As you dig into BOINC crunching you will notice that there are two types of projects, those that use your computers CPU and those that can take advantage of some graphics cards. You will also notice that a CPU project will award perhaps 100 to 500 credits per CPU core per day, GPU projects may award 200,000 credits per card per day! If GPU computing is 1000 times faster, why don’t all projects use the GPU?
There are two main reasons for this. The first and most important is that your CPU is much more flexible in it’s abilities and has a richer programming infrastructure. Complex tasks that require extensive code are better served (and in many cases must be served) by the flexibility of your CPU. Secondly, the vast majority of software tools and trained programmers are aimed at CPU computing, not GPU computing. For example, I co-own an e-commerce and software company, in our 12 years in business we have never created a GPU based application. Therefore even if in theory a task could utilize GPUs the project may not have access to programmers who can implement the solution in a GPU computing environment.
This is War AMD (ATI) vs NVIDIA
As if a project did not have enough to worry about when deciding on CPU or GPU processing, there is a third decision to be made. Should they use AMD or Nvidia graphics cards. If you Google AMD vs Nvidia you will see a nearly endless list of comparisons and reviews and arguments regarding which is the better gaming platform. It is largely a Chevy vs Ford type argument as most games will play just fine on most modern graphics cards from either vendor.
Fortunately there are some technical differences that largely remove this decision from the us verses them mindset of gamers. Nvidia was earliest to the BOINC party and GPU computing in general. The first GPU computing projects were all aimed at Nvidia graphics cards, the CUDA ( Compute Unified Device Architecture ) and is now entering it’s 4th generation. Nvidia has made a conscious decision to go after the scientific computing market with it’s GPU products, AMD has largely concentrated on the straight graphics market with it’s products. AMD has responded with it’s Stream / OpenCL GPU computing environment. Each project looks at the type of computing that it needs to do, and the resources it has to write it’s GPU application and then chooses the platform that it thinks will work best for it.Â At this time there are more Nvidia compatible projects than AMD ones, but there is even one project, Milkyway@home that uses both! A current list of GPU projects is here.
If you already have a higher end graphics card, welcome to the world of BOINC, join team USA and welcome aboard! You will be choosing from the projects that support the card you already have. If you are a new cruncher or want to add a major kick in the pants to your daily credit total, then you have a larger decision to make. We would suggest that you look at the projects in the list above first, and make your GPU flavor decision based on which projects are most interesting to you. The chart below will show you the raw processing power of some recent graphics cards that we know about. If you have a card that is not on the list, please let us know by posting to our forums and we will add it to the list.
GPU BOINC benchmarks
If all these credits sound great to you, check out our GPU Benchmarks The Gflop ratings will tell you how many credits your can expect from one card to another. Your exact values may differ depending on your settings, the card manufacture, and other factors. It is designed to be a general guide only.
The postman dropped my energy bill and broke my foot!
If you are going to be a heavy duty cruncher like many of us, you will soon learn that buying hardware is only a part of the cost of entry into “The Big Leagues”. You will also have to plan on powering and cooling your crunching machine(s). For instance, looking at my stats.
You will see a BSrac (BOINCSTATS recent average credit) of 712,114 as of March 30th 2011. For me to achieve this RAC it takes 4 computers with multiple high end graphics cards, and an additional 2 machines with a single GPU each. I tend to like more scientific crunching than the math based ones, which lowers the point totals per GPU. For instance, the total Collatz and DNETC RAC value is 430K, and that is done with 3 AMD GPUS, whereas GPUGRID and Milkyway, both scientific GPU projects have 7 GPU cards (NVIDIA 460 and 475’s) and produce about 260K credits a day. (The balance are CPU projects running on these and a few other machines).
What does this mean, it means that, not counting the PC’s themselves, I am running 10 GPU’s at about 200 Watts each, 24/7. That is a 2KW electrical load every day all day long. If you are paying 12 cents a KW/hr, a fairly typical rate, that adds up to $5.76 a day, doesn’t sound bad, how about when you think about it as $172 a month or $2,073 a year!
What to do? Well, of course the easiest solution is to just be rich Unfortunately that has not worked out so well for me! It can be quite enough to throw $300 at a video card from time to time without the monthly grind of a huge energy bill.
To address this I rely on Scripture, specifically Acts 20:35 which states “In everything I did, I showed you that by this kind of hard work we must help the weak, remembering the words the Lord Jesus himself said: ‘It is more blessed to give than to receive.’ ” Or as I like to paraphrase it “It is more blessed to give (a video card) than to receive (a large electric bill).
I have found that most everyone knows friends and family that would like a free upgrade, consider offering a GPU for their family PC that they can use for gaming, tell them about BOINC, and away you go. They will be thrilled with the great gaming experience they now have, are generally interested in helping advance science and knowledge in general, and running a single GPU will not have an effect on the family finances. This is a win-win all around.
Of course there are other creative ideas as well. I have updated a couple PC’s at my office, but of course I co-own the company so I have permission. Still if you are in a situation where you can do that, that is one idea. In any case, do not make modifications or run BOINC on any machine that you don’t have permission to do so.
Next, we will look at how to build a credit crunching champion.