Discussion Thread for the Continuous GW Search known as O2MD1 (now O2MDF - GPUs only)
Language
Copyright © 2024 Einstein@Home. All rights reserved.
Copyright © 2024 Einstein@Home. All rights reserved.
tullio wrote:They all fail on
)
Are you running the GW CPU tasks at the same time you are trying to run the GW GPU tasks? If so, try running GW GPU tasks only, no CPU tasks, and see whether you still get the fails. Also suspend Asteroids@home GPU tasks to remove that variable while testing runs for GW GPU.
Ideas are not fixed, nor should they be; we live in model-dependent reality.
I run only one GPU task an
)
I run only one GPU task an the time. Now an Asteroids@home GPU task runs alongside four Einstein@home CPU tasks and 3 Asteroids@home CPU Tasks with no problem.
Tullio
In 2019 this same PC ws
)
In 2019 this same PC ws running the GW 01 GPU tasks with no errors. Now it fails on the GW 02 tasks. my nVidia drivers are the latest as provided by Geforce. Evidently your software does not like them, this is the only reason I can see.
Tullio
Tullio it has been
)
Tullio it has been established that some of those require more than 3GB of memory on Nvidea cards. I get the same thing on my GTX1060 3GB cards. The only saving grace is they bomb out without wasting too much time.
I have a GTX 1050Ti with 4 GB
)
I have a GTX 1050Ti with 4 GB video RAM but it has no additional power connector and it probably caused some faults on another PC with Windows 8.1. Now that PC is dead and I don't wanto to cause problems on this PC which hosts also a Linux Virtual Machine with SuSE Tumbleweed development version, frequently upgraded and now running a 5.6.6-1 kernel. It runs also Einstein@home CPU tasks but since it is enlisted in Science United they don't appear here.
Tullio
tullio wrote: I have a GTX
)
I detest the idea of science united. No control any more.
If this page takes an hour to load, reduce posts per page to 20 in your settings, then the tinpot 486 Einstein uses can handle it.
Betreger wrote: Tullio it
)
Mine are AMD, but with 3GB they do not crash with large GW tasks, they use the system RAM (although more slowly). Can Nvidia cards really not do this?
If this page takes an hour to load, reduce posts per page to 20 in your settings, then the tinpot 486 Einstein uses can handle it.
Peter Hucker wrote: Betreger
)
This goes back to my old statement that Nvidia cards use 27% of available RAM of GPU for OpenCl where ATI and Intel use between 56-67% of Ram of GPU for OpenCl. So you can see why a ATI of 3GB can run where a Nvidia of 3GB can't.
Will be interesting to see
)
Will be interesting to see where VRAM usage settles out with the new OpenCL 3.0 specification that is being finalized. It is based on the old 1.2 specification that Nvidia settled on.
Will AMD cards in the future be limited to 27% of memory like Nvidia cards currently are?
Zalster wrote: This goes
)
Are you saying that a Nvidia card cannot use more than a quarter of it's RAM for OpenCL? I find it strange that either make of card would have any limit on RAM usage, and should use all of it for whatever language is running.
But as I said, I've made my AMD cards run out of RAM, by running multiple GW WUs. And they have no problem simply borrowing some RAM from the main system. It's slower, but it works, no computation errors.
If this page takes an hour to load, reduce posts per page to 20 in your settings, then the tinpot 486 Einstein uses can handle it.