I got 6 units of the latest batch and all errored out after 4 seconds because of an access code violation. Same fate for my wingmen. Windows 10, RTX 4080, 32 GB RAM. Anybody who managed to finish off a bunch of them care to share your system details?

i completed all 20 tasks from last batch
win 10 64; RTX 3060
i had folding@home in parallel

---

--
I crunch for Ukraine

Hello guys
Are you foolish or what?

Not as foolish as you for running two at time.
Per Task runtime doesn't matter- the number per hour is what matters.

On my RTX 2060 & RTX 2060 Super the GPU load is around 75% to 80% for these Tasks.
Given that the GPU load is over 50%, and it requires about 1.3 CPU cores to support one Task, even if you give them 3 CPU cores to support 2 GPU Tasks when running two at a time, and the fact that the application has not been optimised much (if at all) as shown by the frequent dips in the GPU load & less frequent but very regular large dips in GPU load, it doesn't make much sense to run 2 at a time.

Run one at a time & see how it goes.


As for the Credit per Task- it's Alpha work, things will change as they progress. At least for the last batch the Credit per Task was much more consistent than the earlier batches & applications.
And as for their Runtime- while they take longer than the last batch, they are still taking less time than the first batch i was able to process on the GPU.

On my RTX4090 1 Task of the Wus from 3.7.2024 longs about 440 sec 2 Tasks longs about 650 sec. At 1 Task the GPU Utilisation was abot 50-60% with 2 tasks it was about 98%
1 Stk 440-450 sec = 100%
2 Stk 610-640 sec = 142,8%
3 Stk 904-910 sec = 149%
4 Stk 1199-1202 sec = 149,8%

But now 2 tasks are 1,5 longer

I got 6 units of the latest batch and all errored out after 4 seconds because of an access code violation. Same fate for my wingmen. Windows 10, RTX 4080, 32 GB RAM. Anybody who managed to finish off a bunch of them care to share your system details?

I had the same thing happening with my systems initially.
i7-8700K, 32GB RAM, Win10 Pro, 552.22 video driver, RTX 2060 & RTX 2060 Super on other system with same other specs.

I reset the project, it re-downloaded all the files, and then they worked OK (no idea why downloading the same things all over again made any difference, but it did).

---

Grant
Darwin NT

But now 2 tasks are 1,5 longer

As i mentnioned in my other post- while these Tasks take longer to process than the last batch, they are still taking less time than the initial batch that i was able to process.
For my RTX 2060

First batch   27min 50 sec
Last batch    16min 50 sec
Current batch 25min 45 sec

How many CPU cores/threads of your system are in use?
Either try 1 Task at a time again, or limit the number of running CPU Tasks to free up a CPU core/thread, or reserve 3 cores/threads for your 2 GPU Ralph Tasks.
The default for Ralph is to reserve only 0.997 cores/threads per GPU Task, but the actual usage is more like 1.3, with frequent bumps of 2.5 every 20 seconds or so (corresponding with the drop in GPU load & power draw).
See if reserving 3 cores/threads helps significantly enough to make a difference.

I tried reserving 2 cores/threads, but the improvement in Ralph processing time was so small as to not come close to offsetting the lost output of CPU work from that core/thread. For your high end GPU and running 2 Tasks at a time, it might be.

---

Grant
Darwin NT

Momently i have NO CPU tasks running

Ok, so now my three Windows hosts run these wus successfully. It was necessary to detach from Ralph on all of them and to rejoin the project after a reboot to make sure like it was mentioned on a previous post. Evidently the data from the project folder got corrupted. Now they all work flawlessly.

Another thing: My RTX 3070ti was initially rejected because of not enough VRAM which is nonsense because it’s got 8 GB. The solution was to install the latest version of BOINC manager.

something strange is going with the tasks

my tasks page shows that i have 3 tasks in progress, but BOINC on the computer shows only one. Nothing in transfers

i definitely have all tasks visible, not only active...

will research logs carefully tomorrow. did anyone see such issue ?

---

--
I crunch for Ukraine

my tasks page shows that i have 3 tasks in progress, but BOINC on the computer shows only one. Nothing in transfers

Sounds like what's known as Ghost Tasks.
Some sort of network issue during contacting the Scheduler & getting the work. The Scheduler thinks you got it, but you didn't.


Looking at my Tasks i've had some weirdness as well.
I had a Task error out with the reason being "Timed out - no response", that was after only 8 minutes after getting it...

---

Grant
Darwin NT

my tasks page shows that i have 3 tasks in progress, but BOINC on the computer shows only one. Nothing in transfers

Sounds like what's known as Ghost Tasks.
Some sort of network issue during contacting the Scheduler & getting the work. The Scheduler thinks you got it, but you didn't.


Looking at my Tasks i've had some weirdness as well.
I had a Task error out with the reason being "Timed out - no response", that was after only 8 minutes after getting it...

Thank you

I think you are right - i had some networking issues yesterday due to work on some other project..

Those 2 extra tasks still hang In Progress and will time out in few days

---

--
I crunch for Ukraine

I've got few tasks and they all fail after 4000 sec with

C:ProgramDataBOINCprojectsralph.bakerlab.orgcv1rf2aautil.py:450: UserWarning: Using torch.cross without specifying the dim arg is deprecated.
Please either pass the dim explicitly or simply use torch.linalg.cross.
The default value of dim will change to agree with that of linalg.cross in a future release. (Triggered internally at C:cbpytorch_1000000000000workatensrcATennativeCross.cpp:66.)
  Z = torch.cross(Xn,Yn)

---

--
I crunch for Ukraine

I've got few tasks and they all fail after 4000 sec with

C:ProgramDataBOINCprojectsralph.bakerlab.orgcv1rf2aautil.py:450: UserWarning: Using torch.cross without specifying the dim arg is deprecated.
Please either pass the dim explicitly or simply use torch.linalg.cross.
The default value of dim will change to agree with that of linalg.cross in a future release. (Triggered internally at C:cbpytorch_1000000000000workatensrcATennativeCross.cpp:66.)
  Z = torch.cross(Xn,Yn)

They didn't necessarily fail- that error message was occurring with Taks that completed and validated on the previous runs.
It looks like they completed processing OK, but there was a problem with Validation. I'm getting the same issue.

Either they didn't process correctly, or there is an issue with the Validators.

---

Grant
Darwin NT

I'm getting this error:

655	ralph@home	02/09/2024 23:21:46	Requesting new tasks for NVIDIA GPU	
656	ralph@home	02/09/2024 23:21:47	Scheduler request completed: got 0 new tasks	
657	ralph@home	02/09/2024 23:21:47	A minimum of 5120 MB (preferably 5120 MB) of video RAM is needed to process tasks using your computer's NVIDIA GPU	
658	ralph@home	02/09/2024 23:21:47	Project requested delay of 31 seconds

I've got a Quadro P2200 with exactly that (5120MB) according to GPU-Z. Anyone else getting that error?

D

---

I've got few tasks and they all fail after 4000 sec with

C:ProgramDataBOINCprojectsralph.bakerlab.orgcv1rf2aautil.py:450: UserWarning: Using torch.cross without specifying the dim arg is deprecated.
Please either pass the dim explicitly or simply use torch.linalg.cross.
The default value of dim will change to agree with that of linalg.cross in a future release. (Triggered internally at C:cbpytorch_1000000000000workatensrcATennativeCross.cpp:66.)
  Z = torch.cross(Xn,Yn)

They didn't necessarily fail- that error message was occurring with Taks that completed and validated on the previous runs.
It looks like they completed processing OK, but there was a problem with Validation. I'm getting the same issue.

Either they didn't process correctly, or there is an issue with the Validators.

Mine seem to be finishing with the WARNING message and then fail to VALIDATE. They are taking 7GB of 24GB of dedicated GPU memory.



Stderr output

<core_client_version>8.0.2</core_client_version>
<![CDATA[
<stderr_txt>
C:ProgramDataBOINCprojectsralph.bakerlab.orgcv1rf2aautil.py:450: UserWarning: Using torch.cross without specifying the dim arg is deprecated.
Please either pass the dim explicitly or simply use torch.linalg.cross.
The default value of dim will change to agree with that of linalg.cross in a future release. (Triggered internally at C:cbpytorch_1000000000000workatensrcATennativeCross.cpp:66.)
Z = torch.cross(Xn,Yn)
16:09:10 (7564): called boinc_finish(0)

</stderr_txt>
]]>

I've got few of these tasks today and they are successfully validated!

---

--
I crunch for Ukraine