Continuing observations on my Radeon VII running 0.11.
It hit 99% reported progress after about 80 minutes elapsed time, at which point GPU usage dropped to zero.
GPU-Z reports the GPU clock at 25 MHz(!!!), memory clock at 349, temperature at 30/31, power at 20W, but memory usage still at 1541/189.
That lasted for a bit over 5 minutes (toplist statistics recalculation), after which the task completed and uploaded, with a reported elapsed time of 1:24:51.
If you care to take a look at the task it is 839004421. As the second task to fulfill the quorum is unsent, it may be some time before we can see validation or not.
Naturally running this task smashed my DCF up by over a factor of ten. As I've unsuspended the GRP work in my queue, that work is getting burned off in panic mode. It may be some time before I can try 2X or 3X on the GW work.
With my Radeon VII (Win 7, AMD 19.4.1, einstein_O1OD1E_0.11_windows_x86_64__GW-opencl-ati-V1.exe) I can confirm the observation of archae86 - very low GPU load and clock frequency (~ 400 MHz) with a single WU.
When starting a second WU my PC crashed after a few minutes. So back to Gamma 1.18.
... engineering 0.11 GPU WUs running between 1 and 6 hours on fast NVIDIA !! GPU load varying between 1 % and max 10% !! Much effort for 144,000 GFLOPS.
Some WUs getting ERROR at address x'......0005'
Running WIN7 and WIN10.
I'm stopping 0.11 for now and will wait for "improvements".
Now the GPU is used between 50 to 55%. But the clock is not at max , I believe. It can run at like 1800MHz, but its between 1600 and 1700 MHz most of the time.
For the next tasks I will try 3 tasks at once.
Edit: 2 tasks finished in about 7700s, compared to the 5500s for a single one.
Looks like these are similar to the first revisions of the current GPU tasks. Low GPU utilization due to being heavily CPU dependent. Hopefully it'll improve.
Running the GW 0.11 Windows AMD application at 3X changed things materially, although the Radeon VII still is very lightly used.
1X 3X Variable
64% 73% GPU Load
39W 53W GPU only Power Draw
463 880 GPU Clock MHz
798 839 Memory Clock MHZ
37C 44C GPU temperature
39C 49C Hot Spot GPU temperature
While the memory usage reported barely budged--which seems odd.
As you might suppose, the machine is considerably more productive at 3X than 1X on this work. I've tampered with some things mid-stream, and don't have much results, but on this machine a 1X run this morning took elapsed time of 5091 seconds, while a set of three which just finished (yes, I offset them some, but not enough) took only about 6200 seconds, so a huge productivity boost. One of those validated on completion, which is comforting. This host has only got four cores, and is not hyperthreaded, so although 4X might well work, and might be slightly more productive, I'm not tempted to try it.
My brand new (today) RX 570 host has six physical cores. If I get some days of stable running out of it one simpler stuff, I might give 4X on this work a try on it. That machine has so far run two of these tasks at 1X, with elapsed time around 3780 seconds. The much faster 1X time may mean the 570 is better for this work than a Radeon VII, but more likely the 9th generation 6-core CPU burns through the CPU portion of the job much faster than the older CPU on my Radeon VII host.
Meanwhile, my primary host is indicated as having 29 days of work on board, as I unintentionally allowed more of the new GW work to download when a spate of running GRP had driven the completion estimates way back down. I'm afraid a mass abort is in my future but I currently plan to run pure GW GPU for another half day.
Meanwhile, my primary host is indicated as having 29 days of work on board, as I unintentionally allowed more of the new GW work to download when a spate of running GRP had driven the completion estimates way back down. I'm afraid a mass abort is in my future but I currently plan to run pure GW GPU for another half day.
Yeah, The extremly different runtimes of 15 min vs 2h is messing with the work I got. I almost ran out of tasks. I changed it from 0.25 to 0.5d of work buffer. But if I run only the 15 min tasks it might download way to many of the 2h ones. But so far I havent gotten any new GW units.
Looks like these are similar to the first revisions of the current GPU tasks. Low GPU utilization due to being heavily CPU dependent. Hopefully it'll improve.
Yes, that is the way it was, and it will get better.
But if they are having that much of a problem with OpenCl, I wonder what the chances are for CUDA?
The name is listed in the client_state.xml file under the project section in the <app> <name> declaration. The name is what you input into your app_config file.
Continuing observations on my
)
Continuing observations on my Radeon VII running 0.11.
It hit 99% reported progress after about 80 minutes elapsed time, at which point GPU usage dropped to zero.
GPU-Z reports the GPU clock at 25 MHz(!!!), memory clock at 349, temperature at 30/31, power at 20W, but memory usage still at 1541/189.
That lasted for a bit over 5 minutes (toplist statistics recalculation), after which the task completed and uploaded, with a reported elapsed time of 1:24:51.
If you care to take a look at the task it is 839004421. As the second task to fulfill the quorum is unsent, it may be some time before we can see validation or not.
Naturally running this task smashed my DCF up by over a factor of ten. As I've unsuspended the GRP work in my queue, that work is getting burned off in panic mode. It may be some time before I can try 2X or 3X on the GW work.
With my Radeon VII (Win 7,
)
With my Radeon VII (Win 7, AMD 19.4.1, einstein_O1OD1E_0.11_windows_x86_64__GW-opencl-ati-V1.exe) I can confirm the observation of archae86 - very low GPU load and clock frequency (~ 400 MHz) with a single WU.
When starting a second WU my PC crashed after a few minutes. So back to Gamma 1.18.
On the NV 1050Ti in my second host https://einsteinathome.org/de/host/12247194/tasks/2/0
3 WUs run in parallel, time about 7800 s. No problems at the moment.
... engineering 0.11 GPU WUs
)
... engineering 0.11 GPU WUs running between 1 and 6 hours on fast NVIDIA !! GPU load varying between 1 % and max 10% !! Much effort for 144,000 GFLOPS.
Some WUs getting ERROR at address x'......0005'
Running WIN7 and WIN10.
I'm stopping 0.11 for now and will wait for "improvements".
crashtech schrieb:I'm not
)
I just started to set it to use more than one task. I set it to 2 now but changing the value for the GW app to 0.5 in the settings on the website. https://einsteinathome.org/de/account/prefs/project
Now the GPU is used between 50 to 55%. But the clock is not at max , I believe. It can run at like 1800MHz, but its between 1600 and 1700 MHz most of the time.
For the next tasks I will try 3 tasks at once.
Edit: 2 tasks finished in about 7700s, compared to the 5500s for a single one.
Looks like these are similar
)
Looks like these are similar to the first revisions of the current GPU tasks. Low GPU utilization due to being heavily CPU dependent. Hopefully it'll improve.
Running the GW 0.11 Windows
)
Running the GW 0.11 Windows AMD application at 3X changed things materially, although the Radeon VII still is very lightly used.
While the memory usage reported barely budged--which seems odd.
As you might suppose, the machine is considerably more productive at 3X than 1X on this work. I've tampered with some things mid-stream, and don't have much results, but on this machine a 1X run this morning took elapsed time of 5091 seconds, while a set of three which just finished (yes, I offset them some, but not enough) took only about 6200 seconds, so a huge productivity boost. One of those validated on completion, which is comforting. This host has only got four cores, and is not hyperthreaded, so although 4X might well work, and might be slightly more productive, I'm not tempted to try it.
My brand new (today) RX 570 host has six physical cores. If I get some days of stable running out of it one simpler stuff, I might give 4X on this work a try on it. That machine has so far run two of these tasks at 1X, with elapsed time around 3780 seconds. The much faster 1X time may mean the 570 is better for this work than a Radeon VII, but more likely the 9th generation 6-core CPU burns through the CPU portion of the job much faster than the older CPU on my Radeon VII host.
Meanwhile, my primary host is indicated as having 29 days of work on board, as I unintentionally allowed more of the new GW work to download when a spate of running GRP had driven the completion estimates way back down. I'm afraid a mass abort is in my future but I currently plan to run pure GW GPU for another half day.
archae86 schrieb:Meanwhile,
)
Yeah, The extremly different runtimes of 15 min vs 2h is messing with the work I got. I almost ran out of tasks. I changed it from 0.25 to 0.5d of work buffer. But if I run only the 15 min tasks it might download way to many of the 2h ones. But so far I havent gotten any new GW units.
mmonnin wrote:Looks like
)
Yes, that is the way it was, and it will get better.
But if they are having that much of a problem with OpenCl, I wonder what the chances are for CUDA?
I am curious to try running
)
I am curious to try running this app starting with 4 at a time, with each instance having its own physical core, like so:
Does anyone think this might work,
and if so, does anyone know the right project name to place in the app_config?Edit: App name added, thanks to Keith Myers for the valuable information.
The project name is listed in
)
The name is listed in the client_state.xml file under the project section in the <app> <name> declaration. The name is what you input into your app_config file.