Printable View
I was wondering, I know someone started something once on CPH of projects. I would like get into an indebt chart. I have made a FB chart and am looking for the most logical FB/Team choices. My chart is currently made for best over thrown points but FB can also be gear tords over all #1. I personally am going to try and balance both. So that me and my FB team (hahaha) can work tords the goal and still get the best turnout for our self’s and the team.
I remember that this was worked on, but by who or where it is now, you would need access to the old forum to do the search, or hopefully someone has a better memory than I.
Didn't STM start working on something like this here on these boards?
I saw this posted a while back, it doesn't give you credits per hour but goes give a ratio of credits per project (projects that support gpu will be skewed).
http://boincstats.com/stats/project_cpcs.php
Wow that looks awesome! someone tell me how to read it?Quote:
Originally Posted by -GER-
Those are ratios. The ratio of the (row project)/(column project). So, in row 2 col 1, you see you get 0.57 AlmereGrid credits for every ABC@Home credit. In row 1 col 2, you see the inverse of that, you get 1.75 ABC credits for every AlmereGrid credit.Quote:
Originally Posted by Dandasarge
But yeah, you can't compare any CPU only project to any project with a GPU app. It's like comparing apples and chainsaws.
After reviewing the available data I have learned I need a lot more data. Or I really need to go shopping. I have some great start points. I’m also more paranoid the deep I get into this. The CPH list will not be public; the team will still have the info. I’ve already put 100+ hours into this over the last month. It will most benefit the top guns. I am going to 6 categories ATI/Nvida/32bit Intel/32bit AMD/64bit Intel/64bit AMD. I have aprox 40 projects, some have sub projects. I’m looking to compare aprox 60 projects x4 and 10 or so projects x6 there is no way I can do this with telling the difference between stock and over clocked.
I am not going to personally test Every Project. I have a top 10 list that I’m going to start with based on what I have. They should give me the #1 for each platform. After that I will continue to create my FB best choices list, on a needs/credit formula. Once I do that I will try and convince the black opps to do races to help my cause.
My uneducated list for in no order is
GPUGrid
AQUA
Collatz
DNETC
Einstein
Milkyway
PrimeGrid
SETI
Virtual Prairie
QMC
The stats you gather will work really, really well for you and keeping track of which projects have the best credits per hour for your machines is a great idea. How useful the stats will be to others will depend upon how closely their machines match up to your machines.
For example, I have several old P4s that have barely enough RAM to run. They therefore need projects that don't require several hundred megabytes of RAM or else they start swapping everything to disk. So, even though Project X may give more CPH, it won't on that box because of the small amount of RAM it has. So, that machine may have to run the 5th best project as far as CPH goes because it isn't capable of running faster ones. However, someone else may have a machine with the needed amount of RAM for Project X and agree that it is the best project for P4s. Whether hyperthreading is enabled affects some projects, or how often the box downloads work, etc. CPH need to use the runtime of the WU and not the CPU time. Almeregrid used to spend 20 minutes unzipping files, run for 2 minutes, and then spend 20 minutes zipping everything back up again. Total runtime was like an hour but it only ran for 2 minutes (and didn't pay any credit for the other 58 minutes either). I'm told that project is fixed, but those are all things to bear in mind when determining the best CPH. And, no matter what you find, someone else will disagree because on their particular setup, they get different results.
Things to consider: GPU vs CPU, Type of GPU: SP vs DP, CPU, Type of CPU: AMD vs Intel verses hyperthreaded or not, etc., amount of RAM, amount of disk, whether or not hte GPU or CPU is overclocked, the type of motherboard (can be 5-10% difference just in that), whether the GPU task was done on PCIe 0 (at 16x) or PCIx slot 1 (at 8x) or PCIe slot 4 (at 2x) as the speed for the slots varies from MB to MB and not all results show which slot it ran on or what the speed is of the slot.
Then again, if you've spent 100+ hours, you probably already have taken a bunch of that stuff into account already.
I can't acount for everything, I have bought a like Intel computer to match the one I curently run. so should be pretty close. People can make any reason they want. Do I think what i'm doing is going to give 100% results? no, do i think they are going to be, 98% yes. I am doing this for the team but yes each part can change the results 1-2% 5-6 major parts so my results can be 10% off. There really isn't a way to get closer without forking out 2 grand. I'm working on 500 bucks. Not going to be the same as someone working it out on 5k. I will give realistic anwsers on what the better prodjects are. I know this wasn't what you were saying and this is just the beer talking. I am comparing 2 3.2 dual core systems. the video cards are still in the works.
I have an old p4 2.2 that my kids watch tv on i could do testing on it... But i feel that with the 500+ test i'm going to work on in the next month that I will have done enough for a bit. i'm not working on hyper threading