Well. There may be a bug in new ver. 1.05 and 1.06 that cause HUGE slowdown on older GPU (or may be older drivers). I see same behavior as described in previous post with ver 1.06.
I tried 3 tasks parallel and it run seemingly fast for about two hours, but then progress restarted from the beginning. Now progress is extremely slow and the estimate for remaining time is 21 days. MSI Afterburner says GPU usage is still over 40 % and temps tell the card is clearly doing something.
GPU driver on this machine is a force-installed state of the art piece... by Windows... Adrenalin 19.20.17.01 DCH. This same GPU driver seems to work fine on another machine with a little bit newer card, RX 580: https://einsteinathome.org/host/12329446
I think this problem is not about GPU drivers being too old. Looks like these older generation cards get confused of the current app.
I have three S9000 (HD7950 equivalent) that all show 34-45 days to complete. Normally they take less than one minute for a single Milkyway task. Temperatures are in the low 40s which is essentially idle for this GPU. Will have to abort.
OTOH I have six RX560 on risers and same 1.06 tasks take about 2 hours with an occasional at 6 or 7 hours
This is unfortunate as Einstein is my fallback project for the S9000.. Not sure if I can exclude those tasks. what make this really strange is that Milkyway also had a run of bad work units recently.
[EDIT] This is terrible!! Milkyway also has an ongoing problem. After aborting my 3 Einstein tasks, Milkyway started up and within seconds 200+ errored out. However, problem is recognized and is being worked on supposedly.
Some data with 1.06 GW-opencl-ati: Host i5-3570k, Radeon VII, Win7:
1 WU ~5.790 s
3 WUs ~7.680 s
4 WUs ~9,280 s
Even with 4 WUs at the same time, the Radeon VII is not fully utilized, max. 80 % at ~1200 MHz.
I only have 4 cores, so I can't test more. It's time for a new computer.
V 1.04 GW-opencl-nvidia: Host Xeon Xeon(R) CPU E3-1225 v3, nvidia 1050Ti, Linux:
2 WUs ~7800 s
With 3 WUS the runtime is much longer, 2 WUs is optimal for this GPU.
Is there any other work available than these O2 All-Sky Search WU's right now? My RAC is going down very fast every time, when my computer crunches these. 45 000 seconds and 1000 points per WU?
On macOS with an AMD GPU they don't run at all by the way.
Until today noon they all did abort after 3 seconds.
Now I see the computer got 150 new Gravitational GPU Tasks and they run and run and run without any progress, in the end it said one Task needs 350 days while the GPU was at 50% load.
Aborted them all and disabled Gravity Wave tasks for now. Looking forward for the fix, I love gravity :)
This is unfortunate as Einstein is my fallback project for the S9000.. Not sure if I can exclude those tasks. what make this really strange is that Milkyway also had a run of bad work units recently.
Sure you can.
If you need to disable particular app for one or few (but not all) of your computers there are few option for it:
1 - server side: you can disable this app in project preferences (Continuous Gravitational Wave search O2 All-Sky) in one of "venues" (work/home/school) and assign computer(s) to it
2 - client side: you can use cc_config.xml option to disable any particular app on some GPUs.
Example of config. Device_num is a number of GPU you want to exclude from working on the app. if omitted - all GPUs of this type (ATI/AMD in example) :
Mad_Max wrote:Well. There may
)
I see this too with this machine (Radeon R9 270X 4GB): https://einsteinathome.org/host/12768123
I tried 3 tasks parallel and it run seemingly fast for about two hours, but then progress restarted from the beginning. Now progress is extremely slow and the estimate for remaining time is 21 days. MSI Afterburner says GPU usage is still over 40 % and temps tell the card is clearly doing something.
GPU driver on this machine is a force-installed state of the art piece... by Windows... Adrenalin 19.20.17.01 DCH. This same GPU driver seems to work fine on another machine with a little bit newer card, RX 580: https://einsteinathome.org/host/12329446
I think this problem is not about GPU drivers being too old. Looks like these older generation cards get confused of the current app.
I have three S9000 (HD7950
)
I have three S9000 (HD7950 equivalent) that all show 34-45 days to complete. Normally they take less than one minute for a single Milkyway task. Temperatures are in the low 40s which is essentially idle for this GPU. Will have to abort.
OTOH I have six RX560 on risers and same 1.06 tasks take about 2 hours with an occasional at 6 or 7 hours
This is unfortunate as Einstein is my fallback project for the S9000.. Not sure if I can exclude those tasks. what make this really strange is that Milkyway also had a run of bad work units recently.
[EDIT] This is terrible!! Milkyway also has an ongoing problem. After aborting my 3 Einstein tasks, Milkyway started up and within seconds 200+ errored out. However, problem is recognized and is being worked on supposedly.
Some data with 1.06
)
Some data with 1.06 GW-opencl-ati:
Host i5-3570k, Radeon VII, Win7:
1 WU ~5.790 s
3 WUs ~7.680 s
4 WUs ~9,280 s
Even with 4 WUs at the same time, the Radeon VII is not fully utilized, max. 80 % at ~1200 MHz.
I only have 4 cores, so I can't test more. It's time for a new computer.
V 1.04 GW-opencl-nvidia:
Host Xeon Xeon(R) CPU E3-1225 v3, nvidia 1050Ti, Linux:
2 WUs ~7800 s
With 3 WUS the runtime is much longer, 2 WUs is optimal for this GPU.
Host i3-6100, nvidia GT 1030, Win 10:
1 WU ~8100 s
Only one WU at a time makes sense.
I find some of my WUs from today not in my tasklist (https://einsteinathome.org/de/host/12779294/tasks/0/49), but only in my log file, for example:
h1_0501.65_O2C02Cl1In0__O2AS20-500_501.80Hz_343_1
h1_0501.65_O2C02Cl1In0__O2AS20-500_501.80Hz_348_1
h1_0501.65_O2C02Cl1In0__O2AS20-500_501.80Hz_354_1
h1_0501.65_O2C02Cl1In0__O2AS20-500_501.80Hz_344_1
But maybe it's just too hot today.
Is there any other work
)
Is there any other work available than these O2 All-Sky Search WU's right now? My RAC is going down very fast every time, when my computer crunches these. 45 000 seconds and 1000 points per WU?
I'm disabling the GPU
)
I'm disabling the GPU versions for now for further investigation, the validation rates look worrying.
BM
On macOS with an AMD GPU they
)
On macOS with an AMD GPU they don't run at all by the way.
Until today noon they all did abort after 3 seconds.
Now I see the computer got 150 new Gravitational GPU Tasks and they run and run and run without any progress, in the end it said one Task needs 350 days while the GPU was at 50% load.
Aborted them all and disabled Gravity Wave tasks for now. Looking forward for the fix, I love gravity :)
See here for the tasks of my host:
https://einsteinathome.org/host/3426951/tasks/0/0
B.I.G wrote:... the computer
)
I trust you reprimanded it appropriately ... :-).
Cheers,
Gary.
This is unfortunate as
)
Sure you can.
If you need to disable particular app for one or few (but not all) of your computers there are few option for it:
1 - server side: you can disable this app in project preferences (Continuous Gravitational Wave search O2 All-Sky) in one of "venues" (work/home/school) and assign computer(s) to it
2 - client side: you can use cc_config.xml option to disable any particular app on some GPUs.
Example of config. Device_num is a number of GPU you want to exclude from working on the app. if omitted - all GPUs of this type (ATI/AMD in example) :
On a Fury-X in Linux I am
)
On a Fury-X in Linux I am seeing >50% invalids.
https://einsteinathome.org/host/12768835/tasks/5/49
System:
Arch Linux | Kernel 5.2.3
Intel i7-4790k
R9 Fury-X | 1125MHz core 500 MHz mem | Latest kernel driver with OpenCL from AMD driver 19.20.812932
Average run-time v1.06 (73 tasks): 3970 seconds
Valid: 13
Invalid: 20
Pending: 40
I see there is a new CPU
)
I see there is a new CPU application named 1.01 (GWold).
Does anyone know what GWold means?
Thanks!