBM temperature > 80 C has also a negative impact on performance.
Currently I'm happy with -20% power limit, run times are only 10-20s longer that with 0%, but reduced temperature, power and fan speeds are of more benefit.
My experimentation agrees with Mumak's results. Dropping the power limit to -20% has increased runtimes by only ~15 seconds. I also left my fan and temperature settings at my previously reported values...
min, max and average GPU temps have been 54C since tweaking the power limit. GPU Only Power draw values are now 135W min, 180W max and 173.6W average which represents quite a healthy saving on energy cost over my previous +50 power limit results. I will start 'wall power' measurements tomorrow...
With 1WU per GPU im seeing an average of 56,9% load on GPU for this 60 minute timeframe. Average power consumption over the same one period is 111,3W. The numbers are from GPUz log, 1sec interval - so not as accurate as with clamp on PCIe cables. But consumption from the wall confirms the power usage is somewhere around there.
With 2WU per GPU the load will go up and consumption with that, quite sure of this. :)
Tomshardware did a writeup on Vega undervolting, i used it as a baseline for tweaking. http://www.tomshardware.com/reviews/amd-radeon-rx-vega-64-undervolting,5234.html
And regarding HBM2, i think it's overvolted in default spec, since in BIOS bootup it states HBM2 945Mhz 950mv but WattMan shows it as 1050mv. Undervolting it gave ~10W reduction in power consumption, no changes in stability.
This morning I have re
)
This morning I have re measured the GPU Only Power draw, I'm seeing 158W min, 276W max and 244.5W average.
I will drop the power limit now and see what happens!
I've done some more tests and
)
I've done some more tests and it seems temperature was the critical factor (as mentioned so many times elsewhere). So the improved run times were due to reduced temperature - raised fan speeds and lower power limit. Apparently HBM temperature > 80 C has also a negative impact on performance.
Currently I'm happy with -20% power limit, run times are only 10-20s longer that with 0%, but reduced temperature, power and fan speeds are of more benefit.
-----
My
)
My experimentation agrees with Mumak's results. Dropping the power limit to -20% has increased runtimes by only ~15 seconds. I also left my fan and temperature settings at my previously reported values...
min, max and average GPU temps have been 54C since tweaking the power limit. GPU Only Power draw values are now 135W min, 180W max and 173.6W average which represents quite a healthy saving on energy cost over my previous +50 power limit results. I will start 'wall power' measurements tomorrow...
Gav.
Average wall power
)
Average wall power consumption measured over eight and a quarter hours today = 295Watts :-)
Some feedback on my Vega
)
Some feedback on my Vega 56.
Stock clock. Running 1 WU.
Average time - 450sec
Average temp - 70c
Looks like the WU's have changed recently? Now showing each WU will take 31min. Or Einstein playing up?
My configuration
)
vega64 https://www.techpowerup.com/gpuz/64g
Single WU per gpu run, GPU and HBM2 both undervolted ( mem@default 1050mv to 950mv, gpu@default 1200mv to 1025mv),
With these settings it's 100% stable. From the wall it's using ~160W (peak) over GPU idle, average 135W.
WU times 370-406s.
60 minute timeframe stats,
)
60 minute timeframe stats, same settings. Computer crunching continuously for 12h before below timeframe.
How can the GPU (TDP 220W)
)
How can the GPU (TDP 220W) with a Power Limit +50% peak only at 160W ?
I have tried similar undervolting, but even with -20% power limit mine peaks at 176W.
Running 2 WUs.
-----
With 1WU per GPU im seeing an
)
With 1WU per GPU im seeing an average of 56,9% load on GPU for this 60 minute timeframe. Average power consumption over the same one period is 111,3W. The numbers are from GPUz log, 1sec interval - so not as accurate as with clamp on PCIe cables. But consumption from the wall confirms the power usage is somewhere around there.
With 2WU per GPU the load will go up and consumption with that, quite sure of this. :)
Tomshardware did a writeup on Vega undervolting, i used it as a baseline for tweaking. http://www.tomshardware.com/reviews/amd-radeon-rx-vega-64-undervolting,5234.html
And regarding HBM2, i think it's overvolted in default spec, since in BIOS bootup it states HBM2 945Mhz 950mv but WattMan shows it as 1050mv. Undervolting it gave ~10W reduction in power consumption, no changes in stability.
Would be very interested to
)
Would be very interested to see your results running 2 work units at a time...