Message boards : GPUs : BOINC only recognising 4GB RAM on a Quadro K6000
Message board moderation
Author | Message |
---|---|
![]() Send message Joined: 21 Jun 14 Posts: 3 ![]() |
Hi everyone. Can anyone tell me why BOINC only recognises 4GB RAM on my Quadro K6000 graphics card which has 12GB RAM? Is there a setting I can change so it sees all 12GB? Also what are the best setting to choose to maximise GPU usage? I am using Win 8.1 Pro 64-bit on a HP Z820 workstation with Dual Xeon E5 2687W v2 CPU's, 32GB ECC RAM and SSD. I am running BOINC Manager 7.2.42 (x64). |
![]() Send message Joined: 29 Aug 05 Posts: 15618 ![]() |
The algorithm used is 32bit, which has a maximum addressable memory range of 4,096MiB. Most all project applications for the GPU are 32bit applications. There's also no benefit from switching to 64bit applications, because accessing the memory on a GPU is rather slow. At least, this was the reasoning a year ago. I'll see if there's an update on that. |
![]() Send message Joined: 21 Jun 14 Posts: 3 ![]() |
Thanks, I suspected it may have been a 32-bit code issue but was not sure as BOINC is listed as 64-bit. I am also finding BOINC is only using 45-50% of the GPU. |
![]() Send message Joined: 29 Aug 05 Posts: 15618 ![]() |
I am also finding BOINC is only using 45-50% of the GPU. BOINC isn't using your GPU, the project's science application is. So if they don't utilize the GPU completely, or put it under complete load, you'll have to ask about that at their forums. |
![]() Send message Joined: 29 Aug 05 Posts: 15618 ![]() |
I have had an answer from the developers. This is a shortcoming in the openCL specifications. OpenCL has only 32 bits for global_mem_size, so it can report a max of only 4GB. See the description for CL_GLOBAL_MEM_SIZE in http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/ |
![]() Send message Joined: 23 Feb 12 Posts: 198 ![]() |
Thanks, I suspected it may have been a 32-bit code issue but was not sure as BOINC is listed as 64-bit. cocopop05, what you will need to do is go to the project you are running and look into running an app_config.xml file that will allow you to run more than one work unit at a time on your card. This will make a much more efficient output in many cases. However, keep an eye on it because not all apps will benefit from more than one ran at a time. You may have to do some experimenting. Also, keep in mind that if your run time goes up on work units to look at the total time it takes to complete more than one work unit. Compare those numbers to completing that many work units when running just one at a time. This will tell you if you are increasing productivity or just slowing things down. ![]() |
![]() Send message Joined: 21 Jun 14 Posts: 3 ![]() |
Thank you very much everyone for all your responses and advice. I am learning a lot thanks to you all. |
Copyright © 2025 University of California.
Permission is granted to copy, distribute and/or modify this document
under the terms of the GNU Free Documentation License,
Version 1.2 or any later version published by the Free Software Foundation.