[Diffident]

I only got a total of 2 GPU Amicable tasks. The site says there is only 2374 tasks available, but it doesn't break it down to CPU or GPU. Now I'm getting "Internet access OK - project servers may be temporarily down." messages.

 

[tictoc]

I only got a total of 2 GPU Amicable tasks. The site says there is only 2374 tasks available, but it doesn't break it down to CPU or GPU. Now I'm getting "Internet access OK - project servers may be temporarily down." messages.

The same thing happened on my main machine and my primary BOINC instance. Additionally, it assigned me a new cpid for some reason. Added the project to a different instance, and it started downloading tasks right away. :headscrat

 

[Diffident]

The same thing happened on my main machine and my primary BOINC instance. Additionally, it assigned me a new cpid for some reason. Added the project to a different instance, and it started downloading tasks right away. :headscrat

I turned on CPU tasks, it download a handful of tasks that would use 16 threads and take 16 hours....so I said nope...and aborted them all. I changed it to 2 threads and it downloaded 4 tasks that would take 2 days 6 hours to complete so that's it for CPU tasks for me. I now have 18 GPU tasks, and I don't know what to do. Should I bother bunkering these few tasks or should I let them upload in hopes that I can get a lot more.

 

[tictoc]

I think the max is 20 per GPU.

 

[tictoc]

Amicable Numbers Project Homepage
​

May 12th, 0:00 UTC - May 17th, 0:00 UTC​

Countdown to Project Start | Countdown to Project End​

Server Status​

Official Team Stats

Project Specific Information:
upload server IP: 5.9.17.234; 2a01:4f8:160:91ef::2

• The GPU apps for Amicable Numbers will max out one CPU thread at the start of each task as it loads the task into memory
  
• Each GPU task will use a large amount of system memory, depending on the GPU that is running the task. For example a Radeon VII uses 8GB of system memory per task
• You can tune the tasks on your Amicable Numbers account page: Account->Preferences->Project
  • Max # of CPUs for this project (for CPU tasks)
  • Kernel size for AMD/ATI and NVIDIA GPUs
    • Currently my settings are 21 for AMD and 23 for NVIDIA. I'm not sure what the settings should be for maximum performance.​

app_config.xml:

<app_config>
  <app>
    <name>
    </name>
  </app>
</app_config>

*Note* app_config.xml is currently empty, and will be updated if there are any optimizations for the project.

 

[bfromcolo]

On my main system with Linux and a 980ti 4 of 6 tasks failed with this around the 1 minute mark when it goes from all CPU to start using the GPU. The other 2 completed fine.
Amicable/OpenCL.cpp, line 323: clCreateContext returned error -6

They seem to be running fine in Windows with RX-470/480 GPUs, but can't get enough to build any kind of bunker.

 

[tictoc]

On my main system with Linux and a 980ti 4 of 6 tasks failed with this around the 1 minute mark when it goes from all CPU to start using the GPU. The other 2 completed fine.
Amicable/OpenCL.cpp, line 323: clCreateContext returned error -6

They seem to be running fine in Windows with RX-470/480 GPUs, but can't get enough to build any kind of bunker.

I can't find the source code for the app, but I think you might be running into a host memory error. clCreateContext error -6, should be CL_OUT_OF_HOST_MEMORY. Which is the error if the OpenCL kernel is unable to allocate enough system memory for the task.

 

[bfromcolo]

Yes a memory error sounds right, I have 2 VMs and some unniverse also running. Been trying to adjust the kernel size which is supposed to reduce UI latency and memory usage, but it isn't having much impact on either that I can tell.


Do you have a URL to block for uploads, even it it turns out to only be 20 per GPU?

 

[bfromcolo]

Fine with me, thanks

 

[Diffident]

I can't get a bunker at all with my 1070ti in my 2P. It only downloads 1 task at a time. I have my full 20 on my desktop with the VII.

 

[tictoc]

I can't get a bunker at all with my 1070ti in my 2P. It only downloads 1 task at a time. I have my full 20 on my desktop with the VII.

I just added it to an NVIDIA machine, and it picked up 20 right off the bat. My other machines picked up one on the first try, and then 20 the next time it talked to the Amicable server.

 

[Diffident]

I'm spamming the update button, I'm not getting any. You must have taken them all.

 

[mmonnin]

is the GPU load still junk? It used to load the CPU for a full thread and use very little, like single digit GPU util percent.

 

[tictoc]

is the GPU load still junk? It used to load the CPU for a full thread and use very little, like single digit GPU util percent.

They have moved to a new data set that is pretty small, so GPUs are once again viable. It will use a full CPU thread at the beginning, and then load the GPU 100%.

 

[mmonnin]

Going to need like a 100 clients to keep GPUs busy for 3 days with a 20 task limit. Ain't nobody got time for that.

 

[bfromcolo]

For those using host file to block uploads this seems to be working:


127.0.1.1 sech.me

 

[Starbomba]

Well, it seems the app is set to gobble 8 GB RAM per GPU, no matter what. Both my RX 550 and RX 580, on different PCs, have allocated 8 GB RAM.

This presents an issue for my server, it is running only 8 GB RAM at the moment. My disk is getting hit hard with pagefile, but so far hasn't rejected anything, after i set the pagefile to be slightly more flexible.

Also, there is definitely a limit of WUs per GPU, i temporarily added my 550 into my main PC, and it got 20 WUs more.

 

[Diffident]

Looks like this is the one project the Radeon VII isn't very good....unless there is something I need to adjust. My VII completed tasks between 14:45 and 17:14 minute per task while my 1070ti was between 5:41 and 10:34. I don't know if the CPU tasks I was crunching on each contributed to the scores. My 2P with the 1070ti was also running Rosetta while my desktop with the VII was running Universe.

 

[tictoc]

NVIDIA is king at Amicable Numbers. Top 32 hosts are all running NVIDIA GPUs.