GPU – do you? Jon Sonntag (Collatz) helps you find out #2
GPU – do you?
Jon Sonntag (Collatz) helps you find out – Part 2
What's all the fuss about GPU (video card) computing, and is it worth the effort? Do you still use exclusively your CPU for Distributed Computing and BOINC? Did you consider using your video card or do you already have some experience in GPU computing and consider doing upgrades? Is your old video card still helpful for computing? Jon Sonntag, project developer and administrator of Collatz Conjecture, helps you find out.
Do or don't? About monsters, laptops and old mules
Is there anyone who definitely should not use his GPU, for whatever reason?
1. Laptops are made for portability and because of that, there are trade offs in the amount of cooling and the number of calculations their GPUs can do. If the temperatures are too high (check out GPU-Z) one can either try underclocking or not use the GPU for computing. There are some aftermarket laptop coolers available which blow cool air onto the bottom of the laptop which, if the laptop uses the case as a heat sink, may be cool it enough.
2. Computers that are not your own and/or you do not have written permission to use. A verbal agreement is not enough. Running BOINC, whether on a CPU or a GPU uses electricity and causes wear and tear on a system as it is running 24x7 instead of only 8 hours a day.
3. Anyone from team Sicituradastra. ;-)
http://imageshack.us/a/img839/7540/6...14c15f7970.jpg
PCB of an AMD 7970. Source: overclock.net
Regarding crunching on laptops, you would consider TThrottle an option for mobile GPUs?
If the laptop fan can't keep the temperatures at a reasonable level, BOINC has settings to throttle the CPU but it works like a pulse-width modulator (e.g. a setting of 75% means it will run at 100% for 3 seconds and then 0% for 1 second). BOINC's doesn't have any options for the GPUs and the on-off method it uses for the CPUs isn't the best. TThrottle [http://www.efmer.eu/boinc/] actually monitors the temperatures and throttles back only when it needs to. It is a much better solution than the one provided in the BOINC Manager.
http://imageshack.us/a/img545/9151/gk104die.jpg
Worth the money? A brand new NVIDIA GK 104 chip
Source: legitreviews.com
Now, if I can do so much with so little effort, would you say I should buy a monster GPU just for science? Or can I achieve something with the hardware I have at home?
That really depends upon how much money you have and how much you love volunteer computing. If you can afford a monster GPU and the electricity to run it, go for it. You will be contributing anywhere from 10 to 100 times the work that you do now. But, since a monster GPU can easily cost as much as an entire PC and since most people don't have $1000 to spend on a GTX 690 or have other priorities (pay the rent or buy a GPU.....hmmm.....decisions, decisions.....)
Luckily, there is no need to purchase a new GPU. Almost any GPU made in the last several years can be used for computing. Even the slowest GPU will often match the output of the CPU and double the amount of work that the computer contributes to a project.
About Slicker, Indiana Jones and about the guy behind Collatz...
We 'talked' a lot about GPUs, so finally I'd like to ask some questions about you. I reckon I should ask you who Slicker is, if that's not too personal?
Like Indiana Jones, who took the name Indy from his dog, Slicker was our pure bred Weimaraner. His bout with cancer that had gone on for several years ended this past summer.
http://imageshack.us/a/img811/6642/s...gsquirrles.jpg
Slicker chasing squirrels
Jon, do you work in IT or mathematics, just thinking of your BOINC project?
Once upon a time, I was an elementary school teacher. I planned to get my master's degree and teach high school computer science. In Illinois, that meant getting state certified to teach high school mathematics because at the time, Illinois lumped computer science into the mathematics department. I taught a few IT courses at the college level while working at the university's computer center, but after a couple years, worked strictly in IT. That was twenty years ago and I've been in IT/IT Management ever since.
How did you come to the idea developing and running a project like Collatz?
It was a combination of things. Several BOINC projects supported NVIDIA GPUs but only MilkyWay supported AMD GPUs and that's not because of the project administrator, but because on a single volunteer (Crystal Physik) who wrote the AMD application for them. Unfortunately, the MilkyWay server couldn't keep up with AMD GPUs. I owned 4 of them and they would often be idle because there was no work for them.
At the same time, I worked at a publishing company which scored and reported standardized tests. I proposed using BOINC to manage the scoring of the tests and using a GPU application to do the actual scoring computations would have greatly reduced the processing time considerably. So, I started looking for a simple problem that I could use to learn GPU programming and set up a prototype to convince my boss of the merits of distributed computing. Gipsel, as he is known on the Collatz Conjecture project, who was the volunteer from MilkyWay, really helped me out with the GPU applications and tutored me on the do's and don'ts. I had very little Linux experience, so Crunch3r helped with porting the applications to Linux.
Jon, would you do it again? And – will you eventually come up with a new GPU project?
The SETI.USA team was considering hosting a BOINC project as a team. I found a project and wrote applications for it. Since the team lost interest in running their own project, I may add it to the Collatz project as it is a mathematics based problem. Right now my focus is on sorting out some issues with the OpenCL applications. Later in Fall I expect to have more free time, so maybe I'll get around to adding it then.
How to speed up and how to get started
From your practical experience with Collatz, could you give us an example of the speed differences between crunching on GPUs vs crunching on CPUs?
HD 6970 vs Intel i7 950: The GPU is 132 times faster than a single i7 core, and about 17 times faster than all 8 cores combined. Many of the top crunchers have 2, 3, or even 4 GPUs in a single computer.
Is BOINC really up for supporting GPUs? Will I have to rack my head to make it work?
It is very, very easy to do. Install the latest video driver for your video card (a.k.a. GPU). Then join a BOINC project which has GPU applications for your type of GPU.
http://imageshack.us/a/img525/9109/gtx480triplesli.jpg
Take it easy: you don't need a triple-SLI for getting started. Source: Vizworld.com
Which projects would you recommend for getting started?
That depends upon your reasons for volunteer computing. If you have an area in which you are intrigued, find a project in that area that matches your type of GPU. If you don't know which type you have, the BOINC event log (Control+ Shift+E) will display the GPUs it recognizes when it first starts up. I believe Collatz supports the largest number and types of GPUs so if it can be used, it should work on Collatz.
If you are competitive and want to get the most cobblestones (a.k.a. BOINC credits) you can, POEM, Donate, DistrRTGen, and Moo!Wrapper pay the most credits per hour. Credits sometimes change, but your team's message board will likely have suggestions as to where your GPU can be best utilized.
Last but not least: What are your ultimate three tips for anyone who wants to start crunching on his GPU?
1. If using an aftermarket GPU, make sure the power supply in the PC is powerful enough to handle the extra load. Most GPUs' specifications will list the minimum size of the power supply in watts. Make sure yours is 20-30% higher than the minimum. If you need to upgrade the power supply, look for an 80-plus efficient rated one. While it will cost a little more, it will pay for itself by using less electricity. Most computers with embedded video (e.g. a VGA or HDMI jack right by the keyboard, USB, etc. jacks rather than on a separate card) will not be able to be used. A separate GPU can be added for as little as $15 that will crunch.
2. Keep it clean and cool. Your GPU will last longer and run cooler if you blow the dust off it every once or twice every year. Several of my GPUs are over 5 years old.
3. If you are having trouble, ask for help on the project's or your team's message boards. Make sure to give as much information as possible (project, type of GPU, video driver version, BOINC version, and specifically what isn't working). You will most likely get a reply within a day or two and be up and running in no time at all.
Jon, thank you – once more - for taking your time and helping find out. I think you may have inspired some long-term crunchers to try what they can do on their GPU, no matter if old or brand new.
The complete interview with Jon Sonntag will be available as blog and as pdf download.
Re: GPU – do you? Jon Sonntag (Collatz) helps you find out #2
The blog is ready, including the download link to the full pdf version. Kudos to Bok for hosting the full pdf.
Unfortunately the originally used forum fonts seem gone, so you can just see the orignal layout in the pdf version. Also, I could not extend my blog to more than 10000 words, so that's just enough for part 1. Sorry about that.
If you want to read the full version please use this download link.
Re: GPU – do you? Jon Sonntag (Collatz) helps you find out #2
A wonderful read. Thank you Matt and Jon for taking the time to put this together.
Re: GPU – do you? Jon Sonntag (Collatz) helps you find out #2
Thank you for the great receiption this interview has found so far. I am already eyeing a repost of the article for the Polish National Team.
