It was turned off for me by default, so I turned it on and crunched for a short while...
The average time per WU was 23% slower on a 3060ti LHR 8GB so I turned it off rather quickly after the first few WUs were returned being fully crunched under the new setting.
Interesting, thanks, didn't notice that new option.
"While the buffering of GPU commands ensure good CPU-GPU execution parallelism, the side effect of this approach is increased latency or user input lag.
(...)
"The hardware-accelerated GPU scheduling may prove useful when gaming on low and mid-tier CPUs."
I tried rendering in blender and it's faster without that option, too.
Just saw this value: my MSI RTX 2070 Ventus (6GB) did it in ~254 s on Windows 10 without OC. Can this difference be a consequence of another brand / version of that card? 17% seem a lot. That was a few months ago, though. Different tasks?
I also have a RTX2060super in the mix but there is something wrong there as it is taking 452s @1995MHz (+50) under Ubuntu when it should be somewhere between the 2060 and 2070.
the problem with your 2060 Super card is not the card itself, but the old driver version. same with your 1070Ti system.
most of the speed comes from the application. the fastest app is the 1.28 application, and you can see that these two systems are running the older app. the 1.28 application requires OpenCL 2.0+ support. Einstein will only send this application if you have compatible drivers. OpenCL 3.0 was added to Nvidia drivers in the 465 branch and later, with drivers before that only supporting OpenCL 1.2. so your 460 drivers don't make the cut.
update to 470 branch or later drivers, and you will get the 1.28 application, and you will see faster processing times as a result.
the problem with your 2060 Super card is not the card itself, but the old driver version. same with your 1070Ti system.
most of the speed comes from the application. the fastest app is the 1.28 application, and you can see that these two systems are running the older app. the 1.28 application requires OpenCL 2.0+ support. Einstein will only send this application if you have compatible drivers. OpenCL 3.0 was added to Nvidia drivers in the 465 branch and later, with drivers before that only supporting OpenCL 1.2. so your 460 drivers don't make the cut.
update to 470 branch or later drivers, and you will get the 1.28 application, and you will see faster processing times as a result.
Thanks - I'd noticed the driver discrepancy and just updated them for the 2060s before I saw your message. It's performing better now 7:04 per task on _764 tasks vs. 7:29 but still not even close to the 5:20 the 2060 is seeing on _756 and _754 tasks.
Updating the rest of the nodes now ...
edit - Thanks! It is doing much better now at 4:30 per LATeah3012L02_764 task. I guess I just had to wait for the current job cache to finish as presumable it still had jobs with the older application version.
Hello Anandbhat, I have
)
Hello Anandbhat,
I have tried the scheduling setting for a little bit on Win10Pro, wasn't good for me.
WinMenu>Settings>System>Display tab>Graphics settings>Hardware-accelerated GPU scheduling ON/OFF
It was turned off for me by default, so I turned it on and crunched for a short while...
The average time per WU was 23% slower on a 3060ti LHR 8GB so I turned it off rather quickly after the first few WUs were returned being fully crunched under the new setting.
Thanks HWpecker, that is a
)
Thanks HWpecker, that is a significant drop in numbers.
Interesting, thanks, didn't
)
Interesting, thanks, didn't notice that new option.
I tried rendering in blender and it's faster without that option, too.
gordonbb
)
Just saw this value: my MSI RTX 2070 Ventus (6GB) did it in ~254 s on Windows 10 without OC. Can this difference be a consequence of another brand / version of that card? 17% seem a lot. That was a few months ago, though. Different tasks?
Could be. Different cards
)
Could be. Different cards boost to different amounts.
Could be slight difference in tasks though the GR tasks are much less variable in their parameter sets than those for the GW tasks for comparison.
gordonbb wrote: I also have
)
the problem with your 2060 Super card is not the card itself, but the old driver version. same with your 1070Ti system.
most of the speed comes from the application. the fastest app is the 1.28 application, and you can see that these two systems are running the older app. the 1.28 application requires OpenCL 2.0+ support. Einstein will only send this application if you have compatible drivers. OpenCL 3.0 was added to Nvidia drivers in the 465 branch and later, with drivers before that only supporting OpenCL 1.2. so your 460 drivers don't make the cut.
update to 470 branch or later drivers, and you will get the 1.28 application, and you will see faster processing times as a result.
_________________________________________________________________________
Ian&Steve C. wrote:the
)
Thanks - I'd noticed the driver discrepancy and just updated them for the 2060s before I saw your message. It's performing better now 7:04 per task on _764 tasks vs. 7:29 but still not even close to the 5:20 the 2060 is seeing on _756 and _754 tasks.
Updating the rest of the nodes now ...
edit - Thanks! It is doing much better now at 4:30 per LATeah3012L02_764 task. I guess I just had to wait for the current job cache to finish as presumable it still had jobs with the older application version.
Hi,Here comes some run
)
Hi,
Here comes some run times after blowing the dust away from my GPUs.
EDIT added: The task times are from LATeah3012L02_796 series.
TITAN V: about 250W @ 86C, GPU&MEM +60MHz , 3 tasks in 5:34 -> 112 s per one.
RTX 2080 Ti: mem@15000 MHz, GPU@1990 MHz, 204-280W, 3 tasks in 6:12 - 6:20 -> 123 - 127 s per one.
I have set EAH_SLEEP set to 1 and my version is AIO without libsleeep.
Most other projects have a
)
Most other projects have a standard list of all their machine where one can find a good comparison.
e.g. Milkyway:
https://milkyway.cs.rpi.edu/milkyway/gpu_list.php (yes, it seems to have not been updated since some years)
https://milkyway.cs.rpi.edu/milkyway/top_hosts.php
Where are these lists hidden for Einstein? Do they not exist?
MagicEye schrieb: Where are
)
Couldn't find anything for Einstein, this is why I made this list.