...
- changed power limitation from -15% to 0%
...
My motivation for installing an older driver is not that I think it likely suddenly to return me to Einstein GRP success, but because I don't know where the overclock/power limitation section is on this new one.
...
From my recent Vega tests I can recommend two things.
For any changes in power settings or else, don't touch the driver or AMD tool. Instead, download OverdriveNTool, set power target to +50 and then undervolt for power saving there.
Also, I'd like to affirm what Gary said about the PSU. High end cards really need independent stable 12V supplies (through cables plus mainboard). So it may well be worth to take a second look. :-)
After a couple of weeks of working OK with recent AMD drivers (19.12.2 I think), my RX 570 under Win7 64-bits started behaving so badly (locking up, freezing PC, etc.) that I thought it was defective. So I pulled it out and was about to throw it out.
I will try it again, this time with the 18.3.4 drivers, which are the ones that came with the ASRock card. They may not have been quite so fast, but they worked.
I will try it again, this time with the 18.3.4 drivers, which are the ones that came with the ASRock card. They may not have been quite so fast, but they worked.
I was able to get it up and running again with the 19.9.2 drivers, which are WHQL for whatever that does.
It is running very nicely on the GW; the first one (at 1087.50 Hz) finished in 28 minutes (supported by two cores of an i7-4771 under Win7 64-bit).
It runs at 66% GPU load, and a cool 55 C at 60 watts (GPU-Z). I think that is roughly twice as efficient as CPU work units, though I have no way to compare them directly.
After owning Radeon VII for a year, I think I figure out how to run the card at 3x WUs, corroborated by having my host running for a week nonstop.
The trouble of running 3x is the voltage requirement is much higher. Normally for Nvidia cards if you lower the max stable core clock for benchmark by 50 MHz you can run E@H safely. For Radeon VII to run 3x you have to overvolt. Otherwise you will have the card run into zombie state described in previous discussions.
For example one of my card defaults at 1040 mV and 1800 MHz. It can run benchmark overclocked to 1100 mV and 2000 MHz. To run FGRPB1G at 2x the core voltage needs to be raised to 1130 mV while the max core clock is 2000 MHz with an average core clock at 1750 MHz. To run 3x though, I have to lower the max core clock to 1800 MHz with the voltage remains at 1130 mV. The average core clock is also 1750 MHz.
Concurrency
Max Core/MHz
Average Core/MHz
Voltage/mV
Memory/MHz
Elapsed/s
2
2000
1750
1130
1200
320
3
1800
1750
1130
1200
465
In conclusion what determines the required voltage is the average core clock and the required voltage to run E@H is much higher than that running benchmark. If you want to run 3x at default core clock of 1800 MHz try core voltage of about 1150 mV. Also use driver 19.10.1 or prior. AMD fixed OpenCL for 5700 XT in driver 19.12 or 20.1 but they ruined Radeon VII in them.
Can someone please write a short review about VII, after owning and using them for a while? Is it worth buying now? Or better wait for a new model? It seems in Europe it's pretty hard to find these cards, except for amazon + ebay (which i avoid).
The trouble of running 3x is the voltage requirement is much higher. Normally for Nvidia cards if you lower the max stable core clock for benchmark by 50 MHz you can run E@H safely. For Radeon VII to run 3x you have to overvolt. Otherwise you will have the card run into zombie state described in previous discussions.
For example one of my card defaults at 1040 mV and 1800 MHz. It can run benchmark overclocked to 1100 mV and 2000 MHz. To run FGRPB1G at 2x the core voltage needs to be raised to 1130 mV while the max core clock is 2000 MHz with an average core clock at 1750 MHz. To run 3x though, I have to lower the max core clock to 1800 MHz with the voltage remains at 1130 mV. The average core clock is also 1750 MHz.
Hi! A few questions: How do you actually overvolt/overclock with precision? Do you use the atrociously designed "Global WattMan" garbage tool? I find it horrible. And what a terrible name. How do you adjust the overclock curve to get it to stay at average of 1750 MHz?
shuhui1990 wrote:
In conclusion what determines the required voltage is the average core clock and the required voltage to run E@H is much higher than that running benchmark. If you want to run 3x at default core clock of 1800 MHz try core voltage of about 1150 mV. Also use driver 19.10.1 or prior. AMD fixed OpenCL for 5700 XT in driver 19.12 or 20.1 but they ruined Radeon VII in them.
How do you actually roll back drivers? Also, have you tried running linux? Wouldn't that speed things up a bit?
The trouble of running 3x is the voltage requirement is much higher. Normally for Nvidia cards if you lower the max stable core clock for benchmark by 50 MHz you can run E@H safely. For Radeon VII to run 3x you have to overvolt. Otherwise you will have the card run into zombie state described in previous discussions.
For example one of my card defaults at 1040 mV and 1800 MHz. It can run benchmark overclocked to 1100 mV and 2000 MHz. To run FGRPB1G at 2x the core voltage needs to be raised to 1130 mV while the max core clock is 2000 MHz with an average core clock at 1750 MHz. To run 3x though, I have to lower the max core clock to 1800 MHz with the voltage remains at 1130 mV. The average core clock is also 1750 MHz.
Hi! A few questions: How do you actually overvolt/overclock with precision? Do you use the atrociously designed "Global WattMan" garbage tool? I find it horrible. And what a terrible name. How do you adjust the overclock curve to get it to stay at average of 1750 MHz?
shuhui1990 wrote:
In conclusion what determines the required voltage is the average core clock and the required voltage to run E@H is much higher than that running benchmark. If you want to run 3x at default core clock of 1800 MHz try core voltage of about 1150 mV. Also use driver 19.10.1 or prior. AMD fixed OpenCL for 5700 XT in driver 19.12 or 20.1 but they ruined Radeon VII in them.
How do you actually roll back drivers? Also, have you tried running linux? Wouldn't that speed things up a bit?
/Aron
WattMan is enough. Pull the midpoint of the curve to the highest. Just DDU and reinstall drivers.
I will come back with a couple Linux results on the exact same hardware.
Tom Miller
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
archae86 schrieb: ... So far
)
From my recent Vega tests I can recommend two things.
For any changes in power settings or else, don't touch the driver or AMD tool. Instead, download OverdriveNTool, set power target to +50 and then undervolt for power saving there.
Also, I'd like to affirm what Gary said about the PSU. High end cards really need independent stable 12V supplies (through cables plus mainboard). So it may well be worth to take a second look. :-)
Stefan Ledwina
)
Might be interesting to find out whether the older driver itself was helpful or did it just come along with different default settings?
After a couple of weeks of
)
After a couple of weeks of working OK with recent AMD drivers (19.12.2 I think), my RX 570 under Win7 64-bits started behaving so badly (locking up, freezing PC, etc.) that I thought it was defective. So I pulled it out and was about to throw it out.
I will try it again, this time with the 18.3.4 drivers, which are the ones that came with the ASRock card. They may not have been quite so fast, but they worked.
Jim1348 wrote:I will try it
)
I was able to get it up and running again with the 19.9.2 drivers, which are WHQL for whatever that does.
It is running very nicely on the GW; the first one (at 1087.50 Hz) finished in 28 minutes (supported by two cores of an i7-4771 under Win7 64-bit).
It runs at 66% GPU load, and a cool 55 C at 60 watts (GPU-Z). I think that is roughly twice as efficient as CPU work units, though I have no way to compare them directly.
So I have a card again.
After owning Radeon VII for a
)
After owning Radeon VII for a year, I think I figure out how to run the card at 3x WUs, corroborated by having my host running for a week nonstop.
The trouble of running 3x is the voltage requirement is much higher. Normally for Nvidia cards if you lower the max stable core clock for benchmark by 50 MHz you can run E@H safely. For Radeon VII to run 3x you have to overvolt. Otherwise you will have the card run into zombie state described in previous discussions.
For example one of my card defaults at 1040 mV and 1800 MHz. It can run benchmark overclocked to 1100 mV and 2000 MHz. To run FGRPB1G at 2x the core voltage needs to be raised to 1130 mV while the max core clock is 2000 MHz with an average core clock at 1750 MHz. To run 3x though, I have to lower the max core clock to 1800 MHz with the voltage remains at 1130 mV. The average core clock is also 1750 MHz.
In conclusion what determines the required voltage is the average core clock and the required voltage to run E@H is much higher than that running benchmark. If you want to run 3x at default core clock of 1800 MHz try core voltage of about 1150 mV. Also use driver 19.10.1 or prior. AMD fixed OpenCL for 5700 XT in driver 19.12 or 20.1 but they ruined Radeon VII in them.
Hi,Can someone please write
)
Hi,
Can someone please write a short review about VII, after owning and using them for a while? Is it worth buying now? Or better wait for a new model? It seems in Europe it's pretty hard to find these cards, except for amazon + ebay (which i avoid).
Thanks and happy crunching!
shuhui1990 wrote: The
)
Hi! A few questions: How do you actually overvolt/overclock with precision? Do you use the atrociously designed "Global WattMan" garbage tool? I find it horrible. And what a terrible name. How do you adjust the overclock curve to get it to stay at average of 1750 MHz?
How do you actually roll back drivers? Also, have you tried running linux? Wouldn't that speed things up a bit?
/Aron
Aron wrote:shuhui1990
)
WattMan is enough. Pull the midpoint of the curve to the highest. Just DDU and reinstall drivers.
Some notes. Windows 10 - >
)
Some notes.
Windows 10 - > Radeon VII
Gamma Ray 1 task ~ 3.6 minutes
Gamma Ray 7 tasks ~ 21 minutes
3.6 X 7 = ~ 25 minutes
25- 21 = 4 minutes advantage.
It seems to be 15% improvement at 7 tasks.
For 2 Gamma Ray tasks: ~ 6.26 minutes
3.6 X 2 = 7.2 minutes
It seems to be a 13% improvement.
A couple of WAGS
7 tasks = ~ 1,660,000 RAC
2 tasks = ~ 1,590,000 RAC
1 task = ~ 1,386,000 RAC
I will come back with a couple Linux results on the exact same hardware.
Tom Miller
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
Tom M wrote:Some
)
For Linux/Ubuntu -> Radeon VII
Gamma Ray 1 task ~ 3.58 minutes.
So it is a wash as far as Windows/Drivers vs. Linux/Drivers are concerned.
Linux/Ubuntu GW gpu 1 task ~ 16 minutes - 17 minutes
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!