Message boards : Promotion : Boinc in a box.
Message board moderation
Author | Message |
---|---|
Send message Joined: 26 Dec 21 Posts: 2 |
Hello, With these larger arm systems coming out with hundreds of cores would it be possible to create boinc nodes with multiple gpu's cores and run them off of tesla power walls and a reasonably small solar array. Making the entire package self sustainable? Creating a server rack with multiple nodes? Run them on linux with optimized kernels. Then use it as a tax write off? Make a solution and then bang for your buck it. |
Send message Joined: 10 May 07 Posts: 1450 |
... Then use it as a tax write off? I don't know of any country that would consider the electric consumed for work done on our 'computers' for BOINC a 'charity or not-for-profit' tax write-off. I would suggest anyone thinking of doing such consult a very good attorney/tax advisor for professional legal advice before attempting that. In response to your other suggestions about BOINC on ARM based GPU cluster computing and a custom *nix kernel: Various types of cluster/node computing proposals have been wished upon many times over the years in the forums here and the various BOINC projects. No one has got past just a wish. Maybe it's easier said than actually doable in real world computing. |
Send message Joined: 8 Nov 19 Posts: 718 |
Very few, if any, ARM systems have access to dGPUs, and if they do, there are even less projects that support them. Most GPU WUs are written to be ran on an X86 CPU and a PCIE GPU; or on an X86 CPU or ARM CPU. There are no ARM GPU projects I know of. But even if there were, Most X86/64 equivalent systems out there, use anywhere between 1 to 3 GPUs on a desktop (16-24PCIE lanes), and 1 to 6 GPUs on a Server (128PCIE lanes), with the exception of some exotic hardware being able to run 8-12 dGPUs. Both X86/64 CPUs as well as ARM CPUs, are manufactured at 12, 10, 8, 7, 6, 5, and pretty soon 3nm. The power difference between a 5nm x86/64 and a 5nm ARM CPU is about the same, both at idle and full load. X86 uses more power when it has more PCIE lanes; which ARM doesn't have. So you might as well install those solar panels, get that $70k Tesla battery pack, and run yourself a $200k Nvidia DGX A100 if you want to save the environment. If not, you can probably save a good $300k on buying a simple Linux desktop pc with 3x RTX 3090 GPUs, for a good $7k, and do 1/3rd to 1/4th the work at half the power consumption. |
Copyright © 2024 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.