supercomputing power from e-waste - feasible?

Message boards : The Lounge : supercomputing power from e-waste - feasible?
Message board moderation

To post messages, you must log in.

AuthorMessage
noderaser
Avatar

Send message
Joined: 2 Jan 14
Posts: 276
United States
Message 64779 - Posted: 8 Oct 2015, 23:06:16 UTC

I used to have a small cluster of machines (4-6, it varied a bit) when I was in my dorm room at college and therefore had "free" power... They were all "salvaged" from other people who were giving them up after upgrading, I even found one in the dumpster at the dorm. When I moved out of the dorms and into my own place, I had to stop that practice as my electric bill was running near a quarter of my rent at the time.

I was never in the grain of the serious cluster crunchers, just trying to give what I could at the time... Many of them are assembling their clusters from slightly used machines, but if you go too "old" you fall below a FLOPs per watt number that some people think is not worth the money and effort. I've thought about adding a few machines back now that I'm on a more stable financial footing, currently I only have one machine (my DVR) that runs 24/7, the rest are only on while they're in use or for periods where I'm trying to meet a crunching goal.

Personally, I'm of the opinion that "every machine is a good machine", and the "value" of having an older machine crunching is a relative thing to the person paying to keep it running. I have quite a collection of old computers (mostly Macs) that still get used every now and then for various purposes, and any that are BOINC capable still push in a few tasks every now and then.

In the case of an organization where resources are an issue, you could certainly assemble an "ugly" cluster of donated or salvaged machines. However, I think that the challenge for such an organization would be keeping the lights on; it's easy to write grants to obtain equipment or to meet a particular goal, but basic utilities and operating costs is not a very sexy topic to a grant writer. I think MilkyWay is a good example of this, having lost their major NSF grant, they have taken to crowdfunding to keep things afloat.
My Detailed BOINC Stats
ID: 64779 · Report as offensive
noderaser
Avatar

Send message
Joined: 2 Jan 14
Posts: 276
United States
Message 64780 - Posted: 8 Oct 2015, 23:14:37 UTC

Another thought; if you were trying to assemble a cluster you would probably get more "bang" for your buck if you were to take some older machines and upgrade them with inexpensive GPUs. They even made some HD 3000/4000/5000 series GPUs on PCI and AGP; they might not be hot performers by today's standards, but they should still do better than many CPUs.
My Detailed BOINC Stats
ID: 64780 · Report as offensive
Jazzop

Send message
Joined: 19 Dec 06
Posts: 90
United States
Message 65174 - Posted: 1 Nov 2015, 6:21:00 UTC

I've been doing this for years, although it's getting harder as fewer projects support orphaned hardware (e.g., MIPS, PPC, SPARC, DEC Alpha), and the workunits are more demanding of advanced CPU instruction sets, memory, and so forth. SETI is pretty much the only recourse for really old, slow, or weird machines.

I'm just about to fire up an old dual-CPU Pentium II server and see what I can get it to do. This tends to be a winter thing, since the heat generated by several old machines crunching at maximum load is unbearable in the summer.

It also helps to have all utilities included in your rent!
ID: 65174 · Report as offensive
noderaser
Avatar

Send message
Joined: 2 Jan 14
Posts: 276
United States
Message 65189 - Posted: 2 Nov 2015, 0:05:18 UTC - in response to Message 65174.  

SETI is pretty much the only recourse for really old, slow, or weird machines.

The loss of SIMAP was a huge blow to those of us who insist on keeping archaic platforms (mostly PowerPC) running, though I can understand why projects are reluctant to support them when they submit a minimal portion of the returned work. Still, it's nice to have them doing something useful--although I don't run any of them 24/7. I only have one host (my DVR, an old-ish AMD Phenom X3 8400) that is a 24/7 cruncher, aside from my old phone (Samsung Galaxy S3) that is really a non-factor when it comes to power consumption.

CPU computing is a waste of time unless you are talking multi core i7 or Xeons.

I guess that depends on your goal; if you're talking about pure credit, then it may well be... But there are many projects that still don't have GPU apps.
My Detailed BOINC Stats
ID: 65189 · Report as offensive
noderaser
Avatar

Send message
Joined: 2 Jan 14
Posts: 276
United States
Message 65284 - Posted: 5 Nov 2015, 23:51:46 UTC - in response to Message 65195.  

For some years now I have been running Wuprop@home which is a non cpu intensive (NCI) project and racked up over 1/2 million credit. It will run on virtually anything. I wish there was more NCI stuff.

Running WUprop without any other projects is pretty pointless, though... The purpose of the project is to collect data about how the other projects run on various hardware. Most other NCI apps require external hardware, like QCN and Radioactive@Home.
My Detailed BOINC Stats
ID: 65284 · Report as offensive

Message boards : The Lounge : supercomputing power from e-waste - feasible?

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.